sgjoen@nyx.net
md Kernel Patch
cachefs
md)
fstab
/mnt
For unclear reasons this brand new release is codenamed the Taylor3 release.
New code names will appear as per industry standard guidelines to emphasize the state-of-the-art-ness of this document.
This document was written for two reasons, mainly because I got hold of 3 old SCSI disks to set up my Linux system on and I was pondering how best to utilise the inherent possibilities of parallelizing in a SCSI system. Secondly I hear there is a prize for people who write documents...
This is intended to be read in conjunction with the Linux Filesystem Structure Standard (FSSTND). It does not in any way replace it but tries to suggest where physically to place directories detailed in the FSSTND, in terms of drives, partitions, types, RAID, file system (fs), physical sizes and other parameters that should be considered and tuned in a Linux system, ranging from single home systems to large servers on the Internet.
The followup to FSSTND is called the Filesystem Hierarchy Standard (FHS) and covers more than Linux alone. FHS versions 2.0, 2.1 and 2.2 have been released but there are still a few issues to be dealt with. Many recent distributions are now aiming for FHS compliance.
It is also a good idea to read the Linux Installation guides thoroughly and if you are using a PC system, which I guess the majority still does, you can find much relevant and useful information in the FAQs for the newsgroup comp.sys.ibm.pc.hardware especially for storage media.
This is also a learning experience for myself and I hope I can start the ball rolling with this HOWTO and that it perhaps can evolve into a larger more detailed and hopefully even more correct HOWTO.
First of all we need a bit of legalese. Recent development shows it is quite important.
This document is Copyright 1996 Stein Gjoen. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.1 or any later version published by the Free Software Foundation with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts.
If you have any questions, please contact <{linux-howto@metalab.unc.edu}>
Use the information in this document at your own risk. I disavow any potential liability for the contents of this document. Use of the concepts, examples, and/or other content of this document is entirely at your own risk.
All copyrights are owned by their owners, unless specifically noted otherwise. Use of a term in this document should not be regarded as affecting the validity of any trademark or service mark.
Naming of particular products or brands should not be seen as endorsements.
You are strongly recommended to take a backup of your system before major installation and backups at regular intervals.
This is a major upgrade featuring a new copyright statement that is intended to be Debian compliant and allow for inclusion in their distribution. A number of mistakes are corrected and new features added such as descriptions of recent ATA features and more.
On the development front people are concentrating their energy towards completing Linux 2.4 and until that is released there is not going to be much news on disk technology for Linux.
Also now the document is available in postscript both for US letter as well as European A4 formats.
The latest version number of this document can be gleaned from my plan entry if you finger my Nyx account.
Also, the latest version will be available on my web space on Nyx in a number of formats:
A European mirror of the Multi Disk HOWTO just went on line.
In this version I have the pleasure of acknowledging even more people who have contributed in one way or another:
ronnej (at ) ucs.orst.edu
cm (at) kukuruz.ping.at
armbru (at) pond.sub.org
R.P.Blake (at) open.ac.uk
neuffer (at) goofy.zdv.Uni-Mainz.de
sjmudd (at) redestb.es
nat (at) nataa.fr.eu.org
sundbyk (at) oslo.geco-prakla.slb.com
ggjoeen (at) online.no
mike (at) i-Connect.Net
roth (at) uiuc.edu
phall (at) ilap.com
szaka (at) mirror.cc.u-szeged.hu
CMckeon (at) swcp.com
kris (at) koentopp.de
edick (at) idcomm.com
pot (at) fly.cnuce.cnr.it
earl (at) sbox.tu-graz.ac.at
ebacon (at) oanet.com
vax (at) linkdead.paranoia.com
tschenk (at) theoffice.net
pjfarley (at) dorsai.org
jean (at) stat.ubc.ca
johnf (at) whitsunday.net.au
clasen (at) unidui.uni-duisburg.de
eeslgw (at) ee.surrey.asc.uk
adam (at) onshore.com
anikolae (at) wega-fddi2.rz.uni-ulm.de
cjaeger (at) dwave.net
eperezte (at) c2i.net
yesteven (at) ms2.hinet.net
cj (at) samurajdata.se
tbotond (at) netx.hu
russel (at) coker.com.au
lars (at) iar.se
GALLAGS3 (at) labs.wyeth.com
morimoto (at) xantia.citroen.org
shulegaa (at) gatekeeper.txl.com
roman.legat (at) stud.uni-hannover.de
ahamish (at) hicks.alien.usr.com
hduff2 (at) worldnet.att.net
mbaehr (at) email.archlab.tuwien.ac.at
adc (at) postoffice.utas.edu.au
pjm (at) bofh.asn.au
jochen.berg (at) ac.com
jpotts (at) us.ibm.com
jarry (at) gmx.net
LeBlanc (at) mcc.ac.uk
masy (at) webmasters.gr.jp
karlheg (at) hegbloom.net
goeran (at) uddeborg.pp.se
wgm (at) telus.net
Special thanks go to nakano (at) apm.seikei.ac.jp for doing the
Japanese translation,
general contributions as well as contributing an example of
a computer in an academic setting, which is included at the end of this
document.
There are now many new translations available and special thanks go to the translators for the job and the input they have given:
chewie (at) nuernberg.netsurf.de
jonah (at) swipnet.se
Patrick.Loiseleur (at) lri.fr
yesteven (at ) ms2.hinet.net
bigpaul (at) flashnet.it
ICP Vortex is gratefully acknowledges for sending in-depth information on their range of RAID controllers.
Also DPT is acknowledged for sending me documentation on their controllers as well as permission to quote from the material. These quotes have been approved before appearing here and will be clearly labelled. No quotes as of yet but that is coming.
Not many still, so please read through this document, make a contribution and join the elite. If I have forgotten anyone, please let me know.
New in this version is an appendix with a few tables you can fill in for your system in order to simplify the design process.
Any comments or suggestions can be mailed to my mail address on Nyx: sgjoen@nyx.net.
So let's cut to the chase where swap and /tmp are
racing along hard drive...
As this type of document is supposed to be as much for learning as a technical reference document I have rearranged the structure to this end. For the designer of a system it is more useful to have the information presented in terms of the goals of this exercise than from the point of view of the logical layer structure of the devices themselves. Nevertheless this document would not be complete without such a layer structure the computer field is so full of, so I will include it here as an introduction to how it works.
It is a long time since the mini in mini-HOWTO could be defended as proper but I am convinced that this document is as long as it needs to be in order to make the right design decisions, and not longer.
This is based on how each layer access each other, traditionally with the application on top and the physical layer on the bottom. It is quite useful to show the interrelationship between each of the layers used in controlling drives.
___________________________________________________________
|__ File structure ( /usr /tmp etc) __|
|__ File system (ext2fs, vfat etc) __|
|__ Volume management (AFS) __|
|__ RAID, concatenation (md) __|
|__ Device driver (SCSI, IDE etc) __|
|__ Controller (chip, card) __|
|__ Connection (cable, network) __|
|__ Drive (magnetic, optical etc) __|
-----------------------------------------------------------
In the above diagram both volume management and RAID and concatenation are optional layers. The 3 lower layers are in hardware. All parts are discussed at length later on in this document.
Most users start out with a given set of hardware and some plans on what they wish to achieve and how big the system should be. This is the point of view I will adopt in this document in presenting the material, starting out with hardware, continuing with design constraints before detailing the design strategy that I have found to work well. I have used this both for my own personal computer at home, a multi purpose server at work and found it worked quite well. In addition my Japanese co-worker in this project have applied the same strategy on a server in an academic setting with similar success.
Finally at the end I have detailed some configuration tables for use in your own design. If you have any comments regarding this or notes from your own design work I would like to hear from you so this document can be upgraded.
Although not the biggest HOWTO it is nevertheless rather big already and I have been requested to make a reading plan to make it possible to cut down on the volume
(aka the elite). If you are familiar with Linux as well as disk drive technologies you will find most of what you need in the appendices. Additionally you are recommended to read the FAQ and the Bits'n'pieces chapter.
(aka Competent). If you are familiar with computers in general you can go straight to the chapters on technologies and continue from there on.
(mostly harmless). You just have to read the whole thing. Sorry. In addition you are also recommended to read all the other disk related HOWTOs.
A far more complete discussion on drive technologies for IBM PCs can be found at the home page of The Enhanced IDE/Fast-ATA FAQ which is also regularly posted on Usenet News. There is also a site dedicated to ATA and ATAPI Information and Software.
Here I will just present what is needed to get an understanding of the technology and get you started on your setup.
This is the physical device where your data lives and although the operating system makes the various types seem rather similar they can in actual fact be very different. An understanding of how it works can be very useful in your design work. Floppy drives fall outside the scope of this document, though should there be a big demand I could perhaps be persuaded to add a little here.
Physically disk drives consists of one or more platters containing data that is read in and out using sensors mounted on movable heads that are fixed with respects to themselves. Data transfers therefore happens across all surfaces simultaneously which defines a cylinder of tracks. The drive is also divided into sectors containing a number of data fields.
Drives are therefore often specified in terms of its geometry: the number of Cylinders, Heads and Sectors (CHS).
For various reasons there is now a number of translations between
Basically it is a mess and a source of much confusion. For more information you are strongly recommended to read the Large Disk mini-HOWTO
The media technology determines important parameters such as read/write rates, seek times, storage size as well as if it is read/write or read only.
This is the typical read-write mass storage medium, and as everything else in the computer world, comes in many flavours with different properties. Usually this is the fastest technology and offers read/write capability. The platter rotates with a constant angular velocity (CAV) with a variable physical sector density for more efficient magnetic media area utilisation. In other words, the number of bits per unit length is kept roughly constant by increasing the number of logical sectors for the outer tracks.
Typical values for rotational speeds are 4500 and 5400 RPM, though 7200 is also used. Very recently also 10000 RPM has entered the mass market. Seek times are around 10 ms, transfer rates quite variable from one type to another but typically 4-40 MB/s. With the extreme high performance drives you should remember that performance costs more electric power which is dissipated as heat, see the point on Power and Heating.
Note that there are several kinds of transfers going on here, and that these are quoted in different units. First of all there is the platter-to-drive cache transfer, usually quoted in Mbits/s. Typical values here is about 50-250 Mbits/s. The second stage is from the built in drive cache to the adapter, and this is typically quoted in MB/s, and typical quoted values here is 3-40 MB/s. Note, however, that this assumed data is already in the cache and hence for maximum readout speed from the drive the effective transfer rate will decrease dramatically.
Optical read/write drives exist but are slow and not so common. They were used in the NeXT machine but the low speed was a source for much of the complaints. The low speed is mainly due to the thermal nature of the phase change that represents the data storage. Even when using relatively powerful lasers to induce the phase changes the effects are still slower than the magnetic effect used in magnetic drives.
Today many people use CD-ROM drives which, as the name suggests, is read-only. Storage is about 650 MB, transfer speeds are variable, depending on the drive but can exceed 1.5 MB/s. Data is stored on a spiraling single track so it is not useful to talk about geometry for this. Data density is constant so the drive uses constant linear velocity (CLV). Seek is also slower, about 100 ms, partially due to the spiraling track. Recent, high speed drives, use a mix of CLV and CAV in order to maximize performance. This also reduces access time caused by the need to reach correct rotational speed for readout.
A new type (DVD) is on the horizon, offering up to about 18 GB on a single disk.
This is a relatively recent addition to the available technology and has been made popular especially in portable computers as well as in embedded systems. Containing no movable parts they are very fast both in terms of access and transfer rates. The most popular type is flash RAM, but also other types of RAM is used. A few years ago many had great hopes for magnetic bubble memories but it turned out to be relatively expensive and is not that common.
In general the use of RAM disks are regarded as a bad idea as it is normally more sensible to add more RAM to the motherboard and let the operating system divide the memory pool into buffers, cache, program and data areas. Only in very special cases, such as real time systems with short time margins, can RAM disks be a sensible solution.
Flash RAM is today available in several 10's of megabytes
in storage and one might be tempted to use it for fast, temporary
storage in a computer. There is however a huge snag with this: flash
RAM has a finite life time in terms of the number of times you can
rewrite data, so putting
swap, /tmp or /var/tmp on such
a device will certainly shorten its lifetime dramatically.
Instead, using flash RAM for directories that are read often but
rarely written to, will be a big performance win.
In order to get the optimum life time out of flash RAM you will need to use special drivers that will use the RAM evenly and minimize the number of block erases.
This example illustrates the advantages of splitting up your directory structure over several devices.
Solid state drives have no real cylinder/head/sector addressing but for compatibility reasons this is simulated by the driver to give a uniform interface to the operating system.
There is a plethora of interfaces to chose from widely ranging in price and performance. Most motherboards today include IDE interface which are part of modern chipsets.
Many motherboards also include a SCSI interface chip made by Symbios (formerly NCR) and that is connected directly to the PCI bus. Check what you have and what BIOS support you have with it.
Once upon a time this was the established technology, a time when 20 MB was awesome, which compared to todays sizes makes you think that dinosaurs roamed the Earth with these drives. Like the dinosaurs these are outdated and are slow and unreliable compared to what we have today. Linux does support this but you are well advised to think twice about what you would put on this. One might argue that an emergency partition with a suitable vintage of DOS might be fitting.
Actually, ESDI was an adaptation of the very widely used SMD interface used on "big" computers to the cable set used with the ST506 interface, which was more convenient to package than the 60-pin + 26-pin connector pair used with SMD. The ST506 was a "dumb" interface which relied entirely on the controller and host computer to do everything from computing head/cylinder/sector locations and keeping track of the head location, etc. ST506 required the controller to extract clock from the recovered data, and control the physical location of detailed track features on the medium, bit by bit. It had about a 10-year life if you include the use of MFM, RLL, and ERLL/ARLL modulation schemes. ESDI, on the other hand, had intelligence, often using three or four separate microprocessors on a single drive, and high-level commands to format a track, transfer data, perform seeks, and so on. Clock recovery from the data stream was accomplished at the drive, which drove the clock line and presented its data in NRZ, though error correction was still the task of the controller. ESDI allowed the use of variable bit density recording, or, for that matter, any other modulation technique, since it was locally generated and resolved at the drive. Though many of the techniques used in ESDI were later incorporated in IDE, it was the increased popularity of SCSI which led to the demise of ESDI in computers. ESDI had a life of about 10 years, though mostly in servers and otherwise "big" systems rather than PC's.
Progress made the drive electronics migrate from the ISA slot card over to the drive itself and Integrated Drive Electronics was borne. It was simple, cheap and reasonably fast so the BIOS designers provided the kind of snag that the computer industry is so full of. A combination of an IDE limitation of 16 heads together with the BIOS limitation of 1024 cylinders gave us the infamous 504 MB limit. Following the computer industry traditions again, the snag was patched with a kludge and we got all sorts of translation schemes and BIOS bodges. This means that you need to read the installation documentation very carefully and check up on what BIOS you have and what date it has as the BIOS has to tell Linux what size drive you have. Fortunately with Linux you can also tell the kernel directly what size drive you have with the drive parameters, check the documentation for LILO and Loadlin, thoroughly. Note also that IDE is equivalent to ATA, AT Attachment. IDE uses CPU-intensive Programmed Input/Output (PIO) to transfer data to and from the drives and has no capability for the more efficient Direct Memory Access (DMA) technology. Highest transfer rate is 8.3 MB/s.
These 3 terms are roughly equivalent, fast-ATA is ATA-2 but EIDE additionally includes ATAPI. ATA-2 is what most use these days which is faster and with DMA. Highest transfer rate is increased to 16.6 MB/s.
A new, faster DMA mode that is approximately twice the speed of EIDE PIO-Mode 4 (33 MB/s). Disks with and without Ultra-ATA can be mixed on the same cable without speed penalty for the faster adapters. The Ultra-ATA interface is electrically identical with the normal Fast-ATA interface, including the maximum cable length.
The ATA/66 was superceeded by ATA/100 and very recently we have now gotten ATA/133. While the interface speed has iproved dramatically the disks are often limited by platter-to-cache limites which today stands at about 40 MB/s.
For more information read up on these overviews and whitepapers from Maxtor: Fast Drives Technology on the ATA/133 interface and Big Drives Technology on breaking the 137 GB limit.
A new, standard has been agreed upon, the Serial-ATA
interface, backed by the
The Serial ATA
group who made the announcement in August 2001.
Advantages are numerous: simple, thin connectors rather than old cumbersome cable mats that also obstructued air flow, higher speeds (about 150 MB/s) and backward compatibility.
The ATA Packet Interface was designed to support CD-ROM drives using the IDE port and like IDE it is cheap and simple.
The Small Computer System Interface is a multi purpose interface that can be used to connect to everything from drives, disk arrays, printers, scanners and more. The name is a bit of a misnomer as it has traditionally been used by the higher end of the market as well as in work stations since it is well suited for multi tasking environments.
The standard interface is 8 bits wide and can address 8 devices. There is a wide version with 16 bit that is twice as fast on the same clock and can address 16 devices. The host adapter always counts as a device and is usually number 7. It is also possible to have 32 bit wide busses but this usually requires a double set of cables to carry all the lines.
The old standard was 5 MB/s and the newer fast-SCSI increased this to 10 MB/s. Recently ultra-SCSI, also known as Fast-20, arrived with 20 MB/s transfer rates for an 8 bit wide bus. New low voltage differential (LVD) signalling allows these high speeds as well as much longer cabling than before.
Even more recently an even faster standard has been introduced: SCSI 160 (originally named SCSI 160/m) which is capable of a monstrous 160 MB/s over a 16 bit wide bus. Support is scarce yet but for a few 10000 RPM drives that can transfer 40 MB/s sustained. Putting 6 such drives on a RAID will keep such a bus saturated and also saturate most PCI busses. Obviously this is only for the very highest end servers per today. More information on this standard is available at The Ultra 160 SCSI home page
Adaptec just announced a Linux driver for their SCSI 160 host adapter. More information will come when more information becomes available.
Now also SCSI/320 is available.
The higher performance comes at a cost that is usually higher than for (E)IDE. The importance of correct termination and good quality cables cannot be overemphasized. SCSI drives also often tend to be of a higher quality than IDE drives. Also adding SCSI devices tend to be easier than adding more IDE drives: Often it is only a matter of plugging or unplugging the device; some people do this without powering down the system. This feature is most convenient when you have multiple systems and you can just take the devices from one system to the other should one of them fail for some reason.
There is a number of useful documents you should read if you use SCSI, the SCSI HOWTO as well as the SCSI FAQ posted on Usenet News.
SCSI also has the advantage you can connect it easily to tape drives for backing up your data, as well as some printers and scanners. It is even possible to use it as a very fast network between computers while simultaneously share SCSI devices on the same bus. Work is under way but due to problems with ensuring cache coherency between the different computers connected, this is a non trivial task.
SCSI numbers are also used for arbitration. If several drives request service, the drive with the lowest number is given priority.
Note that newer SCSI cards will simultaneously support an array of different types of SCSI devices all at individually optimized speeds.
I do not intend to make too many comments on hardware but I feel I should make a little note on cabling. This might seem like a remarkably low technological piece of equipment, yet sadly it is the source of many frustrating problems. At todays high speeds one should think of the cable more of a an RF device with its inherent demands on impedance matching. If you do not take your precautions you will get a much reduced reliability or total failure. Some SCSI host adapters are more sensitive to this than others.
Shielded cables are of course better than unshielded but the price is much higher. With a little care you can get good performance from a cheap unshielded cable.
Cable Select (ATA drives) you set the drive jumpers
to cable select and use the cable to determine master and slave. This
is not much used.
Bus Speed (MHz) | Max Length (m)
--------------------------------------------------
5 | 6
10 (fast) | 3
20 (fast-20 / ultra) | 3 (max 4 devices), 1.5 (max 8 devices)
xx (differential) | 25 (max 16 devices
--------------------------------------------------
More information on SCSI cabling and termination can be found at various web pages around the net.
This is the other end of the interface from the drive, the part that is connected to a computer bus. The speed of the computer bus and that of the drives should be roughly similar, otherwise you have a bottleneck in your system. Connecting a RAID 0 disk-farm to a ISA card is pointless. These days most computers come with 32 bit PCI bus capable of 132 MB/s transfers which should not represent a bottleneck for most people in the near future.
As the drive electronic migrated to the drives the remaining part that became the (E)IDE interface is so small it can easily fit into the PCI chip set. The SCSI host adapter is more complex and often includes a small CPU of its own and is therefore more expensive and not integrated into the PCI chip sets available today. Technological evolution might change this.
Some host adapters come with separate caching and intelligence but as this is basically second guessing the operating system the gains are heavily dependent on which operating system is used. Some of the more primitive ones, that shall remain nameless, experience great gains. Linux, on the other hand, have so much smarts of its own that the gains are much smaller.
Mike Neuffer, who did the drivers for the DPT controllers, states that the DPT controllers are intelligent enough that given enough cache memory it will give you a big push in performance and suggests that people who have experienced little gains with smart controllers just have not used a sufficiently intelligent caching controller.
In order to increase throughput it is necessary to identify the most significant bottlenecks and then eliminate them. In some systems, in particular where there are a great number of drives connected, it is advantageous to use several controllers working in parallel, both for SCSI host adapters as well as IDE controllers which usually have 2 channels built in. Linux supports this.
Some RAID controllers feature 2 or 3 channels and it pays to spread the disk load across all channels. In other words, if you have two SCSI drives you want to RAID and a two channel controller, you should put each drive on separate channels.
In addition to having both a SCSI and an IDE in the same machine it is also possible to have more than one SCSI controller. Check the SCSI-HOWTO on what controllers you can combine. Also you will most likely have to tell the kernel it should probe for more than just a single SCSI or a single IDE controller. This is done using kernel parameters when booting, for instance using LILO. Check the HOWTOs for SCSI and LILO for how to do this.
Multi board systems can offer significant speed gains if you
configure your disks right, especially for RAID0. Make sure you
interleave the controllers as well as the drives, so that you
add drives to the md RAID device in the right order.
If controller 1 is connected to drives sda and sdc
while controller 2 is connected to drives sdb and sdd
you will gain more paralellicity by adding in the order of
sda - sdc - sdb - sdd rather than sda - sdb - sdc - sdd
because a read or write over more than one cluster will be more
likely to span two controllers.
The same methods can also be applied to IDE. Most motherboards come with typically 4 IDE ports:
hda primary masterhdb primary slavehdc secondary masterhdd secondary slavehda - hdc - hdb - hdd
as this will most likely parallelise both channels.
The following tables are given just to indicate what speeds are possible but remember that these are the theoretical maximum speeds. All transfer rates are in MB per second and bus widths are measured in bits.
IDE : 8.3 - 16.7
Ultra-ATA : 33 - 66
SCSI :
Bus width (bits)
Bus Speed (MHz) | 8 16 32
--------------------------------------------------
5 | 5 10 20
10 (fast) | 10 20 40
20 (fast-20 / ultra) | 20 40 80
40 (fast-40 / ultra-2) | 40 80 --
--------------------------------------------------
ISA : 8-12
EISA : 33
VESA : 40 (Sometimes tuned to 50)
PCI
Bus width (bits)
Bus Speed (MHz) | 32 64
--------------------------------------------------
33 | 132 264
66 | 264 528
--------------------------------------------------
This is a very, very difficult topic and I will only make a few cautious comments about this minefield. First of all, it is more difficult to make comparable benchmarks that have any actual meaning. This, however, does not stop people from trying...
Instead one can use benchmarking to diagnose your own system, to check it is going as fast as it should, that is, not slowing down. Also you would expect a significant increase when switching from a simple file system to RAID, so a lack of performance gain will tell you something is wrong.
When you try to benchmark you should not hack up your own, instead
look up iozone and bonnie and read the documentation very
carefully. In particular make sure your buffer size is bigger than
your RAM size, otherwise you test your RAM rather than your disks
which will give you unrealistically high performance.
A very simple benchmark can be obtained using hdparm -tT which
can be used both on IDE and SCSI drives.
For more information on benchmarking and software for a number of platforms, check out ACNC benchmark page as well as this one and also The Benchmarking-HOWTO.
There are also official home pages for bonnie, bonnie++ and iozone.
Trivia: Bonnie is intended to locate bottlenecks, the name is a tribute to Bonnie Raitt, "who knows how to use one" as the author puts it.
SCSI offers more performance than EIDE but at a price. Termination is more complex but expansion not too difficult. Having more than 4 (or in some cases 2) IDE drives can be complicated, with wide SCSI you can have up to 15 per adapter. Some SCSI host adapters have several channels thereby multiplying the number of possible drives even further.
For SCSI you have to dedicate one IRQ per host adapter which can control up to 15 drives. With EIDE you need one IRQ for each channel (which can connect up to 2 disks, master and slave) which can cause conflict.
RLL and MFM is in general too old, slow and unreliable to be of much use.
SCSI-3 is under way and will hopefully be released soon. Faster devices are already being announced, recently an 80 MB/s and then a 160 MB/s monster specification has been proposed and also very recently became commercially available. These are based around the Ultra-2 standard (which used a 40 MHz clock) combined with a 16 bit cable.
Some manufacturers already announce SCSI-3 devices but this is currently rather premature as the standard is not yet firm. As the transfer speeds increase the saturation point of the PCI bus is getting closer. Currently the 64 bit version has a limit of 264 MB/s. The PCI transfer rate will in the future be increased from the current 33 MHz to 66 MHz, thereby increasing the limit to 528 MB/s.
The ATA development is continuing and is increasing the performance with the new ATA/100 standard. Since most ATA drives are slower in sustained transfer from platter than this the performance increase will for most people be small.
More interesting is the Serial ATA development, where the flat cable will be replaced with a high speed serial link. This makes cabling far simpler than today and also it solves the problem of cabling obstructing airflow over the drives.
Another trend is for larger and larger drives. I hear it is possible to get 75 GB on a single drive though this is rather expensive. Currently the optimum storage for your money is about 30 GB but also this is continuously increasing. The introduction of DVD will in the near future have a big impact, with nearly 20 GB on a single disk you can have a complete copy of even major FTP sites from around the world. The only thing we can be reasonably sure about the future is that even if it won't get any better, it will definitely be bigger.
Addendum: soon after I first wrote this I read that the maximum useful speed for a CD-ROM was 20x as mechanical stability would be too great a problem at these speeds. About one month after that again the first commercial 24x CD-ROMs were available... Currently you can get 40x and no doubt higher speeds are in the pipeline.
A project to encapsulate SCSI over TCP/IP, called iSCSI has started, and one Linux iSCSI implementation has appeared.
My personal view is that EIDE or Ultra ATA is the best way to start out on your system, especially if you intend to use DOS as well on your machine. If you plan to expand your system over many years or use it as a server I would strongly recommend you get SCSI drives. Currently wide SCSI is a little more expensive. You are generally more likely to get more for your money with standard width SCSI. There is also differential versions of the SCSI bus which increases maximum length of the cable. The price increase is even more substantial and cannot therefore be recommended for normal users.
In addition to disk drives you can also connect some types of scanners and printers and even networks to a SCSI bus.
Also keep in mind that as you expand your system you will draw ever more power, so make sure your power supply is rated for the job and that you have sufficient cooling. Many SCSI drives offer the option of sequential spin-up which is a good idea for large systems. See also Power and Heating.
Linux has been multi tasking from the very beginning where a number of programs interact and run continuously. It is therefore important to keep a file structure that everyone can agree on so that the system finds data where it expects to. Historically there has been so many different standards that it was confusing and compatibility was maintained using symbolic links which confused the issue even further and the structure ended looking like a maze.
In the case of Linux a standard was fortunately agreed on early on called the File Systems Standard (FSSTND) which today is used by all main Linux distributions.
Later it was decided to make a successor that should also support operating systems other than just Linux, called the Filesystem Hierarchy Standard (FHS) at version 2.2 currently. This standard is under continuous development and will soon be adopted by Linux distributions.
I recommend not trying to roll your own structure as a lot of thought has gone into the standards and many software packages comply with the standards. Instead you can read more about this at the FHS home page.
This HOWTO endeavours to comply with FSSTND and will follow FHS when distributions become available.
The various parts of FSSTND have different requirements regarding
speed, reliability and size, for instance losing root is a pain
but can easily be recovered. Losing /var/spool/mail is a
rather different issue. Here is a quick summary of some essential
parts and their properties and requirements. Note that this is
just a guide, there can be binaries in etc and
lib directories, libraries in bin directories
and so on.
Maximum! Though if you rely too much on swap you should consider buying some more RAM. Note, however, that on many old Pentium PC motherboards the cache will not work on RAM above 128 MB.
Similar as for RAM. Quick and dirty algorithm: just as for tea: 16 MB for the machine and 2 MB for each user. Smallest kernel run in 1 MB but is tight, use 4 MB for general work and light applications, 8 MB for X11 or GCC or 16 MB to be comfortable. (The author is known to brew a rather powerful cuppa tea...)
Some suggest that swap space should be 1-2 times the size of the RAM, pointing out that the locality of the programs determines how effective your added swap space is. Note that using the same algorithm as for 4BSD is slightly incorrect as Linux does not allocate space for pages in core.
A more thorough approach is to consider swap space plus RAM as your total working set, so if you know how much space you will need at most, you subtract the physical RAM you have and that is the swap space you will need.
There is also another reason to be generous when dimensioning your swap space: memory leaks. Ill behaving programs that do not free the memory they allocate for themselves are said to have a memory leak. This allocation remains even after the offending program has stopped so this is a source of memory consumption. Only after the program dies is the memory returned. Once all physical RAM and swap space are exhausted the only solution is to kill the offending processes if possible, or failing that, reboot and start over. Thankfully such programs are not too common but should you come across one you will find that extra swap space will buy you extra time between reboots.
Also remember to take into account the type of programs you use. Some programs that have large working sets, such as image processing software have huge data structures loaded in RAM rather than working explicitly on disk files. Data and computing intensive programs like this will cause excessive swapping if you have less RAM than the requirements.
Other types of programs can lock their pages into RAM. This can be for security reasons, preventing copies of data reaching a swap device or for performance reasons such as in a real time module. Either way, locking pages reduces the remaining amount of swappable memory and can cause the system to swap earlier then otherwise expected.
In man 8 mkswap it is explained that each swap partition can
be a maximum of just under 128 MB in size for 32-bit machines
and just under 256 MB for 64-bit machines.
This however changed with kernel 2.2.0 after which the limit is 2 GB. The man page has been updated to reflect this change.
Medium. When it fails you know it pretty quickly and failure will cost you some lost work. You save often, don't you?
Linux offers the possibility of interleaved swapping
across multiple devices, a feature that can gain you much. Check out
"man 8 swapon" for more details. However, software raiding
swap across multiple devices adds more overheads than you gain.
Thus the /etc/fstab file might look like this:
/dev/sda1 swap swap pri=1 0 0
/dev/sdc1 swap swap pri=1 0 0
fstab file is very sensitive to the formatting
used, read the man page carefully and do not just cut and paste
the lines above.
Some people use a RAM disk for swapping or some other file systems. However, unless you have some very unusual requirements or setups you are unlikely to gain much from this as this cuts into the memory available for caching and buffering.
There is once exception: on a number of badly designed motherboards the on board cache memory is not able to cache all the RAM that can be addressed. Many older motherboards could accept 128 MB RAM but only cache the lower 64 MB. In such cases it would improve the performance if you used the upper (uncached) 64 MB RAM for RAMdisk based swap or other temporary storage.
/tmp and /var/tmp)
Very high. On a separate disk/partition this will
reduce fragmentation generally, though ext2fs handles fragmentation
rather well.
Hard to tell, small systems are easy to run with just
a few MB but these are notorious hiding places for stashing files
away from prying eyes and quota enforcement and can grow without
control on larger machines. Suggested: small home machine: 8 MB,
large home machine: 32 MB, small server: 128 MB, and large
machines up to 500 MB (The machine used by the author at work has 1100
users and a 300 MB /tmp directory). Keep an eye on these directories,
not only for hidden files but also for old files. Also be prepared that
these partitions might be the first reason you might have to resize
your partitions.
Low. Often programs will warn or fail gracefully when these areas fail or are filled up. Random file errors will of course be more serious, no matter what file area this is.
Mostly short files but there can be a huge number of
them. Normally programs delete their old tmp files but if somehow an
interruption occurs they could survive. Many distributions have a policy
regarding cleaning out tmp files at boot time, you might want to
check out what your setup is.
In FSSTND there is a note about putting /tmp on
RAM disk. This, however, is not recommended for the same reasons
as stated for swap. Also, as noted earlier, do not use flash RAM
drives for these directories. One should also keep in mind that some
systems are set to automatically clean tmp areas on rebooting.
Older systems had a /usr/tmp but this is no longer
recommended and for historical reasons a symbolic link now makes it
point to one of the other tmp areas.
(* That was 50 lines, I am home and dry! *)
/var/spool/news and /var/spool/mail)
High, especially on large news servers. News transfer and expiring are disk intensive and will benefit from fast drives. Print spools: low. Consider RAID0 for news.
For news/mail servers: whatever you can afford. For
single user systems a few MB will be sufficient if you read
continuously. Joining a list server and taking a holiday is, on the
other hand, not a good idea. (Again the machine I use at work
has 100 MB reserved for the entire /var/spool)
Mail: very high, news: medium, print spool: low. If your mail is very important (isn't it always?) consider RAID for reliability.
Usually a huge number of files that are around a few KB in size. Files in the print spool can on the other hand be few but quite sizable.
Some of the news documentation suggests putting all
the .overview files on a drive separate from the news
files, check out all news FAQs for more information.
Typical size is about 3-10 percent of total news spool size.
/home)
Medium. Although many programs use /tmp for temporary
storage, others such as some news readers frequently update files in the
home directory which can be noticeable on large multiuser systems. For
small systems this is not a critical issue.
Tricky! On some systems people pay for storage so this is usually then a question of finance. Large systems such as Nyx.net (which is a free Internet service with mail, news and WWW services) run successfully with a suggested limit of 100 KB per user and 300 KB as enforced maximum. Commercial ISPs offer typically about 5 MB in their standard subscription packages.
If however you are writing books or are doing design work the requirements balloon quickly.
Variable. Losing /home on a single user machine is
annoying but when 2000 users call you to tell you their home
directories are gone it is more than just annoying. For some their
livelihood relies on what is here. You do regular backups of course?
Equally tricky. The minimum setup for a single user tends to be a dozen files, 0.5 - 5 KB in size. Project related files can be huge though.
You might consider RAID for either speed or reliability. If you want extremely high speed and reliability you might be looking at other operating system and hardware platforms anyway. (Fault tolerance etc.)
Web browsers often use a local cache to speed up browsing and this cache can take up a substantial amount of space and cause much disk activity. There are many ways of avoiding this kind of performance hits, for more information see the sections on Home Directories and WWW.
Users often tend to use up all available space on the
/home partition. The Linux Quota subsystem is capable of
limiting the number of blocks and the number of inode a single user
ID can allocate on a per-filesystem basis. See the
Linux Quota mini-HOWTO by
Albert M.C. Tam bertie (at) scn.org
for details on setup.
/usr/bin and /usr/local/bin)
Low. Often data is bigger than the programs which are demand loaded anyway so this is not speed critical. Witness the successes of live file systems on CD ROM.
The sky is the limit but 200 MB should give you most of what you want for a comprehensive system. A big system, for software development or a multi purpose server should perhaps reserve 500 MB both for installation and for growth.
Low. This is usually mounted under root where all the essentials are collected. Nevertheless losing all the binaries is a pain...
Variable but usually of the order of 10 - 100 KB.
/usr/lib and /usr/local/lib)
Medium. These are large chunks of data loaded often, ranging from object files to fonts, all susceptible to bloating. Often these are also loaded in their entirety and speed is of some use here.
Variable. This is for instance where word processors
store their immense font files. The few that have given me feedback on
this report about 70 MB in their various lib directories.
A rather complete Debian 1.2 installation can take as much as
250 MB which can be taken as an realistic upper limit.
The following ones are some of the largest disk space consumers:
GCC, Emacs, TeX/LaTeX, X11 and perl.
Low. See point Main binaries.
Usually large with many of the order of 1 MB in size.
For historical reasons some programs keep executables in
the lib areas. One example is GCC which have some huge binaries in the
/usr/lib/gcc/lib hierarchy.
Quite low: after all booting doesn't happen that often and loading the kernel is just a tiny fraction of the time it takes to get the system up and running.
Quite small, a complete image with some extras fit on a single floppy so 5 MB should be plenty.
High. See section below on Root.
The most important part about the Boot partition is that on many systems it must reside below cylinder 1023. This is a BIOS limitation that Linux cannot get around.
The above is not necessarily true for recent IDE systems and not for any SCSI disks. For more information check the latest Large Disk HOWTO.
Recently a new boot loader has been written that overcomes the 1023 sector limit. For more information check out this article on nuni.
Quite low: only the bare minimum is here, much of which is only run at startup time.
Relatively small. However it is a good idea to keep some essential rescue files and utilities on the root partition and some keep several kernel versions. Feedback suggests about 20 MB would be sufficient.
High. A failure here will possibly cause a fair bit of grief and you might end up spending some time rescuing your boot partition. With some practice you can of course do this in an hour or so, but I would think if you have some practice doing this you are also doing something wrong.
Naturally you do have a rescue disk? Of course this is updated since you did your initial installation? There are many ready made rescue disks as well as rescue disk creation tools you might find valuable. Presumably investing some time in this saves you from becoming a root rescue expert.
If you have plenty of drives you might consider putting a spare emergency boot partition on a separate physical drive. It will cost you a little bit of space but if your setup is huge the time saved, should something fail, will be well worth the extra space.
For simplicity and also in case of emergencies
it is not advisable to put the root partition on a RAID level 0 system.
Also if you use RAID for your boot partition you have to remember to
have the md option turned on for your emergency kernel.
For simplicity it is quite common to keep Boot and Root
on the same partition. If you do that, then
in order to boot from LILO it is important that the
essential boot files reside wholly within cylinder 1023. This includes
the kernel as well as files found in /boot.
At the danger of sounding heretical I have included this little section about something many reading this document have strong feelings about. Unfortunately many hardware items come with setup and maintenance tools based around those systems, so here goes.
Very low. The systems in question are not famed for speed so there is little point in using prime quality drives. Multitasking or multi-threading are not available so the command queueing facility found in SCSI drives will not be taken advantage of. If you have an old IDE drive it should be good enough. The exception is to some degree Win95 and more notably NT which have multi-threading support which should theoretically be able to take advantage of the more advanced features offered by SCSI devices.
The company behind these operating systems is not famed for writing tight code so you have to be prepared to spend a few tens of MB depending on what version you install of the OS or Windows. With an old version of DOS or Windows you might fit it all in on 50 MB.
Ha-ha. As the chain is no stronger than the weakest link you can use any old drive. Since the OS is more likely to scramble itself than the drive is likely to self destruct you will soon learn the importance of keeping backups here.
Put another way: "Your mission, should you choose to accept it, is to keep this partition working. The warranty will self destruct in 10 seconds..."
Recently I was asked to justify my claims here. First of all I am not calling DOS and Windows sorry excuses for operating systems. Secondly there are various legal issues to be taken into account. Saying there is a connection between the last two sentences are merely the ravings of the paranoid. Surely. Instead I shall offer the esteemed reader a few key words: DOS 4.0, DOS 6.x and various drive compression tools that shall remain nameless.
Naturally the faster the better but often the happy installer of Linux has several disks of varying speed and reliability so even though this document describes performance as 'fast' and 'slow' it is just a rough guide since no finer granularity is feasible. Even so there are a few details that should be kept in mind:
This is really a rather woolly mix of several terms: CPU load, transfer setup overhead, disk seek time and transfer rate. It is in the very nature of tuning that there is no fixed optimum, and in most cases price is the dictating factor. CPU load is only significant for IDE systems where the CPU does the transfer itself but is generally low for SCSI, see SCSI documentation for actual numbers. Disk seek time is also small, usually in the millisecond range. This however is not a problem if you use command queueing on SCSI where you then overlap commands keeping the bus busy all the time. News spools are a special case consisting of a huge number of normally small files so in this case seek time can become more significant.
There are two main parameters that are of interest here:
is usually specified in the average time take for the read/write head to seek from one track to another. This parameter is important when dealing with a large number of small files such as found in spool files. There is also the extra seek delay before the desired sector rotates into position under the head. This delay is dependent on the angular velocity of the drive which is why this parameter quite often is quoted for a drive. Common values are 4500, 5400 and 7200 RPM (rotations per minute). Higher RPM reduces the seek time but at a substantial cost. Also drives working at 7200 RPM have been known to be noisy and to generate a lot of heat, a factor that should be kept in mind if you are building a large array or "disk farm". Very recently drives working at 10000 RPM has entered the market and here the cooling requirements are even stricter and minimum figures for air flow are given.
is usually specified in megabytes per second. This parameter is important when handling large files that have to be transferred. Library files, dictionaries and image files are examples of this. Drives featuring a high rotation speed also normally have fast transfers as transfer speed is proportional to angular velocity for the same sector density.
It is therefore important to read the specifications for the drives very carefully, and note that the maximum transfer speed quite often is quoted for transfers out of the on board cache (burst speed) and not directly from the platter (sustained speed). See also section on Power and Heating.
Naturally no-one would want low reliability disks but one might be better off regarding old disks as unreliable. Also for RAID purposes (See the relevant information) it is suggested to use a mixed set of disks so that simultaneous disk crashes become less likely.
So far I have had only one report of total file system failure but here unstable hardware seemed to be the cause of the problems.
Disks are cheap these days yet people still underestimate the value of the contents of the drives. If you need higher reliability make sure you replace old drives and keep spares. It is not unusual that drives can work more or less continuous for years and years but what often kills a drive in the end is power cycling.
The average file size is important in order to decide the most suitable drive parameters. A large number of small files makes the average seek time important whereas for big files the transfer speed is more important. The command queueing in SCSI devices is very handy for handling large numbers of small files, but for transfer EIDE is not too far behind SCSI and normally much cheaper than SCSI.
Over time the requirements for file systems have increased and the demands for large structures, large files, long file names and more has prompted ever more advanced file systems, the system that accesses and organises the data on mass storage. Today there is a large number of file systems to choose from and this section will describe these in detail.
The emphasis is on Linux but with more input I will be happy to add information for a wider audience.
Most operating systems usually have a general purpose file system for every day use for most kinds of files, reflecting available features in the OS such as permission flags, protection and recovery.
minixThis was the original fs for Linux, back in the days Linux was hosted on minix machines. It is simple but limited in features and hardly ever used these days other than in some rescue disks as it is rather compact.
xiafs and extfsThese are also old and have fallen in disuse and are no longer recommended.
ext2fsThis is the established standard for general purpose in the Linux world. It is fast, efficient and mature and is under continuous development and features such as ACL and transparent compression are on the horizon.
For more information check the ext2fs home page.
ext3fs
This is the name for the upcoming successor to ext2fs due to enter
stable kernel in the near future. Many features are added to
ext2fs but to avoid confusion over the name after such a radical
upgrade the name will be changed too. You may have heard of it already
but source code is now in beta release .
Patches are available at Linux.org.
ufsThis is the fs used by BSD and variants thereof. It is mature but also developed for older types of disk drives where geometries were known. The fs uses a number of tricks to optimise performance but as disk geometries are translated in a number of ways the net effect is no longer so optimal.
efsThe Extent File System (efs) is Silicon Graphics' early file system widely used on IRIX before version 6.0 after which xfs has taken over. While migration to xfs is encouraged efs is still supported and much used on CDs.
There is a Linux driver available in early beta stage, available at Linux extent file system home page.
XFSSilicon Graphics Inc (sgi) has started porting its mainframe grade file system to Linux. Source is not yet available as they are busily cleaning out legal encumbrance but once that is done they will provide the source code under GPL.
More information is already available on the XFS project page at SGI.
reiserfs
As of July, 23th 1997
Hans Reiser reiser (at) RICOCHET.NET
has put up the source to his tree based
reiserfs
on the web. While his filesystem has some very interesting features and
is much faster than ext2fs and is in use by a number of people.
Hopefully it will be ready for kernel 2.4.0 which might be ready at
the end of the year.
enh-fsThe Enhanced File System project is now dead.
Tux2 fs
This is a variation on the ext2fs that adds robustness
in case of unexpected interruptions such as power failure.
After such an event Tux2 fs will restart with the file system
in a consistent, recently recorded state without fsck or
other recovery operations. To achieve this Tux2 fs uses
a newly designed algorithm called Phase Tree.
More information can be found at the project home page.
This company is responsible for a lot, including a number of filesystems that has at the very least caused confusions.
fat
Actually there are 2 fats out there, fat12 and fat16
depending on the partition size used but fortunately the difference
is so minor that the whole issue is transparent.
On the plus side these are fast and simple and most OSes understands it and can both read and write this fs. And that is about it.
The minus side is limited safety, severely limited permission flags
and atrocious scalability. For instance with fat you cannot
have partitions larger than 2 GB.
fat32
After about 10 years Microsoft realised fat was about, well, 10 years
behind the times and created this fs which scales reasonably well.
Permission flags are still limited. NT 4.0 cannot read this file system but Linux can.
vfat
At the same time as Microsoft launched fat32 they also added
support for long file names, known as vfat.
Linux reads vfat and fat32 partitions by mounting with
type vfat.
ntfsThis is the native fs of Win-NT but as complete information is not available there is limited support for other OSes.
These take a radically different approach to file updates by logging modifications for files in a log and later at some time checkpointing the logs.
Reading is roughly as fast as traditional file systems that always update the files directly. Writing is much faster as only updates are appended to a log. All this is transparent to the user. It is in reliability and particularly in checking file system integrity that these file systems really shine. Since the data before last checkpointing is known to be good only the log has to be checked, and this is much faster than for traditional file systems.
Note that while logging filesystems keep track of changes made to both data and inodes, journaling filesystems keep track only of inode changes.
Linux has quite a choice in such file systems but none are yet in production quality. Some are also on hold.
Note that ext3fs, XFS and reiserfs also have
features for logging or journaling.
Read-only media has not escaped the ever increasing complexities seen in more general file systems so again there is a large choice to choose from with corresponding opportunities for exciting mistakes.
Note that ext2fs works quite well on a CD-ROM
and seems to save space while offering the normal file system features
such as long file names and permissions that can be retained when
copying files across to read-write media. Also having
/dev
on a CD-ROM is possible.
Most of these are used with the CD-ROM media but also the new DVD can be used and you can even use it through the loopback device on a hard disk file for verifying an image before burning a ROM.
There is a read-only romfs for Linux but as that is not disk
related nothing more will be said about it here.
High SierraThis was one of the earliest standards for CD-ROM formats, supposedly named after the hotel where the final agreement took place.
High Sierra was so limited in features that new extensions simply
had to appear and while there has been no end to new formats the original
High Sierra remains the common precursor and is therefore still
widely supported.
iso9660
The International Standards Organisation made their extensions and
formalised the standard into what we know as the iso9660 standard.
The Linux iso9660 file system supports both High Sierra as well as
Rock Ridge extensions.
Rock Ridge
Not everyone accepts limits like short filenames and lack of permissions
so very soon the Rock Ridge extensions appeared to rectify these
shortcomings.
Joliet
Microsoft, not be be outdone in the standards extension game, decided
it should extend CD-ROM formats with some internationalisation features
and called it Joliet.
Linux supports this standards in kernels 2.0.34 or newer. You need to enable NLS in order to use it.
Joliet is a city outside Chicago; best known for being the site of the prison where Jake was locked up in the movie "Blues Brothers." Rock Ridge (the UNIX extensions to ISO 9660) is named after the (fictional) town in the movie "Blazing Saddles."
UDF
With the arrival of DVD with up to about 17 GB of storage capacity
the world seemingly needed another format, this time ambitiously named
Universal Disk Format (UDF).
This is intended to replace iso9660 and will be required for DVD.
Currently this is not in the standard Linux kernel but a project is underway to make a http://trylinux.com/projects/udf/index.html name="UDF driver"> for Linux. Patches and documentation are available.
More information is also available at the Linux and DVDs page.
There is a large number of networking technologies available that lets you distribute disks throughout a local or even global networks. This is somewhat peripheral to the topic of this HOWTO but as it can be used with local disks I will cover this briefly. It would be best if someone (else) took this into a separate HOWTO...
NFS
This is one of the earliest systems that allows mounting a file space
on one machine onto another. There are a number of problems with NFS
ranging from performance to security but it has nevertheless become
established.
AFSThis is a system that allows efficient sharing of files across large networks. Starting out as an academic project it is now sold by Transarc whose home page gives you more details.
Derek Atkins, of MIT, ported AFS to Linux and has also set up the Linux AFS mailing List ( linux-afs@mit.edu) for this which is open to the public. Requests to join the list should go to linux-afs-request@mit.edu and finally bug reports should be directed to linux-afs-bugs@mit.edu.
Important: as AFS uses encryption it is restricted software and cannot easily be exported from the US.
IBM who owns Transarc, has announced the availability of the latest version of client as well as server for Linux.
Arla is a free AFS implementation, check the Arla homepage for more information as well as documentation.
A networking filesystem similar to AFS is underway and is called
Coda.
This is designed to be more robust and fault tolerant than AFS,
and supports mobile, disconnected operations.
Currently it does not scale very well, and does not really have
proper administrative tools, as AFS does and ARLA is
beginning to.
nbd
The
Network Block Device
(nbd) is available in Linux kernel 2.2
and later and offers reportedly excellent performance. The interesting
thing here is that it can be combined with RAID (see later).
enbd
The
http://www.it.uc3m.es/~ptb/nbd
name="Enhanced Network Block Device">
(enbd) is a project to enhance the nbd with
features such as block journaled multi channel communications,
internal failover and automatic balancing between channels
and more.
The intended use is for RAID over the net.
The Global File System is a new file system designed for storage across a wide area network. It is currently in the early stages and more information will come later.
In addition to the general file systems there is also a number of more specific ones, usually to provide higher performance or other features, usually with a tradeoff in other respects.
tmpfs and swapfs
For short term fast file storage SunOS offers tmpfs which is
about the same as the swapfs on NeXT.
This overcomes the inherent slowness in ufs by caching file data
and keeping control information in memory. This means that data on such
a file system will be lost when rebooting and is therefore mainly
suitable for /tmp area but not /var/tmp which is where
temporary data that must survive a reboot, is placed.
SunOS offers very limited tuning for tmpfs and the number of
files is even limited by total physical memory of the machine.
Linux now features tmpfs since kernel version 2.4 and is
enabled by turning on virtual memory file system support (former shm fs).
Under certain circumstances tmpfs can lock up the system in
early kerbel versions, make sure you use version 2.4.6 or later.
userfs
The user file system (userfs) allows a number of extensions to
traditional file system use such as
FTP based file system, compression (arcfs) and fast prototyping
and many other features. The docfs is based on this filesystem.
Check the
userfs homepage
for more information.
devfs
When disks are added, removed or just fail it is likely that
disk device names of the remaining disks will change.
For instance if sdb fails then the old sdc becomes sdb,
the old sdc becomes sdb and so on.
Note that in this case hda, hdb etc will remain unchanged.
Likewise if a new drive is added the reverse may happen.
There is no guarantee that SCSI ID 0 becomes sda and that adding
disks in increasing ID order will just add a new device name without
renaming previous entries, as some SCSI drivers assign from ID 0 and up
while others reverse the scanning order.
Likewise adding a SCSI host adapter can also cause renaming.
Generally device names are assigned in the order they are found.
The source of the problem lies in the limited number of bits available
for major and minor numbering in the device files used to describe the
device itself. You an see these in the
/dev
directory, info
on the numbering and allocation can be found in man MAKEDEV.
Currently there are 2 solutions to this problem in various stages of
development:
works by creating a database of drives and where they belong, check man scsifs and the scsidev home page for more information
is a more long term project aimed at getting around the whole business of device numbering by making the /dev directory a kernel file system in the same way as /proc is. More information will appear as it becomes available.
smugfs
For a number of reasons it is currently difficult to have files
bigger than 2 GB. One file system that tries to overcome this
limit is smugfs which is very fast but also simple. For instance
there are no directories and the block allocation is simple.
It is available as compressed tarred source code and while it worked with kernel version 2.1.85 it is quite possible some work is required to make it fit into newer kernels. Also the low version number (0.0) suggests extra care is required.
There is a jungle of choices but generally it is recommended to
use the general file system that comes with your distribution.
If you use ufs and have some kind of tmpfs available
you should first start off with the general file system to get
an idea of the space requirements and if necessary buy more
RAM to support the size of tmpfs you need. Otherwise you
will end up with mysterious crashes and lost time.
If you use dual boot and need to transfer data between the two
OSes one of the simplest ways is to use an appropriately sized
partition formatted with fat as most systems can reliably
read and write this.
Remember the limit of 2 GB for fat partitions.
For more information of file system interconnectivity you can check out the file system page which has been superseded by file system and the article Kragen's Amazing List of Filesystems.
That guide is being superseded by a HOWTO which is underway and a link will be added when it is ready.
To avoid total havoc with device renaming if a drive fails
check out the scanning order of your system and try to keep
your root system on hda or sda and removable media
such as ZIP drives at the end of the scanning order.
In order to decide how to get the most of your devices you need to know what technologies are available and their implications. As always there can be some tradeoffs with respect to speed, reliability, power, flexibility, ease of use and complexity.
Many of the techniques described below can be stacked in a number of ways to maximise performance and reliability, though at the cost of added complexity.
This is a method of increasing reliability, speed or both by using multiple disks in parallel thereby decreasing access time and increasing transfer speed. A checksum or mirroring system can be used to increase reliability. Large servers can take advantage of such a setup but it might be overkill for a single user system unless you already have a large number of disks available. See other documents and FAQs for more information.
For Linux one can set up a RAID system using either software
(the md module in the kernel), a Linux compatible
controller card (PCI-to-SCSI) or a SCSI-to-SCSI controller. Check the
documentation for what controllers can be used. A hardware solution is
usually faster, and perhaps also safer, but comes at a significant cost.
A summary of available hardware RAID solutions for Linux is available at Linux Consulting.
SCSI-to-SCSI controllers are usually implemented as complete cabinets with drives and a controller that connects to the computer with a second SCSI bus. This makes the entire cabinet of drives look like a single large, fast SCSI drive and requires no special RAID driver. The disadvantage is that the SCSI bus connecting the cabinet to the computer becomes a bottleneck.
A significant disadvantage for people with large disk farms is that there is a limit to how many SCSI entries there can be in the /dev directory. In these cases using SCSI-to-SCSI will conserve entries.
Usually they are configured via the front panel or with a terminal connected to their on-board serial interface.
Some manufacturers of such systems are CMD and Syred whose web pages describe several systems.
PCI-to-SCSI controllers are, as the name suggests, connected to the high speed PCI bus and is therefore not suffering from the same bottleneck as the SCSI-to-SCSI controllers. These controllers require special drivers but you also get the means of controlling the RAID configuration over the network which simplifies management.
Currently only a few families of PCI-to-SCSI host adapters are supported under Linux.
The oldest and most mature is a range of controllers from DPT including SmartCache I/III/IV and SmartRAID I/III/IV controller families. These controllers are supported by the EATA-DMA driver in the standard kernel. This company also has an informative home page which also describes various general aspects of RAID and SCSI in addition to the product related information.
More information from the author of the DPT controller drivers (EATA* drivers) can be found at his pages on SCSI and DPT.
These are not the fastest but have a good track record of proven reliability.
Note that the maintenance tools for DPT controllers currently run under DOS/Win only so you will need a small DOS/Win partition for some of the software. This also means you have to boot the system into Windows in order to maintain your RAID system.
A very recent addition is a range of controllers from ICP-Vortex featuring up to 5 independent channels and very fast hardware based on the i960 chip. The Linux driver was written by the company itself which shows they support Linux.
As ICP-Vortex supplies the maintenance software for Linux it is not necessary with a reboot to other operating systems for the setup and maintenance of your RAID system. This saves you also extra downtime.
This is one of the latest entries which is out in early beta. More information as well as drivers are available at Dandelion Digital's Linux DAC960 Page.
Another very recent entry and currently in beta release is the Smart-2 driver.
IBM has released their driver as GPL.
A number of operating systems offer software RAID using ordinary disks and controllers. Cost is low and performance for raw disk IO can be very high. As this can be very CPU intensive it increases the load noticeably so if the machine is CPU bound in performance rather then IO bound you might be better off with a hardware PCI-to-RAID controller.
Real cost, performance and especially reliability of software vs. hardware RAID is a very controversial topic. Reliability on Linux systems have been very good so far.
The current software RAID project on Linux is the md system
(multiple devices) which offers much more than RAID so it is
described in more details later.
RAID comes in many levels and flavours which I will give a brief overview of this here. Much has been written about it and the interested reader is recommended to read more about this in the Software RAID HOWTO.
There are also hybrids available based on RAID 0 or 1 and one other level. Many combinations are possible but I have only seen a few referred to. These are more complex than the above mentioned RAID levels.
RAID 0/1 combines striping with duplication which gives very high transfers combined with fast seeks as well as redundancy. The disadvantage is high disk consumption as well as the above mentioned complexity.
RAID 1/5 combines the speed and redundancy benefits of RAID5 with the fast seek of RAID1. Redundancy is improved compared to RAID 0/1 but disk consumption is still substantial. Implementing such a system would involve typically more than 6 drives, perhaps even several controllers or SCSI channels.
Volume management is a way of overcoming the constraints of fixed sized partitions and disks while still having a control of where various parts of file space resides. With such a system you can add new disks to your system and add space from this drive to parts of the file space where needed, as well as migrating data out from a disk developing faults to other drives before catastrophic failure occurs.
The system developed by Veritas has become the defacto standard for logical volume management.
Volume management is for the time being an area where Linux is lacking.
One is the virtual partition system project VPS that will reimplement many of the volume management functions found in IBM's AIX system. Unfortunately this project is currently on hold.
Another project is the Logical Volume Manager project that is similar to a project by HP.
md Kernel Patch
The Linux Multi Disk (md) provides a number of block level features in various stages of development.
RAID 0 (striping) and concatenation are very solid and in production quality and also RAID 4 and 5 are quite mature.
It is also possible to stack some levels, for instance mirroring (RAID 1) two pairs of drives, each pair set up as striped disks (RAID 0), which offers the speed of RAID 0 combined with the reliability of RAID 1.
In addition to RAID this system offers (in alpha stage) block level
volume management and soon also translucent file space.
Since this is done on the block level it can be used in combination
with any file system, even for fat using Wine.
Think very carefully what drives you combine so you can operate all drives
in parallel, which gives you better performance and less wear. Read more
about this in the documentation that comes with md.
Unfortunately The Linux software RAID has split into two trees, the old stable versions 0.35 and 0.42 which are documented in the official Software-RAID HOWTO and the newer less stable 0.90 series which is documented in the unofficial Software RAID HOWTO which is a work in progress.
A patch for online growth of ext2fs is available in early stages and related work is taking place at the ext2fs resize project at Sourceforge.
Hint: if you cannot get it to work properly you have forgotten to set
the persistent-block flag. Your best documentation is currently
the source code.
Disk compression versus file compression is a hotly debated topic especially regarding the added danger of file corruption. Nevertheless there are several options available for the adventurous administrators. These take on many forms, from kernel modules and patches to extra libraries but note that most suffer various forms of limitations such as being read-only. As development takes place at neck breaking speed the specs have undoubtedly changed by the time you read this. As always: check the latest updates yourself. Here only a few references are given.
e2compr is a package that extends ext2fs with compression
capabilities. It is still under testing and will therefore mainly be of
interest for kernel hackers but should soon gain stability for wider use.
Check the
http://e2compr.memalpha.cx/e2compr/
name="e2compr homepage">
for more information. I have reports of speed and good stability
which is why it is mentioned here.
Access Control List (ACL) offers finer control over file access
on a user by user basis, rather than the traditional owner, group
and others, as seen in directory listings (drwxr-xr-x). This
is currently not available in Linux but is expected in kernel 2.3
as hooks are already in place in ext2fs.
cachefs
This uses part of a hard disk to cache slower media such as CD-ROM. It is available under SunOS but not yet for Linux.
This is a copy-on-write system where writes go to a different system than the original source while making it look like an ordinary file space. Thus the file space inherits the original data and the translucent write back buffer can be private to each user.
There is a number of applications:
SunOS offers this feature and this is under development for Linux.
There was an old project called the Inheriting File Systems (ifs)
but this project has stopped.
One current project is part of the md system and offers
block level translucence so it can be applied to any file system.
Sun has an informative page on translucent file system.
It should be noted that Clearcase (now owned by Rational) pioneered and popularized translucent filesystems for software configuration management by writing their own UNIX filesystem.
This trick used to be very important when drives were slow and small, and some file systems used to take the varying characteristics into account when placing files. Although higher overall speed, on board drive and controller caches and intelligence has reduced the effect of this.
Nevertheless there is still a little to be gained even today. As we know, "world dominance" is soon within reach but to achieve this "fast" we need to employ all the tricks we can use .
To understand the strategy we need to recall this near ancient piece of knowledge and the properties of the various track locations. This is based on the fact that transfer speeds generally increase for tracks further away from the spindle, as well as the fact that it is faster to seek to or from the central tracks than to or from the inner or outer tracks.
Most drives use disks running at constant angular velocity but use (fairly) constant data density across all tracks. This means that you will get much higher transfer rates on the outer tracks than on the inner tracks; a characteristics which fits the requirements for large libraries well.
Newer disks use a logical geometry mapping which differs from the actual physical mapping which is transparently mapped by the drive itself. This makes the estimation of the "middle" tracks a little harder.
In most cases track 0 is at the outermost track and this is the general assumption most people use. Still, it should be kept in mind that there are no guarantees this is so.
tracks are usually slow in transfer, and lying at one end of the seeking position it is also slow to seek to.
This is more suitable to the low end directories such as DOS, root and print spools.
tracks are on average faster with respect to transfers than inner tracks and being in the middle also on average faster to seek to.
This characteristics is ideal for the most demanding parts such as
swap, /tmp and /var/tmp.
tracks have on average even faster transfer characteristics but like the inner tracks are at the end of the seek so statistically it is equally slow to seek to as the inner tracks.
Large files such as libraries would benefit from a place here.
Hence seek time reduction can be achieved by positioning frequently
accessed tracks in the middle so that the average seek distance and
therefore the seek time is short. This can be done either by using
fdisk or cfdisk to make a partition
on the middle tracks or by first
making a file (using dd) equal to half the size of the entire disk
before creating the files that are frequently accessed, after which
the dummy file can be deleted. Both cases assume starting from an
empty disk.
The latter trick is suitable for news spools where the empty directory structure can be placed in the middle before putting in the data files. This also helps reducing fragmentation a little.
This little trick can be used both on ordinary drives as well as RAID systems. In the latter case the calculation for centring the tracks will be different, if possible. Consult the latest RAID manual.
The speed difference this makes depends on the drives, but a 50 percent improvement is a typical value.
The same mechanical head disk assembly (HDA) is often available with a number of interfaces (IDE, SCSI etc) and the mechanical parameters are therefore often comparable. The mechanics is today often the limiting factor but development is improving things steadily. There are two main parameters, usually quoted in milliseconds (ms):
After voice coils replaced stepper motors for the head movement the improvements seem to have levelled off and more energy is now spent (literally) at improving rotational speed. This has the secondary benefit of also improving transfer rates.
Some typical values:
Drive type
Access time (ms) | Fast Typical Old
---------------------------------------------
Track-to-track <1 2 8
Average seek 10 15 30
End-to-end 10 30 70
This shows that the very high end drives offer only marginally better access times then the average drives but that the old drives based on stepper motors are significantly worse.
Rotational speed (RPM) | 3600 | 4500 | 4800 | 5400 | 7200 | 10000
-------------------------------------------------------------------
Latency (ms) | 17 | 13 | 12.5 | 11.1 | 8.3 | 6.0
As latency is the average time taken to reach a given sector, the formula is quite simply
latency (ms) = 60000 / speed (RPM)
Clearly this too is an example of diminishing returns for the efforts put into development. However, what really takes off here is the power consumption, heat and noise.
There is also a
Linux Yoke Driver
available in beta which
is intended to do hot-swappable transparent binding of
one Linux block device to another. This means that if you
bind two block devices together,
say /dev/hda and /dev/loop0,
writing to one device will mean also writing to the other
and reading from either will yield the same result.
One of the advantages of a layered design of an operating system
is that you have the flexibility to put the pieces together
in a number of ways.
For instance you can cache a CD-ROM with cachefs that is
a volume striped over 2 drives. This in turn can be set up
translucently with a volume that is NFS mounted from another machine.
RAID can be stacked in several layers to offer very fast seek and
transfer in such a way that it will work if even 3 drives fail.
The choices are many, limited only by imagination and, probably
more importantly, money.
There is a near infinite number of combinations available but my recommendation is to start off with a simple setup without any fancy add-ons. Get a feel for what is needed, where the maximum performance is required, if it is access time or transfer speed that is the bottle neck, and so on. Then phase in each component in turn. As you can stack quite freely you should be able to retrofit most components in as time goes by with relatively few difficulties.
RAID is usually a good idea but make sure you have a thorough grasp of the technology and a solid back up system.
Many Linux users have several operating systems installed, often necessitated by hardware setup systems that run under other operating systems, typically DOS or some flavour of Windows. A small section on how best to deal with this is therefore included here.
Leaving aside the debate on weather or not DOS qualifies as an operating system one can in general say that it has little sophistication with respect to disk operations. The more important result of this is that there can be severe difficulties in running various versions of DOS on large drives, and you are therefore strongly recommended in reading the Large Drives mini-HOWTO. One effect is that you are often better off placing DOS on low track numbers.
Having been designed for small drives it has a rather unsophisticated
file system (fat) which when used on large drives will allocate
enormous block sizes. It is also prone to block fragmentation which
will after a while cause excessive seeks and slow effective transfers.
One solution to this is to use a defragmentation program regularly but
it is strongly recommended to back up data and verify the disk before
defragmenting. All versions of DOS have chkdsk that can do some
disk checking, newer versions also have scandisk which is somewhat
better. There are many defragmentation programs available, some versions
have one called defrag. Norton Utilities have a large suite of
disk tools and there are many others available too.
As always there are snags, and this particular snake in our drive paradise is called hidden files. Some vendors started to use these for copy protection schemes and would not take kindly to being moved to a different place on the drive, even if it remained in the same place in the directory structure. The result of this was that newer defragmentation programs will not touch any hidden file, which in turn reduces the effect of defragmentation.
Being a single tasking, single threading and single most other things operating system there is very little gains in using multiple drives unless you use a drive controller with built in RAID support of some kind.
There are a few utilities called join and subst which
can do some multiple drive configuration but there is very little
gains for a lot of work. Some of these commands have been removed in
newer versions.
In the end there is very little you can do, but not all hope is lost.
Many programs need fast, temporary storage, and the better behaved
ones will look for environment variables called TMPDIR or
TEMPDIR which you can set to point to another drive. This is
often best done in autoexec.bat.
SET TMPDIR=E:/TMP SET TEMPDIR=E:/TEMP
Not only will this possibly gain you some speed but also it can reduce fragmentation.
There have been reports about difficulties in removing multiple primary
partitions using the fdisk program that comes with DOS. Should this
happen you can instead use a Linux rescue disk with Linux fdisk to
repair the system.
Don't forget there are other alternatives to DOS, the most well known being DR-DOS from Caldera. This is a direct descendant from DR-DOS from Digital Research. It offers many features not found in the more common DOS, such as multi tasking and long filenames.
Another alternative which also is free is Free DOS which is a project under development. A number of free utilities are also available.
Most of the above points are valid for Windows too, with the exception of Windows95 which apparently has better disk handling, which will get better performance out of SCSI drives.
A useful thing is the introduction of long filenames, to read these from
Linux you will need the vfat file system for mounting these partitions.
Disk fragmentation is still a problem. Some of this can be avoided by doing a defragmentation immediately before and immediately after installing large programs or systems. I use this scheme at work and have found it to work quite well. Purging unused files and emptying the waste basket first can improve defragmentation further.
Windows also use swap drives, redirecting this to another drive can give you some performance gains. There are several mini-HOWTOs telling you how best to share swap space between various operating systems.
The trick of setting TEMPDIR can still be used but not all
programs will honour this setting. Some do, though. To get a good
overview of the settings in the control files you can run sysedit
which will open a number of files for editing, one of which is the
autoexec file where you can add the TEMPDIR settings.
Much of the temporary files are located in the /windows/temp
directory and changing this is more tricky. To achieve this you can
use regedit which is rather powerful and quite capable of
rendering your system in a state you will not enjoy, or more
precisely, in a state much less enjoyable than windows in general.
Registry database error is a message that means seriously bad news.
Also you will see that many programs have their own private temporary
directories scattered around the system.
Setting the swap file to a separate partition is a better idea and much less risky. Keep in mind that this partition cannot be used for anything else, even if there should appear to be space left there.
It is now possible to read ext2fs partitions from Windows,
either by mounting the partition using
FSDEXT2
or by using a file explorer like tool called
Explore2fs.
The only special note here is that you can get file system driver for
OS/2 that can read an ext2fs partition.
Matthieu Willm's ext2fs Installable File System for OS/2 can be found at
ftp-os2.nmsu.edu,
Sunsite,
ftp.leo.org and
ftp-os2.cdrom.com.
The IFS has read and write capabilities.
This is a more serious system featuring most buzzwords known to marketing. It is well worth noting that it features software striping and other more sophisticated setups. Check out the drive manager in the control panel. I do not have easy access to NT, more details on this can take a bit of time.
One important snag was recently reported by acahalan at cs.uml.edu : (reformatted from a Usenet News posting)
NT DiskManager has a serious bug that can corrupt your disk when you have several (more than one?) extended partitions. Microsoft provides an emergency fix program at their web site. See the knowledge base for more. (This affects Linux users, because Linux users have extra partitions)
You can now read ext2fs partitions from NT using
Explore2fs.
Most points regarding Windows NT also applies to its descendant Windows 2000 though at the time of writing this I do not know if the aforementioned bugs have been fixed or not.
While Windows 2000, like its predecessor, features RAID, at least one company, RAID Toolbox, has found the bundled RAID somewhat lacking and made their own commercial alternative.
There is a little bit of confusion in this area between Sun OS vs. Solaris.
Strictly speaking Solaris is just Sun OS 5.x packaged with Openwindows and
a few other things. If you run Solaris, just type uname -a to see your
version. Parts of the reason for this confusion is that Sun Microsystems
used to use an OS from the BSD family, albeight with a few bits and pieces
from elsewhere as well as things made by themselves. This was the situation
up to Sun OS 4.x.y when they did a "strategic roadmap decision" and decided
to switch over to the official Unix, System V, Release 4 (aka SVR5),
and Sun OS 5 was created.
This made a lot of people unhappy. Also this was bundled with other things
and marketed under the name Solaris, which currently stands at release
7 which just recently replaced version 2.6 as the latest and greatest.
In spite of the large jump in version number this is actually a minor
technical upgrade but a giant leap for marketing.
This is quite familiar to most Linux users.
The last release is 4.1.4 plus various patches.
Note however that the file system
structure is quite different and does not conform to FSSTND so any planning
must be based on the traditional structure. You can get some information by
the man page on this: man hier. This is, like most man pages, rather brief
but should give you a good start. If you are still confused by the structure
it will at least be at a higher level.
This comes with a snazzy installation system that runs under Openwindows, it will help you in partitioning and formatting the drives before installing the system from CD-ROM. It will also fail if your drive setup is too far out, and as it takes a complete installation run from a full CD-ROM in a 1x only drive this failure wil