Posting-Frequency: quarterly
Archive-name: chorus-faq Last-modified: 1998/02/09 Revision: 1.32 See reader questions & answers on this topic! - Help others by sharing your knowledge Change of administrator & location: Christian Bac (INT Evry, France) kindly agreed to continue the administration of the newsgroup. He will now do regular postings and maintaining of the HTML version of the FAQ. The Web version of the FAQ will still be available for a while on the Web server at Lancaster but should soon move to a new location at INT, Christian will keep you informed on that. E-mails to chorus-adm@comp.lancs.ac.uk will be redirected appropriately to the group administrators. I hope people will make more use of this newsgroup to discuss issues related to CHORUS technology and more general discussions! -Philippe ---------------- Table of Contents ================= 1. General Information 1.1. Organization and Availability of this FAQ 1.2. What's New? 1.3. What is CHORUS? 1.4. How to Contact Chorus Systems? 1.5. Disclaimer & Copyright 2. Documentation 2.1. Documentation Available through Anonymous FTP 2.2. Papers on CHORUS 2.3. Other references 2.4. Press references 3. Chorus Product Offering 3.1. Overview 3.2. Offering for Universities 4. CHORUS Microkernel 4.1. General 4.2. Supported Microprocessors 4.3. Porting on various platforms 4.4. Scheduling and real-time 4.5. CHORUS on Transputers 4.6. Comparison with other OS 4.7. Perfomances 4.8. Object Oriented issues 5. OS Personalities 5.1. OS personalities available on top of CHORUS 5.2. CHORUS/MiX V.3.2 5.3. CHORUS/COOL-ORB 5.4. CHORUS/ClassiX 6. Other Software -------------- 1. General Information =================== 1.1. Organization and Availability of this FAQ ----------------------------------------- This FAQ contains informations related to the CHORUS operating system, description of the products and contacts. It will be progressively updated according to the discussions held in this newsgroup and the evolution of the products. It is posted every quarter in the following newsgroups:<<comp.os.chorus, news.answers, comp.answers >>. A Japanese version of the FAQ is also available in the <<fj.os.misc>> newsgroup. Copies of the FAQ can also be obtained by e-mail by sending a request to "chorus-adm@comp.lancs.ac.uk". Hypertext version of the FAQ can be found at the following URLs: http://www.comp.lancs.ac.uk/computing/users/pr/www/chorus/faq.html http://www.faqs.org You can make any comments, suggestions or contributions to this FAQ by sending an e-mail to "chorus-adm@comp.lancs.ac.uk" or by discussions in the newsgroup. 1.2. What's New? ----------- Christian Bac (INT Evry, France) kindly agreed to continue the administration of the newsgroup. He will now do regular postings and maintaining of the HTML version of the FAQ. The Web version of the FAQ will still be available for a while on the Web server at Lancaster but should soon move to a new location at INT, Christian will keep you informed on that. E-mails to chorus-adm@comp.lancs.ac.uk will be redirected appropriately to the group administrators. I hope people will make more use of this newsgroup to discuss issues related to CHORUS technology and more general discussions! 1.3. What is CHORUS? -------------- CHORUS is a family of open microkernel-based operating system components to meet advanced distributed computing needs in areas such as telecommunications, internetworking, embedded systems, realtime, "mainframe UNIX", supercomputing and high availability. The CHORUS/MiX multiserver implementations of UNIX allow to dynamically integrate part or all of standard UNIX functionalities and services in the above application areas. CHORUS is designed, developed and marketed by Chorus Systems. 1.4. How to Contact Chorus Systems ----------------------------- North America: Chorus Systems Inc. 1999 South Bascom Avenue, Suite 400 Campbell, CA 95008 United States Phone: +1 (408) 879-4100 Fax: +1 (408) 879-4102 Voice Mail: +1 (408) 291 8832 Email: info@chorus.com Europe: Chorus systemes SA 6 avenue Gustave Eiffel F-78182 St Quentin-en-Yvelines Cedex France Phone: +33 (1) 30 64 82 00 Fax: +33 (1) 30 57 00 66 Email: info@chorus.fr Asia Pacific: Chorus Systems KK Mitsutake Building Ikejiri, 8F 3-22-4 Ikejiri, Setagaya Ku Tokyo 154 Japan Phone: +81 (3) 5430-1131 Fax: +81 (3) 5430-1133 Email: info-ap@chorus.com 1.4. Disclaimer & Copyright --------------------- The author provides no warranty regarding the content of this document. It is provided "as is" without express or implied waranty. Permisson to distribute this document, in part or full, via electronic means (emailed, posted or archived) or printed copy are granted providing that no charges are involved, reasonable attempt is made to use the most current version, and all credits notices are retained. For other distributions including incorporation in commercial products, such as books, magazine articles, or CD-ROMs, you must get permission from the author first (e-mail: chorus-adm@comp.lancs.ac.uk). 2. Documentation ============= 2.1. Documentation Available through Anonymous FTP --------------------------------------------- There are several technical reports on CHORUS available via anonymous FTP from Chorus systemes, France: ftp.chorus.fr [192.33.15.3], directory pub/chorus-reports (see the file "index" for an overview). A set of slides on CHORUS is also available in the directory "pub/chorus-slides", documents CS-TR-92-64 (PostScript, versions 1-up and 2-up). Product Data Sheets are available (in ascii or PostScript format) in the directory "pub/chorus-datasheets". Those reports are available through the World Wide Web at Chorus systemes: "ftp://ftp.chorus.fr/" or from the Web server at the URL:"http://www.chorus.com". 2.2. Papers on CHORUS ---------------- [Bricker, 1991] A. Bricker, M. Gien, M. Guillemont, J. Lipkis, D. Orr and M. Rozier, "A new look at micro-kernel-based UNIX operating systems: Lessons in performance and compatibility". Proc. of the EurOpen Spring'91 Conference, Tromsoe, Norway, 20-24 May 1991. Chorus Systems Technical Report CS-TR-91-7 [Coulson, 1994] Coulson G., and G.S. Blair. "Microkernel Support for Continuous Media in Distributed Systems". Computer Networks and ISDN Systems, Special Issue on Multimedia, 1994; also available as internal report MPG-93-04, Computing Dept., Lancaster University. [Gaultier, 1994] O. Gautier and Y. Metais. "Mise en Place d'une Plateforme CHORUS, Conception et Implementation d'un Ordonnanceur a Echeance au Sein du Noyau Chorus". Memoire CNAM, Paris, March 1994. CS/TR-94-82.1 "CHORUS Kernel v3 r5 for T425/T805 Connection Manager and Aserver Library" (Internal Report) CS/TR-94-81.1 "CHORUS Kernel v3 r5 for T425/T805 Host Server User's Manual, Aserver Guide" (Internal Report) 2.3. Other references ---------------- [Bradley, 1993] J Bradley Chen and Brian N Bershad "The Impact of Operating System Structure on Memory System Performance", ACM SIGOPS Dec. '93 [Coulouris, 1994] G. Coulouris, J. Dollimore and T. Kindberg. "Distributed Systems, Concepts and Design", Addison-Wesley, second edition, 1994. [Douglis et al., 1992] Douglis, F., Kaashoek, M.F., and Tanenbaum, A.S.: "A Comparison of Two Distributed Systems: Amoeba and Sprite," Computing Systems, vol. 4, Fall 1991 (sic). [Dean, 1992] R. Dean and F. Armand. "Data Movement in Kernelized Systems". Proceedings of the USENIX Workshop on Micro-Kernels and Other Kernel Architectures, pp. 243-261, April 1992. [Tanenbaum, 1992] Andrew S. Tanenbaum: "Modern Operating Systems", Prentice-Hall, 1992. [Tanenbaum, 1994] Andrew S. Tanenbaum, "Distributed Operating Systems", Prentice-Hall, ISBN 0-13-219908-4 [Tanenbaum, 1995] Andrew S. Tananbaum, "A Comparison of Three Microkernels", The Journal of Supercomputing, vol 9, number 1, ISSN 0920-8542, 1995. 2.4. Press references ---------------- BYTE, Jan 94 issue: - "Small Kernels Hit it Big" (by Peter Varhol, p. 119, 6 pages), and - "The Chorus Microkernel" (by Dick Pountain, p. 131, 4 pages) (a color reprint of these articles is available upon request from Chorus Systems.) BYTE, Feb 95 issue: - "Novell's Campaign" (by Jon Udell, p. 43, 11 pages) CHORUS is mentioned in this article as being the basis for the microkernel underpinning the UnixWare and NetWare personalities. BYTE, Mar 95 issue: - "Europe's Chip Challenge" (by Dick Pountain, p. 19, 5 pages) CHORUS is mentioned in this article on the CEC's Open Microprocessor Initiative (OMI) as the microkernel being used and enhanced within various OMI projects. 3. Chorus Product Offering ======================= 3.1. Overview -------- CHORUS/Micro is a very small (10K) hard real-time embedded kernel typically used in low-cost, dedicated application environments needing minimal functionality and a minimum memory footprint, such as line cards, portable phones, and hand-held devices. CHORUS/ClassiX is host-target cross-development environment for C++ or C written applications, named C_actors. C_actors can be loaded, unloaded and debugged dynamically from the host (e.g. SPARCstation/SunOS) on the target (e.g. ix86 PC/AT), interconnected via Ethernet. C_actor applications can interoperate with UNIX on the host through TCP/IP sockets and NFS. On the target, there is only the CHORUS/Nucleus microkernel and the CHORUS/C_actor subsystem but no CHORUS/MiX UNIX System V subsystem. CHORUS/ClassiX is available for i386/i486/Pentium (PC/AT), mc68040 (MVME167), mc68360 (QUADS) Micro-SPARC-1/-2 (FORCE CPU-3CE/-5CE, SPARCclassic, SPARCstation 5, SPARCengine 5) and T425/T805/T9000 (SGS-Thomson INMOS boards). More informations can be found at the following URL: http://www.chorus.com/Products/Datasheets/classix.html. CHORUS/MiX V.4 is a distributed multi-server implementation of UNIX SVR4.0 on top of the CHORUS microkernel). E.g. on a 386/486 PC/AT, it offers binary compatibility with native SVR4.0v4. MiX V.4 requires a Novell/USG SVR4.0 source license. More information on this product can be found at http://www.chorus.com/Products/Datasheets/mixv4.html. CHORUS/COOL is a distributed programming environment for object-oriented applications. CHORUS/COOL supports the dynamic creation of C++ objects, these objects can be invoked, using C++ mechanism in a system wide transparent way. Objects can migrate, and remain persistent unless explicitly deleted. The programming model used is based on the Object Management Group's architecture (OMG). CHORUS/COOL-ORB is an OMG-CORBA compliant Object Request Broker. It is available for CHORUS/Fusion, CHORUS/ClassiX, CHORUS/MiX V.4, SCO UNIX, SunOS 4.1, Solaris, Linux, AIX, Windows95 and WindowsNT. More information concerning this product can be found at the following URL: http://www.chorus.com/Products/Datasheets/coolorb.html CHORUS/JaZZ is an implementation of the Java runtime and selected components of the JavaOS that have been integrated into CHORUS/ClassiX as an operating system personality. CHORUS/JaZZ extends CHORUS/ClassiX with a Java "Virtual Machine" personality including all standard Java containers and classes such as windows, threads, I/O, exceptions and network. More information on this product can be found at the following URL: http://www.chorus.com/Products/Datasheets/jazz.html. CHORUS/Harmony includes C , C++, and Embedded C++ optimizing compilers, tool chains (assemblers, linkers, utility programs) profilers, runtime error checkers, simulators and kernel debuggers. All of the Integrated Development Environment components share a common GUI and communicate with each other for enhanced functionality. 3.2. Offering for Universities ------------------------- Chorus Systems has special programs for universities. More information on offerings, conditions, etc is available via ftp (ftp://ftp.chorus.fr) in the following ASCII files: - pub/README - pub/academic/README - pub/academic/offerings If you have questions, you may contact Didier.Irlande@chorus.fr. 4. CHORUS Microkernel ================== 4.1. General ------- * What is a microkernel? A "microkernel" is an operating system with only the essential services, such as interprocess communication, short-term scheduling, and memory management. It basically provides the process abstraction and a means for processes to communicate. It is designed to be portable between computer architectures, using high-level languages such as C or C++ and reducing the machine-dependant component to a minimal bottom layer. The microkernel appears as a layer between the hardware layer and a layer consisting of system components called 'subsystems'. Their size can vary from about 10Kb to several hundred kilobytes of executable code and static data. * Synchronisation primitives offered to CHORUS threads? The CHORUS microkernel (v3 r5.x) offers the following synchronisation primitives: - mutexes - (counting) sempahores - spin locks (supervisor applications only) - mini messages (supervisor applications only) Other synchronisation primitives such as condition variables and reader/writer locks can be built on top of those basic primitives. * Do CHORUS threads support specific data? Yes. The microkernel supports so-called "software registers". Each thread has two software registers which are systematically saved/restored by the microkernel upon a thread context switch. The software registers can be read/written through threadStoreR(K) and threadLoad(K) system calls. A software register typically contains a pointer to a per-thread private data area. Via software registers, one can implement e.g. a per-thread value of "errno". * Distributed synchronization service on top of the CHORUS microkernel? This work is part of a PhD thesis undertaken by Stephane Eranian <eranian@chorus.fr> implementing distributed synchronization service on top of the CHORUS microkernel. It implements pure mutex (no mr/sw). The synchronization is achieved using a token-based algorithm through a server. Its main role is to manage token creation, deletion and sharing among sets of clients. For more information contact S. Eranian. * Initialization of the context of an actor created with the actorCreate(K) system call? To create an actor in CHORUS using actorCreate(K), the actor address space must be empty (i.e. does not include any valid memory region). If you want to "fork" an actor, i.e. the new actor's code and data is a copy of the main actor's code and data. In order perform address space duplication, you can use the rgnDup(K) service, just after having created the actor (and before having created new regions: this operation is only valid on an empty address space). Before invoking rgnDup(K), you must specify for each region how the region will be "duplicated": map (sharing), copy or not duplicated at all. In order to do this, you have to get the list of your regions (rgnStat(K)), and set the inheritance flags using rgnSetInherit(K). Note: an alternative to the rgnSetInherit/rgnDup scheme is to treat each region individually, using rgnMapFromActor(K) for shared regions, and rgnInitFromActor(K) for copied regions. * Encryption mechanisms in CHORUS IPC? CHORUS IPC does not provide any "encryption" mechanisms. Such mechanisms can be implemented by applications. * Given a set of actors and ports, what would happen if one of the actors is stopped (with an actorStop) prior to an ipcSend(). Would all the other members of the port group receive the message or would the behavior of the kernel be different ? Depends on the mode set for the group, but if the group is in broadcast mode all the members of the group should receive the message, including ports belonging to actors in stopped state. * Trap connection in supervisor actors using svTrapConnect(K) or scCallConnect(K)? - svCallConnect(int trapNb, KnCallEntry *table, uint entNb, int flag); Connects a trap handler table to a specific trap number where each entry in the table contains a pointer to a function and the number of arguments it takes. entNb is the number of entries in the table. One thing to note is that svCallConnect()'s trap number is biased by the constant TRAP_BASE (cf. cpu.h). That is, if you want to use interrupt 5, you would pass 5 to svCallConnect() and trigger the trap by executing the assembly instruction: int $(5 + TRAP_BASE) after first loading %eax with the service you want. On ix86 the assembly file would be something like: .globl DoSomething DoSomething: movl $0, %eax ; index of the call in the table in eax int $(TRAP_BASE + trapNb) ret If you have access to CHORUS sources you can look at how the CHORUS libraries are built (i.e. lib/mklib or lib/mkslib). - svTrapConnect(uint trapNb, KnHdl func) connects a trap handler taking two arguments (thread context and interrupt number) to a the hardware trap trapNb (trapNb is an absolute value in this case). The assembly code to generate the trap would be similar to the code mentioned for svCallConnect for ix86 architectures. On other architectures you can either look at how kernel libraries are built if you have access to the source code, otherwise look at the processor reference manual or try disassemble a kernel library. 4.2. Supported Microprocessors ------------------------- Various versions of the CHORUS microkernel have been ported to a variety of microprocessors, either by Chorus Systems or by its clients: - i386/i486/Pentium (various PC/ATs) - mc68030/mc68360/mc68040 (MVME147S, QUADS, MVME167S) - mc88k - SPARC (SPARCstation SLC, SPARCstation Classic, SPARC CPU-3CE) - transputer T425/T805/T9000 - R3000/R4000 (Sony 3410) - PA-RISC (HP 9000/834 and 9000/720) - YMP (Cray YMP) 4.3. Porting on various platforms ---------------------------- * Chorus on Macintosh? The INT (Institut National des Telecommunications, Evry, France) has ported the v3 r3 version of the CHORUS microkernel to a Macintosh II CX (mc68030-based). CHORUS and MacOS coexist and cooperate on the same hardware. The paper "Cohabitation and Cooperation of Chorus and MacOS", by Christian Bac and Edmond Garnier, was presented at the Usenix Symposium on Microkernels and Other Kernel Architectures in Sep 93 in San Diego. You can find the paper in the proceedings. It is available from ftp.int-evry.fr:/pub/systeme. In the same directory you will also find another paper on the same subject: "ChorusToolbox : MacOS running on top of Chorus", by Christian Bac and Hong Quang Nguyen from INT. This paper was presented at SUUG'94 in April 94 in Moscow. * CHORUS on transputers? Archipel, Chorus and SGS/Thomson Inmos have ported the CHORUS microkernel and the CHORUS/MiX V.3.2 subsystem (SVR3.2 compatible) to T425 and T805 transputers. This was done in the context of the Esprit project "Harmony" (EC-funded R&D). Initially, a T9000 port was planned to be available by now. Due to a delay in the availability of the T9000, the CHORUS port (which is underway now), has shifted as well. Inmos and Chorus have been working together in order to assure that CHORUS/MiX (i.e. UNIX) will run in an optimal manner on the T9000. * CHORUS on 64-bit architecture? CHORUS has been ported to DEC's Alpha, Cray Research's YMP and MIPS' R4000. * Port of CHORUS on HP-PA? On December 1st Jon Inouye <jinouye@cse.ogi.edu> wrote: Prof. Jonathan Walpole supervised a port of the CHORUS v3.3 nucleus to the Hewlett-Packard 9000/834 workstation from late 1990 to mid-1991. This was part of a funded research project to evaluate the CHORUS operating system with respect to the Hewlett-Packard PA-RISC architecture. The nucleus did not support any disk/network drivers and performed all console/ keyboard I/O though IODC (PROM) routines. A CHORUS/MiX V.3.2 Process Manager (PM) port was partially completed to the point where UNIX shells and certain system calls were supported ... but not a UNIX file system. Since then, I have been porting Chorus/MiX V.3.2 (with the v3.4 nucleus) to the HP 9000/720. Since I am performing this port in my spare time it is not progressing very fast. The v3.4 nucleus runs along with a serial driver. It lacks other device drivers, FP emulation support (though basic FP operations are supported) and still uses the old HP-UX PDIR structure rather than the more recent HPT. The Ethernet driver is still being debugged as is an ancient version of the MiX V.3.2 PM. The port is being used for virtual memory experiments. Both ports use a considerable amount (over 40,000 lines combined) of HP-UX source code for the assembly language utilities, boot up, I/O initialization, and device drivers. The 834 port uses a Tut (HP-UX 2.0 modified to run Mach 2.0) base and the 720 port uses a HP-UX 8.0 base. For this reason, we have not been able to release anything because of all the legal implications ... HP, Chorus, USL copyrights. The evaluation is available as a series of OGI technical reports which can be obtained via anonymous ftp from cse.ogi.edu (129.95.20.2) in the directory /pub/tech-reports or via the URL: "http://www.cse.ogi.edu/DISC/projects/pa-chorus/pa-chorus.html". * Are PCI-bus devices supported on ix86? PCI devices, like video and IDE, whose I/O mode is compliant with ISA, are supported; they are just seen as ISA adapters. PCI devices which are not ISA compliant (e.g. the SCSI controller and/or the Ethernet controller on some COMPAQs) are not supported; supporting them would require modifications in the driver code (and possibly also in the CHORUS microkernel code). 4.4. Scheduling and real-time ------------------------ * Scheduling mechanisms and scheduling policies The CHORUS microkernel makes a distinction between scheduling mechanism and scheduling policies. The core scheduler within the microkernel does pure preemptive scheduling (SCHED_FIFO in POSIX RT terms). On top of that, different scheduling policies can be implemented in the form of scheduling classes; each class communicates with the core scheduler and can make its own scheduling decisions within that class based upon attributes (priorities, deadlines, etc) and behaviour (time-slicing, SCHED_RR, ...). Today, 4 scheduling classes are provided: a default class and the 3 UNIX SVR4 classes (SVR4_TS, SVR4_RT and SVR4_SYS). Work is in progress for additional classes (deadline, fair-share, etc) cf the work done by Olivier Gaultier and Olivier Metais at CNAM Paris on the implementation of an EDF (Earliest Deadline First) policy in the CHORUS kernel. * Relative cost of context switch between user and supervisor threads? User threads/actors have their own address spaces, and are protected from other user address actors. Supervisor threads/actors all share the supervisor address space, each supervisor actor has its own "slot" in the supervisor address space. For comparison, let's take: [U] a context switch between 2 user threads in different user actors, [S] a context switch between 2 supervisor threads in different supervisor actors Unlike [U], [S] does not require the saving/restoring of the memory context, so [S] is less costly. For the CHORUS/Nucleus v3 r5.2 on a i486/50MHz, the ratio [U]/[S] is 1.57. * How to measure Interrupt latency? The easiest way is to connect an interrupt handler to a hardware timer. As soon as the handler is activated, the handler will measure the current time and then wake-up a thread, e.g. by doing a V on a semaphore. The thread, when returning form its P operation on the same semaphore, will also measure the current time. The difference between these 2 time measurements is the latency. If you are using CHORUS/ClassiX or CHORUS/MiX V.4, you should take a look at one of the example applications in the tutorial, named ILD (Interrupt Latency Demo) which does exactly this work. 4.5. CHORUS on Transputers --------------------- * On transputers, how to communicate from CHORUS to UNIX (SunOS)? There is no standard way to do this with CHORUS. This is the kind of things CHORUS/MiX is there for. Specific to transputer, you can use emulated links. On your workstation run a daemon, the "Aserver", which comunicate with your b300 box. Together they emulate transputer virtual links over TCP/IP. These links can then be used by transputer and Unix applications to communicate. For details on how to do that, see the sections on the Aserver in the CHORUS documentation (CS/TR-94-82.1 and CS/TR-94-81.1). * Use of the INMOS RTL for the C-toolchain to work with sockets? [Answer on Dec. 6, 1994 from B. Wipfel <raw@unislc.slc.unisys.com>]: Guessing that you want to make socket calls from one of your actors, and have the B300 handle them in the normal way, the trouble is that the transputer implementation of CHORUS uses AServer, not IServer. Since AServer communication is encapsulated in IServer MEGA_PACKETS, the B300 never gets to see any of the socket packets and passes everything to the host. A real option is to write a new AServer server to provide your socket service. In this case, all communication will go to the host, and it will make the socket calls on behalf of your actor in the target system. The B300 wouldn't be involved, other than maintaining normal AServer communication with the host. This is kind of a shame, since the B300 has the necessary functionality. A last option might be to use a second root link. Wire up two links from the B300 to your transputer network. Boot the network via one of the links. Configure the second link as a network "EDGE". Have your actor connect to the edge link with one of the AServer routines; something like cmLinkOpen("/dev/raw/00") ? Once the link is open, and you have the channel pointers, it might be possible to attach the socket library channels to these channels. You'll need to do communication via CHORUS' channel I/O routines however. * If we have made ourselves a sockgateway process, configured in the .cfs file as: .... interface(input fromhost, output tohost, input formChorus, output toChorus); Where do we connect the input and outputs ? [Answer on Nov. 25, 1994 from N. Stephen <stephen@osf.org>]: I don't know what release you might have, but look up the #device primitive in the build tool users manual. This primitive allows you to attach native transputer processes' channels (which normally control devices) to the CHORUS world, and does all the necessary wiring so that these channels are accessible from the connection manager. There may also be an example of this being done in one of the tutorials, if you have them with your release - the CHORUS Nucleus Tutorial (2), mixing CHORUS and Native transputer code. 4.6. Comparison with other OS ------------------------ * differences between Mach and CHORUS? There are a lot of similarities between the concepts of the CHORUS v3 r5 microkernel and the Mach 3.0 microkernel: IPC, threads, memory management. See [Dean, 92] (cf. section 2.3) for more elements of comparison between the two kernels. (For comparison with other kernels see [Tanenbaum, 1994], [Tanenbaum, 1995] in section 2.3) 4.7. Performances ------------ * Is there any system for performance analysis? As far as the CHORUS microkernel applications ("actors") are concerned, the CHORUS/Profiler and the CHORUS benchmarks can be used. The CHORUS/Profiler allows one to obtain and display symbolic call-graph profile data for actors (similar to UNIX' gprof(1)): callers, calllees, absolute and relative time spent in different procedures within one actor. Actors are to be compiled with the -p option. The profiler consists of a supervisor actor (PROF) plus 2 CHORUS/MiX utilities. Profiling be can enabled/disabled dynamically using the CHORUS/MiX utility profctl(1). profctl(1) stores the raw profiling data in a UNIX file, which can then be exploited by the report generator profrpg(1) in order to produce a human-readable profile report. See ftp://ftp.chorus.fr/pub/ datasheets/Profiler_v3_r4.{ascii,ps.Z}. CHORUS benchmarks allow you to get performance figures for individual microkernel system calls. For some system calls, like ipcSend(K), you get performance figures for different cases/parameters (small/medium/big message sizes). These basic figures can also help you to analyse and tune the performance of your microkernel applications. In the context of the ESPRIT project Ouverture, Alcatel, Siemens and Chorus have designed and implemented so-called hooks for monitoring and debugging in the CHORUS microkernel. These hooks are a clean set of new microkernel system calls which allow monitoring and debugging tools (e.g., PATOC, PARTAMOS) to be informed about the occurrence of events they're interested in (context switch, message arrival, thread creation, etc). They will be available in a future CHORUS product release. PATOC is a graphical tool (Motif-based) that allows to monitor applications running on CHORUS. It is event based and is able to display its information is various forms (diagrams, bar-charts etc). PATOC is not a product but rather a working prototype. 4.8. Object Oriented Issues ---------------------- * How is CHORUS Object-Oriented? The major part (>90%) of the CHORUS microkernel is written in C++. OO techniques are used in the implementation of the microkernel, but the API exported by the microkernel is a traditional procedure call based interface (like UNIX). 5. OS Personalities ================ 5.1. OS personalities available on top of CHORUS? ------------------------------------------- Chorus Systems has developed the following personalties: - SVR4.0 - SVR3.2 - SCO ODT 3.0 - BSD4.3 - object-oriented (CHORUS/COOL) - POSIX real-time (POSIX 1003.1b/.1c, former .4/.4a) Others have developed, or are developing, personalities for SVR4.2 MP, UNICOS, MacOS, CHILL, ESTEREL, TINA DPE, and a number of (proprietary) real-time OSs. 5.2. CHORUS/MiX V.3.2 ---------------- * What is the link between a u_thread and a kernel thread? A u_thread is an abstraction, created by the CHORUS/MiX V.3.2 subsystem, on top of the microkernel threads (like a UNIX process is created on top of a CHORUS actor). Each u_thread is mapped 1-1 to a microkernel thread. A u_thread has some UNIX-specific attributes, like signal context, which is managed at the CHORUS/MiX subsystem level, as an added value w.r.t. a microkernel thread. * Which synchronisation primitives are offered to u_threads? Sempahores and mutexes. * Do u_threads support Thread Specific Data? Yes, through the threadsafe C library (c_threadPrivate(3CT), c_getPrivate(3CT)). These functions are in fact built on top of the software registers described for the kernel threads (cf 4.1). * How are u_threads scheduled? By the microkernel, just like any other thread. If one want to implement N user level threads on top of 1 kernel thread, he need a user level scheduler in some kind of run-time library (just like an Ada run-time schedules multiple Ada tasks within one Ada program). * Is there a way to calculate the size of a u_thread stack? For dynamic calculation, the best approach is probably to fill the stack you allocate with a specific pattern, and then at run-time (or at thread termination) control which part of the stack still contains the pattern. This allows to calculate which part of the initial stack has (not) been used. To detect a stack overflow, the classical approach is to surround the stack by some chunks of memory that are mapped read-only. * Is it better to put the thread stack in the data segment or in the heap area? The only advantage of putting it on the heap is that the corresponding memory can be allocated (and freed) dynamically, according to the application's run-time behaviour and needs. If you allocate it as data, the corresponding memory is always allocated, even if your thread doesn't exist yet/anymore. * What are the differences between an u_thread and a c_thread? The c_threadXxx(3CT) interface is a library, built upon the u_threadXxx(2C) interface, and is inspired by early drafts of the POSIX pthreads interface (POSIX 1003.4a, later renamed to .1c). It offers a higher level interface than u_threadXxx(2C), e.g.: - when creating a c_thread the library creates the stack for you, while for a u_thread you have to allocate the stack yourself, - there is a routine allowing one c_thread to wait for the termination of another c_thread (c_threadJoin), - there are routines to allocate and access per c_thread private global data. 5.3. CHORUS/COOL-ORB --------------- * How can I write an CORBA::Any to a file and reading it back from file without any knowledge about the type? Reply from <ciceron@chorus.fr> on July 25th: The idea is to fill an area of memory with the any value. The memory must be filled in the same way COOL builds its communication messages. Once the memory image is built, you simply have to write it. On the same idea you can read back the any. With this scheme, you don't need to cope with the type of the any (and you may also not have it). I've written a small persistent generic distributed database on top of CHORUS/COOL ORB. It uses this technique which works very well. (This is not CORBA but... it's cool) Here is how you could write: CORBA_Any any = ... long size = ::marshalSize(any); COOL_ComBufDesc buffer(size); buffer <<= any; write(fd, buffer.buffer(), size); Here is how you could read: struct stat st; fstat(fd, &st); COOL_ComBufDesc buffer(st.st_size); read(fd, buffer.buffer(), st.st_size); CORBA_Any any; buffer >>= any; * Array/sequences manipulation with CHORUS/COOL. From <ciceron@chorus.chorus.fr> Mon Sep 2, 1996: Here is the solution for COOL ORB. With COOL you can take the address of the first element of the array and use it as the beginning of the buffer holding the sequence. For example: // IDL: typedef sequence<octet> OctetSeq; // C++ OctetSeq* seq = ...; // Passed to the server implementation // CORBA [] operator returns a reference to the first element CORBA_Octet& o = (*seq)[0]; // Convert to a pointer and you get the begining of the buffer CORBA_Octet* p = &o; You can then optimize the extraction of your data... * list of platforms supported for CHORUS/COOL? For an up to date list you can look at the chorus web site at <http://www.chorus.com/> in the COOL section. The list of platforms supported by CHORUS/COOL r3.1 is the following: +----------------------------------------------------------------+ | System | Compiler | Type | |----------------------------------------------------------------| | AIX 2.3 | gcc 2.7.2 g++ | mono-threaded | |----------------------------------------------------------------| | ClassiX r2.3 for i386at | Dev System r1.1| multi-threaded| |----------------------------------------------------------------| | ClassiX r2.3 for MVME167 | Dev System r3.1| multi-threaded| |----------------------------------------------------------------| | ClassiX r3.0 for i386at | Dev System r4.0| multi-threaded| |----------------------------------------------------------------| | Fusion r2 | SCO C++ 3.1.1 | multi-threaded| |----------------------------------------------------------------| | Linux 1.2 | gcc 2.7.2 g++ | mono-threaded | |----------------------------------------------------------------| | SCO OpenDesktop 3.0 | SCO C++ 3.1.1 | mono-threaded | | ---------------------------------| | | gcc 2.7.2 g++ | mono-threaded | |----------------------------------------------------------------| | SCO OpenServer 5.0 | SCO C++ 3.1.1 | mono-threaded | | ---------------------------------| | | gcc 2.7.2 g++ | mono-threaded | |----------------------------------------------------------------| | Solaris 2.5 | Sparc C++ V4.1 | mono-threaded | | ---------------------------------| | | Sparc C++ V4.1 | multi-threaded| | ---------------------------------| | | gcc 2.7.2 g++ | mono-threaded | |----------------------------------------------------------------| | SunOS 4.1.3 |Sparc C++ V2.0.1| mono-threaded | | ---------------------------------| | | gcc 2.7.2 g++ | mono-threaded | | ---------------------------------| | |Sparc C++ V4.0.1| mono-threaded | |----------------------------------------------------------------| |Windows NT, Windows 95 | Visual C++ 4.0 | mono-threaded | | ---------------------------------| | | Visual C++ 4.0 | multi-threaded| +----------------------------------------------------------------+ * Implementation of COOL_Mutex and COOL_Sem? Reply from <ciceron@chorus.fr> on April 7, 1997: COOL_Mutex/COOL_Sem are based on the underlying operating system primitives. That is: mutexGet/mutexRel/semP/semV on CHORUS/OS, mutex_lock/mutex_unlock/sema_wait/sema_post on Solaris 2.5, and others on Windows etc... * How to compile the demo examples of CHORUS/COOL-ORB r4.1 for debugging purposes? You must use: make CXXDEBUGFLAGS="-g -O" * We are using Orbix 2.2 on NT 4.0, and want to create many CORBA objects per process (at least several thousands to tens of thousands). From our tests it seems that the time it takes to allocate such objects is in linear relation to the number of objects. Has anyone tackled this problem successfully ? Do other ORBs have a more efficient solution to this problem ? <From ciceron@dimba.chorus.fr Fri Aug 29, 1997>: In COOL ORB, the registration depends on the OA that you are using. The standard OA proposed by COOL uses a hash table for this. This scales quite well. You can imagine to provide a new OA that registers in a binary tree (balanced tree would be even better). To support billions of objects, you should have a look at the Variable Sized Object Reference feature that COOL ORB supports. A single servant can be used by many objects: the object reference always refers to the same servant BUT it also contains additional data which is specific to the object. Have a look at the 'simpleTree' demo which illustrates a binary tree and only one servant for referencing all the nodes. Your tree can contain billions of nodes, you will always have a single servant. In the next release, COOL ORB will propose a new OA that supports compounds of servants. This will allow you to register many servants (any type) in a single registration call to the OA. Note that beside the performance aspect that you are facing, there is a memory problem too. Each time you register a servant, the OA needs to allocate some memory to record it. If you want to support billions of servants, you must consider this. Having a single servant for implementing many objects is a first way to reduce the memory usage. Using the compounds OA is another way. 5.4. CHORUS/ClassiX --------------- * I am trying to write an experimental network protocol on top of CHORUS/Classix and found that I get Segmentation fault... <From maze@crf.canon.fr Tue Sep 30, 1997>: Wrong Imakefile: the ndm library is missing in the libs part of the ClassiXSupProgramTarget macro. As the actor is a relocatable supervisor actor, the ndmGetMajMin routine is left undefined in the binary, leading to the segmentation fault. In you Imakefile you should have: SRCS = hello.c LIBS = $(MERGEDIR)/lib/ndmlib/ndmlib.s.a ClassiXSupLibs ClassiXSupProgramTarget(hello, hello.o, , $(LIBS), ClassiXReloc) DependTarget() 6. Other software ================ * We are looking for a ISAM file library or a lightweight database (in terms of storage, runtime overhead and price!) running under CHORUS (or under BSD UNIX 4.2 if the source code is available). So far, I found two candidate libraries. The first one is the famous C-ISAM; However Informix doesn't sell this product anymore (at least not in Germany). The second one is called CQL++ of Machine Independent Software Corporation, Scottsdale, AZ. Has anyone heard of it or even used it in a software project, respectively? <From scrantr@ix.netcom.com, Wed Aug 27, 1997>: Mix Software (Richardson, Texas) also sells for *cheap* an ISAM database library using b+tree indexing. Source is available, and the package includes a nice manual describing the library. See http://www.mixsoftware.com/product/database.htm User Contributions:
[ Usenet FAQs | Web FAQs | Documents | RFC Index ]
Send corrections/additions to the FAQ Maintainer: pr@comp.lancs.ac.uk
Last Update March 27 2014 @ 02:11 PM
|
Comment about this article, ask questions, or add new information about this topic: