Patent application title: META-EVENT GENERATION BASED ON TIME ATTRIBUTES
Hugh S. Njemanze (Los Altos, CA, US)
Pravin S. Kothari (San Jose, CA, US)
Debabrata Dash (Sunnyvale, CA, US)
Shijie Wang (San Jose, CA, US)
IPC8 Class: AG06F2100FI
Class name: Information security policy
Publication date: 2012-10-11
Patent application number: 20120260306
First stage meta-events are generated based on analyzing time attributes
of base events received from a network component. Second stage
meta-events are generated based on a number of the first stage
meta-events that have a time attribute falling within a time period. An
amount of time that has passed since a most-recent second stage
meta-event was generated is determined, and if a threshold time period
does not exceed the amount of time that has passed since the most-recent
second stage meta-event was detected, a third stage meta-event is
42. A computer-implemented method, comprising: generating first stage meta-events based on analyzing time attributes of base events received from at least one network component; generating second stage meta-events based on a number of the first stage meta-events that have a time attribute falling within a time period; determining an amount of time that has passed since a most-recent second stage meta-event was generated; and if a threshold time period does not exceed the amount of time that has passed since the most-recent second stage meta-event was detected, generating a third stage meta-event.
43. The method of claim 42, wherein generating first stage meta-events comprises: identifying a first rule that indicates a threshold number of base events and a first time period; determining how many base events include a time attribute that falls within the first time period; and when the threshold number of base events does not exceed the number of base events whose time attributes fall within the first time period, generating a first stage meta-event.
44. The method of claim 43, comprising: activating the first rule dynamically.
45. The method of claim 43, comprising: detecting improper rule syntax.
46. The method of claim 43, comprising: detecting a loop condition generated by the first rule.
47. The method of claim 43, comprising: detecting rule feedback.
48. The method of claim 43, comprising: filtering the base events based on a condition before determining how many base events include a time attribute that falls within the first time period.
49. The method of claim 48, wherein filtering the base events based on the condition comprises discarding the base events that do not satisfy the condition.
50. The method of claim 42, wherein the at least one network component comprises an intrusion detection system.
51. The method of claim 42, comprising: aligning time of the base events generated by different network components.
52. The method of claim 42, comprising: performing an action specified by the first rule to notify an individual of the first-stage meta-events.
53. A system, comprising: data storage to store base events received from at least one network component; and a processor to generate first stage meta-events based on analyzing time attributes of base events received from at least one network component, generating second stage meta-events based on a number of the first stage meta-events that have a time attribute falling within a time period, determine an amount of time that has passed since a most-recent second stage meta-event was generated, and if a threshold time period does not exceed the amount of time that has passed since the most-recent second stage meta-event was detected, generate a third stage meta-event.
54. The system of claim 53, wherein the processor to generate first stage meta-events comprises the processor to identify a first rule that indicates a threshold number of base events and a first time period, determine how many base events include a time attribute that falls within the first time period, and when the threshold number of base events does not exceed the number of base events whose time attributes fall within the first time period, generate a first stage meta-event.
55. The system of claim 54, wherein the processor is to activate the first rule dynamically.
56. The system of claim 54, wherein the processor is to detect improper rule syntax.
57. The system of claim 54, wherein the processor is to detect a loop condition generated by the first rule.
58. The system of claim 54, wherein the processor is to filter the base events based on a condition before determining how many base events include a time attribute that falls within the first time period.
59. The system of claim 58, wherein the processor is to discard the base events that do not satisfy the condition.
60. A non-transitory computer readable medium storing machine readable instructions that when executed perform instructions to: generate first stage meta-events based on analyzing time attributes of base events received from at least one network component; generate second stage meta-events based on a number of the first stage meta-events that have a time attribute falling within a time period; determine an amount of time that has passed since a most-recent second stage meta-event was generated; and if a threshold time period does not exceed the amount of time that has passed since the most-recent second stage meta-event was detected, generate a third stage meta-event.
 This application is a continuation of U.S. patent application Ser. No. 10/308,767, filed Dec. 2, 2002, entitled "CORRELATION ENGINE WITH SUPPORT FOR TIME-BASED RULES", which is incorporated by reference in its entirety.
FIELD OF THE INVENTION
 The present invention relates to a computer-based system for the identification and processing of security events from heterogeneous sources, including a correlation engine with support for time--based rules.
 Computer networks and systems have become indispensable tools for modern business. Today terabits of information on virtually every subject imaginable are stored in and accessed across such networks by users throughout the world. Much of this information is, to some degree, confidential and its protection is required. Not surprisingly then, intrusion detection systems (IDS) have been developed to help uncover attempts by unauthorized persons and/or devices to gain access to computer networks and the information stored therein.
 Intrusion detection may be regarded as the art of detecting inappropriate, incorrect or anomalous activity within or concerning a computer network or system. The most common approaches to intrusion detection are statistical anomaly detection and pattern-matching detection. IDS that operate on a host to detect malicious activity on that host are called host-based IDS (and may exist in the form of host wrappers/personal firewalls or agent-based software), and those that operate on network data flows are called network-based IDS. Host-based intrusion detection involves loading software on the system (the host) to be monitored and using log files and/or the host's auditing agents as sources of data. In contrast, a network-based intrusion detection system monitors the traffic on its network segment and uses that traffic as a data source. Packets captured by the network interface cards are considered to be of interest if they match a signature.
 Regardless of the data source, there are two complementary approaches to detecting intrusions: knowledge-based approaches and behavior-based approaches. Almost all IDS tools in use today are knowledge-based. Knowledge-based intrusion detection techniques involve comparing the captured data to information regarding known techniques to exploit vulnerabilities. When a match is detected, an alarm is triggered. Behavior-based intrusion detection techniques, on the other hand, attempt to spot intrusions by observing deviations from normal or expected behaviors of the system or the users (models of which are extracted from reference information collected by various means). When a suspected deviation is observed, an alarm is generated.
 Advantages of the knowledge-based approaches are that they have the potential for very low false alarm rates, and the contextual analysis proposed by the intrusion detection system is detailed, making it easier for a security officer using such an intrusion detection system to take preventive or corrective action. Drawbacks include the difficulty in gathering the required information on the known attacks and keeping it up to date with new vulnerabilities and environments.
 Advantages of behavior-based approaches are that they can detect attempts to exploit new and unforeseen vulnerabilities. They are also less dependent on system specifics. However, the high false alarm rate is generally cited as a significant drawback of these techniques and because behaviors can change over time, the incidence of such false alarms can increase.
 With both knowledge-based and behavior-based systems, matches are detected with the aid of a rules engine. Many current rules engines implement a standard RETE algorithm because the rules engine's performance is demonstrably independent of the number of rules that are used.
 Regardless of whether a host-based or a network-based implementation is adopted and whether that implementation is knowledge-based or behavior-based, an intrusion detection system is only as useful as its ability to discriminate between normal system usage and true intrusions (accompanied by appropriate alerts). If intrusions can be detected and the appropriate personnel notified in a prompt fashion, measures can be taken to avoid compromises to the protected system. Otherwise such safeguarding cannot be provided. Accordingly, what is needed is a system that can provide accurate and timely intrusion detection and alert generation so as to effectively combat attempts to compromise a computer network or system
SUMMARY OF INVENTION
 A rules engine with support for time-based rules is disclosed. A method performed by the rules engine, comprises receiving security events generated by a number of network devices. The security events are aggregated. One or more time-based rules are provided to a RETE engine. The aggregated security events are provided to the RETE engine at specific times associated with the time-based rules. The security events are cross-correlated with the one or more time-based rules; and one or more first stage meta-events are reported.
BRIEF DESCRIPTION OF THE DRAWINGS
 The present invention is illustrated by way of example, and not limitation, in the figures of the accompanying drawings in which:
 FIG. 1 illustrates one embodiment of a computer-based system for capturing, normalizing and reporting security events from heterogeneous sources configured in accordance with the present invention;
 FIG. 2 illustrates procedures followed by an agent configured in accordance with an embodiment of the present invention when collecting, normalizing and reporting security event data;
 FIG. 3 illustrates procedures followed by a manager configured in accordance with an embodiment of the present invention when analyzing security event data and generating alerts based thereon;
 FIG. 4 illustrates one embodiment of a rules engine in accordance with the present invention; and
 FIGS. 5A and 5B illustrate a set of procedures followed by a rules engine in accordance with an embodiment of the present invention.
 Described herein is a computer-based system for the identification and processing of security events from heterogeneous sources, including a correlation engine with support for time-based rules. The system (one embodiment of which is manifest as computer software), implements a method that comprises receiving security events generated by a number of network devices. The security events are aggregated. One or more time-based rules are provided to a RETE engine. The aggregated security events are provided to the RETE engine at specific times associated with the time-based rules. The security events are cross-correlated with the one or more time-based rules; and one or more first stage meta-events are reported.
 Although the present system will be discussed with reference to various illustrated examples, these examples should not be read to limit the broader spirit and scope of the present invention. For example, the examples presented herein describe distributed agents, managers and consoles, which are but one embodiment of the present invention. The general concepts and reach of the present invention are much broader and may extend to any computer-based or network-based security system. Also, examples of the messages that may be passed to and from the components of the system and the data schemas that may be used by components of the system are given in an attempt to further describe the present invention, but are not meant to be all-inclusive examples and should not be regarded as such.
 Some portions of the detailed description that follows are presented in terms of algorithms and symbolic representations of operations on data within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the computer science arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of steps leading to a desired result. The steps are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared and otherwise manipulated. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers or the like. It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise, it will be appreciated that throughout the description of the present invention, use of terms such as "processing", "computing", "calculating", "determining", "displaying" or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.
 As indicated above, one embodiment of the present invention is instantiated in computer software, that is, computer readable instructions, which, when executed by one or more computer processors/systems, instruct the processors/systems to perform the designated actions. Such computer software may be resident in one or more computer readable media, such as hard drives, CD-ROMs, DVD-ROMs, read-only memory, read-write memory and so on. Such software may be distributed on one or more of these media, or may be made available for download across one or more computer networks (e.g., the Internet). Regardless of the format, the computer programming, rendering and processing techniques discussed herein are simply examples of the types of programming, rendering and processing techniques that may be used to implement aspects of the present invention. These examples should in no way limit the present invention, which is best understood with reference to the claims that follow this description.
 Referring now to FIG. 1, an example of a computer-based system 10 architected in accordance with an embodiment of the present invention is illustrated. System 10 includes agents 12, one or more managers 14 and one or more consoles 16 (which may include browser-based versions thereof). In some embodiments, agents, managers and/or consoles may be combined in a single platform or distributed in two, three or more platforms (such as in the illustrated example). The use of this multi-tier architecture supports scalability as a computer network or system grows.
 Agents 12 are software programs that provide efficient, real-time (or near real-time) local event data capture and filtering from a variety of network security devices and/or applications. The primary sources of security events are common network elements including firewalls, intrusion detection systems and operating system logs. Agents 12 can collect events from any source that produces event logs or messages and can operate at the native device, at consolidation points within the network, and/or through simple network management protocol (SNMP) traps.
 Managers 14 are server-based components that further consolidate filter and cross-correlate events received from the agents, employing a rules engine 18 and a centralized event database 20. One role of manager 14 is to capture and store all of the real-time and historic event data to construct (via database manager 22) a complete, enterprise-wide picture of security activity. The manager 14 also provides centralized administration, notification (through one or more notifiers 24), and reporting, as well as a knowledge base 28 and case management workflow. The manager 14 may be deployed on any computer hardware platform and one embodiment utilizes an Oracle® database. Communications between manager 14 and agents 12 may be bi-directional (e.g., to allow manager 14 to transmit commands to the platforms hosting agents 12) and encrypted. In some installations, managers 14 may act as concentrators for multiple agents 12 and can forward information to other managers (e.g., deployed at a corporate headquarters).
 Consoles 16 are computer- (e.g., workstation-) based applications that allow security professionals to perform day-to-day administrative and operation tasks such as event monitoring, rules authoring, incident investigation and reporting. Access control lists allow multiple security professionals to use the same system and event database, with each having their own views, correlation rules, alerts, reports and knowledge base appropriate to their responsibilities. A single manager 14 can support multiple consoles 16.
 In some embodiments, a browser-based version of the console 16 may be used to provide access to security events, knowledge base articles, reports, notifications and cases. That is, the manager 14 may include a web server component accessible via a web browser hosted on a personal computer (which takes the place of console 16) to provide some or all of the functionality of a console 16. Browser access is particularly useful for security professionals that are away from the consoles 16 and for part-time users. Communication between consoles 16 and manager 14 is bi-directional and may be encrypted.
 Through the above-described architecture the present invention can support a centralized or decentralized environment. This is useful because an organization may want to implement a single instance of system 10 and use an access control list to partition users. Alternatively, the organization may choose to deploy separate systems 10 for each of a number of groups and consolidate the results at a "master" level. Such a deployment can also achieve a "follow-the-sun" arrangement where geographically dispersed peer groups collaborate with each other by passing primary oversight responsibility to the group currently working standard business hours. Systems 10 can also be deployed in a corporate hierarchy where business divisions work separately and support a rollup to a centralized management function.
 Examining each of the various components in further detail, we begin with the agents 12. Agents 12 are used to collect, reduce and normalize the enormous amount of data that is generated by a network's security devices before a manager 14 acts on the data. As will become evident, this process goes beyond simple log consolidation. Before presenting those details, however, and to understand why such measures are desirable, some background regarding how analysts currently cope with security event information generated by multiple network devices is useful.
 Conventional intrusion detection systems can help an analyst detect an attack directed at a network resource such as a server. Usually, such investigations are launched in response to an alert generated by the IDS. As a first step after receiving such an alert, an analyst might review perimeter router logs to see if a router associated with the network passed a packet that triggered the alert. If such a packet were discovered, the analyst would likely then want to review one or more firewall logs to see if any existing filters blocked the suspect packet. Assume, for the sake of this example, the suspect packet got past any firewalls; further investigation would be necessary to determine whether the integrity of the server itself was compromised. Such an integrity check may be performed using a conventional software application such as Tripwire, which is a file integrity checker employing MD5 checksums, to see which files, if any, had been accessed or modified. Finally, the analyst may have to examine a Syslog or an EventLog from the subject server, as well as any tcpdump data collected by a dedicated tcpdump host, for the segment of time surrounding the attack to determine what actually happened.
 By this time the analyst has accessed many different systems and looked at several different types of logs in an effort to distill a comprehensive view of the attack. This can be a significant amount of work, and time taken in such review and analysis is time lost from the vitally important tasks of securing the network and restoring the compromised server to make sure that no other systems will be affected. The present invention helps to minimize the time spent on such analysis by consolidating all the relevant information in a single logging facility, allowing the analyst to look at the data in whatever sequence or depth he or she requires.
 More than just consolidation, though, the present agents 12 provide data normalization, which is of great benefit when an analyst must deal with security incidents in a heterogeneous network environment. To understand why normalization is helpful consider a typical enterprise environment, which consists of many different types of network devices ranging from border routers and VPN devices, to firewalls and authentication servers, and a wide range of application servers such as web servers, e-mail servers and database servers. Each of these devices generates logs that, as described above, are sources of data to a security analyst. However, it is seldom, if ever, the case that two manufactures will use the same event logging mechanism or format their event logs identically. For example a Cisco Systems PIX® firewall will not report an accepted packet in the same way as a Check Point firewall or even in the same fashion as a Cisco Systems router.
 An example of the types of various reports that might be generated by different network devices is presented below in Table 1, which shows examples of logs from different network devices, each reporting the same packet traveling across a network. In particular, these logs represent a remote printer buffer overflow that connects to IIS servers over port 80.
TABLE-US-00001 TABLE 1 Examples of Event Logs for Different Network Devices. Network Device Event Log Check Point firewall ''14'' ''21Dec2001'' ''12:10:29'' ''eth-s1p4c0'' ''ip.of.firewall'' ''log'' "accept'' ''www-http'' ''188.8.131.52'' ''10.10.10.10'' ''tcp'' ''4'' ''1355'' '' '' '' '' '' '' '' '' '' '' '' '' '' '' '' '' '' '' ''firewall'' '' len 68'' Cisco Systems router Dec 21 12:10:27: %SEC-6-IPACCESSLOGP: list 102 permitted tcp 184.108.40.206(1355) -> 10.10.10.10(80), 1 packet Cisco Systems PIX Dec 21 2001 12:10:28: %PIX-6-302001: Built inbound TCP connection firewall 125891 for faddr 220.127.116.11/1355 gaddr 10.10.10.10/80 laddr 10.0.111.22/80 Snort [**] [1:971:1] WEB-IIS ISAPI .printer access [**] [Classification: Attempted Information Leak] [Priority: 3] 12/21-12:10:29.100000 18.104.22.168:1355 -> 10.10.10.10:80 TCP TTL:63 TOS:0x0 ID:5752 IpLen:20 DgmLen:1234 DF ***AP*** Seq: 0xB13810DC Ack: 0xC5D2E066 Win: 0x7D78 TcpLen: 32 TCP Options (3) => NOP NOP TS: 493412860 0 [Xref => http://cve.mitre.org/cgi- bin/cvename.cgi?name=CAN-2001-0241] [Xref => http://www.whitehats.com/info/IDS533]
 The Check Point record contains the following fields: event id, date, time, firewall interface, IP address of the firewall interface, logging facility, action, service, source IP, target IP, protocol, source port, some Check Point specific fields and then the size of the datagram. This report is, to say the least, difficult for a human analyst to read (especially with all the empty fields that are represented by double quotes). The Cisco router has a different format: date, time, logging facility, event name, source IP, source port, target address, target port, and number of packets. The Cisco PIX firewall, which is produced by the same manufacturer as the router, uses yet another format: date, time, event name, source IP, source port, translated address or target address, target port, local address, and local port.
 The final record is a Snort alert that claims this traffic was malicious. Snort is a well-known IDS and the fields it populates are: exploit or event name, classification, priority, date, time, source IP, source port, target IP, target port, protocol, TTL (time to live), type of service, ID, IP length, datagram length, tcp flags, sequence number, acknowledgement number, window size, and tcp length. Snort also reports additional data such as references to investigate the exploit.
 Agents 12 may be deployed in connection with some or all of these (and other) network components and applications. For example, in FIG. 1, agent 12a is deployed in connection with an IDS (such as Snort). Agent 12b is deployed in connection with a firewall (such as the Check Point firewall and/or the Cisco PIX firewall). Agent 12c is deployed in connection with other network components or agents (e.g., a router). Each of these agents receives the event information from its associated network device or application in that device's or application's native format and converts (or normalizes) the information to a common schema. This normalization allows for later storage of the event information in a format that can more readily be utilized by an analyst.
 Many normalized schemas can be used and, in general, choosing the fields of a common schema may be based on content rather than semantic differences between device logs and/or manufacturers. To accomplish this normalization, agents 12 are equipped with a parser configured to extract values from the events as reported by the individual network devices/applications and populate the corresponding fields in the normalized schema. Table 2 is an example of a normalized schema for the data reported by the devices in Table 1.
TABLE-US-00002 TABLE 2 Common Schema Representation of Event Data Event Device Date Time Name Src_IP Src_Port Tgt_IP Trg_Port Type Additional data Dec. 21, 2001 12:10:29 accept 22.214.171.124 1355 10.10.10.10 80 Check Point Dec. 21, 2001 12:10:27 list 102 126.96.36.199 1355 10.10.10.10 80 Cisco permitted Router tcp Dec. 21, 2001 12:10:28 built 188.8.131.52 1355 10.10.10.10 80 Cisco inbound PIX tcp connection Dec. 21, 2001 12:10:29 WEB-IIS 184.108.40.206 1355 10.10.10.10 80 Snort TCP TTL: 63 ISAPI. TOS: 0x0 ID: 5752 printer IpLen: 20 access DgmLen: 1234 DF ***AP*** Seq: 0xB13810DC Ack: 0xC5D2E066 Win: 0x7D78 TcpLen: 32 TCP Options (3) =>NOP NOP TS: 493412860 0
 Table 2 reports the same four events described earlier, this time in a normalized fashion. Each of the agents 12 is configured to extract the relevant data from events reported by its associated network device/application and map that data to the corresponding common schema representation. For instance the Check Point firewall reports a target port as www-http, not as port 80 as is the case for most other network devices. Therefore an agent 12 associated with the Check Point firewall is configured with an appropriate lookup mechanism (e.g., a table) to ensure that "www-http" as reported by the firewall gets translated into "port 80" when the agent 12 reports the event to the manager 14.
 Similarly, the agents 12 may need to be configured to convert the date/time stamp formats used by the various network devices/applications into a common date/time representation. That is, because the different network devices/applications all use different date/time formats, the agents cannot simply report the date/time stamps reported by the device/application. Instead, the agents 12 may be configured to convert local date/time stamps to a universal date/time notation, such as Greenwich Mean Time.
 In addition to normalizing event data by fields, agents 12 can parse the event data stream and set field values based on conventions and practices of the organization. For example, the variety of event severity levels that devices produce can all be normalized at the agent level into a single, consistent hierarchy.
 Thus, agents 12 collect and process events generated by heterogeneous network devices/applications throughout an enterprise. Alerts can come from routers, e-mail logs, anti-virus products, firewalls, intrusion detection systems, access control servers, VPN systems, NT Event Logs, Syslogs, and other sources where security threat information is detected and reported. In some embodiments, each event generator has an agent 12 assigned to collect all relevant security information, while in other embodiments agents are shared among two or more event generators. Thus, depending on the device/application to be monitored and the in-place infrastructure, a choice is provided for simple log parsing and loading, network listening (e.g., through SNMP traps), installation on aggregation points (Syslog servers and concentrators) and full distribution to all security-relevant devices.
 In addition to collecting and normalizing data from security devices, the agents 12 intelligently manage the data with:  Filtering: each agent 12 can be configured according to conditions by which data will be collected and sent to the manager 14. This helps to reduce the need to collect and manage large volumes of unwanted data.  Aggregation: Based on the time period selected, the agents 12 can collect duplicate alerts but send only a single message with a count of the total number of such alerts to the manager 14. This helps reduce the amount of traffic transmitted across the network.  Batching: Agents 12 can be configured to send a collection of alerts at one time rather than sending alerts immediately after each occurrence.
 FIG. 2 illustrates the various processes performed by agents 12 from the point of view of the event information. Initially, at step 30, the raw event information is received or collected from the native network device or application in that device's/application's native format. At this point (or, optionally, following normalization), data filters may be applied to reduce the volume of data being passed for further analysis (step 32). Such filtering is optional and may involve assessing the captured data against one or more conditions to determine whether or not the data is relevant for further analysis.
 Thereafter, the event data is normalized at step 34. As indicated above, the normalization may occur at the field and/or the field value level. Further, the normalization may involve translation of the field values into nomenclatures/formats used across an enterprise.
 Following normalization, the event data may, optionally, be aggregated (step 36) before being transmitted to the manager 14 (step 38). The transmissions may occur as the events are captured or may be made on a batched basis. In either case, the messages used to transmit the event data preferably include all of the source fields of an event. By delivering the entire event data set (i.e., all of the source fields) organized in a consistent format (i.e., the common schema), powerful upstream data management, cross-correlation, display and reporting is available to the security team. In some embodiments the event data is discarded after successful transmission to the manager 14, but in other cases the data may be cached for a time at the agent 12 to permit later replay of the data.
 Referring again to FIG. 1, the manager 14 includes one or more agent managers 26, which are responsible for receiving the event data messages transmitted by the agents 12. Where bi-directional communication with the agents 12 is implemented, these agent managers 26 may be used to transmit messages to the agents 12. If encryption is employed for agent--manager communications (which is optional), the agent manager 26 is responsible for decrypting the messages received from agents 12 and encrypting any messages transmitted to the agents 12.
 Once the event data messages have been received (and if necessary decrypted), the event data is passed to the rules engine 18. Rules engine 18 is at the heart of the manager 14 and is used to cross-correlate the event data with security rules in order to generate meta-events. Meta-events, in the context of the present invention, are instances of (usually) multiple individual event data elements (gathered from heterogeneous sources) that collectively satisfy one or more rule conditions such that an action is triggered. Stated differently, the meta-events represent information gathered from different sensors and presented as correlated results (i.e., the decision output of the rules engine 18 indicating that different events from different sources are associated with a common incident as defined by one or more rules).
 The actions triggered by the rules may include notifications transmitted (e.g., via notifier 24) to designated destinations (e.g., security analysts may be notified via the consoles 16, e-mail messages, a call to a telephone, cellular telephone, voicemail box and/or pager number or address, or by way of a message to another communication device and/or address such as a facsimile machine, etc.) and/or instructions to network devices (e.g., via agents 12 or via external scripts or programs to which the notifier 24 may pass arguments) to take action to thwart a suspected attack (e.g., by reconfiguring one or more of the network devices, and or modifying or updating access lists, etc.). The information sent with the notification can be configured to include the most relevant data based on the event that occurred and the requirements of the analyst. In some embodiments, unacknowledged notifications will result in automatic retransmission of the notification to another designated operator.
 As discussed below, when meta-events are generated by the rules engine 18, on-screen notifications may be provided to consoles 16 to prompt users to open cases for investigation of the events which led to the notification. This may include accessing knowledge base 28 to gather information regarding similar attack profiles and/or to take action in accordance with specified procedures. The knowledge base 28 contains reference documents (e.g., in the form of web pages and/or downloadable documents) that provide a description of the threat, recommended solutions, reference information, company procedures and/or links to additional resources. Indeed, any information can be provided through the knowledge base 28. By way of example, these pages/documents can have as their source: user-authored articles, third-party articles, and/or security vendors' reference material.
 The rules engine 18 is based on a RETE engine configured to preserve event information state over configurable time windows so as to provide correlation of the event data according to specified rules. Correlation is generally regarded as a process of bringing information items into mutual relation. In the context of the present invention, correlation through rules engine 18 provides the ability to access, analyze, and relate different attributes of events from multiple sources to bring something to the attention of an analyst that might (or likely would) have otherwise gone unnoticed. In other words, the rules engine 18 provides the ability to determine what type of incident is represented by a collection of events reported by a number of heterogeneous network devices and/or applications. Because the collected event data is normalized into a common event schema, correlation can be performed utilizing any field including, but not limited to, geography, device type, source, target, time thresholds, and/or event type. Based on alerts generated by the rules engine 18, operators are provided with a workflow for investigating these incidents.
 Turning to FIG. 3, the manager 14 receives (step 40) and analyzes (step 42) the event data reported by agents 12 in real-time (or near real-time owing to network latencies and depending upon whether or not batched message transmission is used) according to a set of flexible rules. The rules define which events generate an alert, when those events generate an alert, and what actions are associated with the alert. Hence, the rules may be written to contain event conditions, thresholds, and actions. In some embodiments the rule conditions may be specified using Boolean operators and/or database queries. When incoming events match a particular rule's conditions and thresholds, causing a meta-event to be generated (step 44), the rule automatically fires the action that has been defined (step 46). Such actions can include, but are not limited to: executing a pre-determined command or script, logging the alert, sending the alert to the consoles 16, sending the alert to notification designees, setting custom severity levels for the alert based on cumulative activity, adding a source to a suspicious list or a target to a vulnerable list, and/or a combination of these actions.
 Rules may be created at the manager 14 and/or at the consoles 16 using a flexible scripting language. An example of a rule might be:  If(an ids evasion attack) occurs (from the same source ip address) (3 times) within (2 minutes) then (send message to console) and (notify the security supervisor via pager). In this example, the incoming event data would be compared against the rule conditions and thresholds (in the above example 3 events that satisfy the condition of an IDS evasion attack are required and all must originate from a common source IP address and be detected within 2 minutes of each other), and if those criteria are satisfied the designated actions (here, sending an alert message to the consoles 16 and also notifying a security supervisor via a pager) would be performed. The correlation rules that operate on the events evaluate threats and attacks according to selected criteria (e.g., degree of threat, level of success, vulnerability of target and value of target) and generate alerts according to a security intelligence taxonomy that focuses attention on the most dangerous and potentially most damaging attacks. For example, threats to network assets that are deemed not to have succeeded or that are not likely to succeed may be coded green, while those that have succeeded or have a high probability of success might be coded red. The value of the security information taxonomy lies in its ability to eliminate false positives while clearly identifying real threats to vulnerable and valuable assets.
 In general, the rules may be designed to capture threats and attacks that are typical in large, diverse networks and may be organized to provide multiple lines of defense by detecting specific activities and grouping them according to level of threat:  Reconnaissance zone transfer, port scan, protocol, scanning, etc.  Suspicious illegal outgoing traffic, unusual levels of alerts from the same host, etc.  Attack overflow, IDS evasion, virus, denial of service, etc.  Successful compromise of a backdoor, root compromise, covert channel exploit, etc. Similar events and signatures may be grouped into rule categories that can be utilized by the rules to insulate the rule from changes in vendor-specific event details. For example, event names may change between product releases or new devices may be added to the network infrastructure with a new set of nomenclature. Since the rule categories map similar signatures into a single name that is used by the rules engine, if an individual network device changes taxonomy, only the mapping is changed, not the rule definition. Therefore, despite changes in individual devices, the investment in custom defined rules is preserved.
 After the events are processed by rules engine 18, the raw event data as well as any meta-events that were generated are stored in database 20 (step 48). In some embodiments, the raw event data may be stored prior to or concurrently with processing of the data by rules engine 18. Regardless of the sequence, such storage of the event data (and the meta events generated by the rules engine 18) preserves a historical record of the event traffic and allows for replaying of the events through an existing or a new rule set (either at the manager 14 or the consoles 16) in order to assess the efficacy of new rules, for training purposes, and/or for case investigation.
 Correlation via the rules ensures that credible threats and attacks come to the attention of the security staff on a high-priority basis. Hence once an alert is received, the operator can perform in-depth analysis and take aggressive action secure in the knowledge that the effort is well spent. When a rule match is reported to a console 16, the analyst can quickly drill down (through an associated graphical user interface) to see all of the individual events that caused the rule to fire. If necessary, the analyst can investigate even further to see all of the individual data elements captured for those events.
 When action is required, the present invention provides a full set of tools and services for the operator. Resources such as the rule definition, a knowledge base article containing company policies and recommended actions, and the development of a complete case docket describing the problem assist the operator in responding immediately to critical security threats. If necessary, the operator can proactively deal with an attack by launching specific applications or scripts from the console 16 to reconfigure device settings or change access privileges.
 The console 16 provides a centralized view into the security status of an enterprise and gives administrators, analysts, and operators an interface to perform security management tasks. In various embodiments, the console provides event display in real-time or in replay mode (i.e., the ability to playback events from a given time period according to a VCR or DVD metaphor). Replay may be had from the events stored in database 20 or, in some instances, from caches associated with agents 12. This latter form of replay is especially useful because it provides improved simulation of actual network conditions as the events are played out across the same network as during the original attack.
 Consoles 16 also provide operators with complete drill-down capability from the highest level of detail (e.g., the entire rage of events) to the lowest level of detail (e.g., fields within a single event). This allows analysts to probe at whatever level of detail is required to gain further insight into an attack and assess vulnerability. This varying level of detailed analysis is made possible because the agents 12 report all of the event data fields, not merely a subset thereof. By way of example, one tool provides analysts with the ability to quickly see similar characteristics of events using a cursor control operation, such as a mouse click. For example, if analysts are presented with a meta-event alert that consists of, say, twenty or more individual events reported by several different agents associated with different network devices, the present user interface associated with consoles 16 allows the analyst to quickly visualize only the common fields of these events (e.g., such as a source IP address) by simply highlighting the events and performing a mouse click/select operation.
 Once security personnel have been notified of a meta-event, they can utilize the knowledge base to determine the appropriate actions. In addition, security analysts may undertake investigations of events and/or meta-events. In general, such matters can be assigned to so-called cases. Stated differently, cases create a workflow and oversight environment for situations where there are suspicious events requiring further investigation. Once a case is created, it can be assigned to an operator, investigated, and resolved based on the business policies and practices of the enterprise (e.g., as documented in knowledge base 28). The security staff can also add narration and event information to a case, or view open cases to determine their status and any required next steps.
 Consoles 16 also provide a front-end for the administration of the entire system 10. This may include system configuration such as setting up operators, notification, agent behavior, etc. User management (such as creating and modifying users, access, roles, and responsibilities), rules management (e.g., authoring, viewing, and updating rules), and workflow management (e.g., setting up the flow of actions taken when an event is received) may also be handled through the consoles 16. Finally, the consoles 16 allow for remote access, thus supporting divisional responsibility and "follow-the-sun" management.
 Having thus described the elements of system 10, it is helpful to present a more in depth look at rules engine 18. As described above, rules engine 18 is used to cross-correlate the event data with security rules in order to generate meta-events. FIG. 4 illustrates one embodiment of a rules engine 18 configured in accordance with the present invention. Rules engine 18 includes six components; a partial matcher 410, a memory manager 420, a RETE engine 430, a time tracker 440, a rules manager 490, and an action engine 450.
 The partial matcher 410, receives events via the agent manager 26. The events may be gathered from different security devices. Partial matcher 410 determines which rule in the system is interested in a particular event it received. An event is considered to be interesting if a rule used by the system mentions one or more attributes of the event. For example, an event can be considered interesting if the event has a particular source address from a particular subnet. Partial matcher 410 groups, batches, or aggregates interesting events together that are related to one another and one or more conditions of a particular rule. Additionally, the partial matcher 410 is aware of time windows associated with a rule used by the system. By knowing the time window associated with a rule, partial matcher 410 can compute the last moment in time at which that rule remains interesting, and should be processed by RETE engine 430.
 The memory manager 420 keeps track of the events that are operated on by RETE engine 430. Aggregated events are passed from the partial matcher 410 to the memory manager 420. The aggregated events can have an expiration time, (i.e., the aggregated event is only of interest for a period of time.) Memory manager 420 provides events to RETE Engine 430 and deletes the events when the event reaches its expiration time. In other words, memory manager 420 feeds and deletes events to and from the RETE engine 430 as needed to provide statefulness.
 RETE engine 430 implements the RETE algorithm which scales to many hundreds of rules while its performance is independent of the number of rules it considers. In operation, RETE engine 430 loads user-written rules, that are time-based and once the rule is active, engine 430 analyzes the events provided by memory manager 420 and generates a first stage meta-event which can result in the performance of an action in response to the correlated events. More specifically, the RETE engine 430, reports instances where the rules are satisfied.
 Rules manager 490 provides user-defined, time-based rules to the RETE engine 430. The user-defined rules are generated via console/browser interfaces 16. Furthermore, a user can provide instructions to rules manager 490 for activating or deactivating rules dynamically.
 Time tracker 440 allows the RETE engine 430 to process time-based rules. Time-based rules are triggered when events that occur over a period of time collectively are recognized as being associated with similar occurrences. Time tracker 440 receives meta-events generated by RETE engine 430 and groups related meta-events together. This correlation of the meta-events is used to determine if the threshold of a time-based rule is reached. For example, time-based rules can require that an event occur ten times in an hour to signify that an action need be performed. If that threshold is not reached, the group of events is terminated. If the threshold is reached, then a second stage meta-event is generated. Time tracker 440 communicates with action engine 450, which executes the actions specified in the rule loaded into RETE engine 430. Meta-events are a hypothesized description of the real world scenario of what various sensors of security devices independently report as events. These meta-events may then be fed back into rules engine 18 to be used as an event. Action engine 450 can notify the user that a meta-event occurred via email, website, or with notifier on the console. Additionally, time tracker 440 reports a third stage meta-event if no second stage meta-events occur or when second stage meta-events cease to occur. The third stage meta-event detects the end of a security attack and indicates the magnitude of the attack.
 Although the functional blocks of a rules engine are depicted in one embodiment within rules engine 18, one or more of these functional blocks can be distributed in other systems. Rules engine 18 has additional functionalities, such as detection of improper rule syntax, loop detection, rule feedback detection, aggregation of joint events, and timeline alignment. Detection of improper rule syntax allows for the deactivation of rules that are abusive to the system, such as inefficient memory consumption, or CPU usage. Rules engine 18 can also detect if a user defined rule generates a loop condition in which the same events are provided repeatedly to the RETE engine 430. Similarly, rules engine 18 can detect rule feedback. Rule feedback occurs when a meta-event that is generated by rules engine 18 is fed back into rules engine 18 and results in an abusive or destructive consumption of memory or CPU processing.
 FIG. 5 illustrates procedures followed by a rules engine 18 in accordance with an embodiment of the present invention. Partial matcher 410 receives an event (step 510) and determines if the event is interesting as described above (decision block 530). If the event is not of interest, the next event is received by partial matcher 410 (step 520) and the process determines if the next event is interesting (decision block 530). If an event is determined to be of interest, these interesting events are aggregated by partial matcher 410 (step 530). In parallel with the processing performed by partial matcher 410, rules manager 490 compiles a user-defined, time-based rule (step 545) associated with the aggregated events of interest. The compiled rule and aggregated events are provided to RETE engine 430 (step 550) for a time period specified by the rule. If the time period expires, the time-based rule is no longer provided to the RETE engine 430.
 Upon receipt of the compiled rule and interesting events, the RETE engine 430 determines if one or more of the aggregated events matches a processed rule (decision block 560). If no matches occur, then new events are received at rules engine 18 (step 520), and the process described above is repeated. If a match occurs, then a first stage meta-event is generated (step 570). Although not shown, first stage meta-events can be reported, via the console 16 or with another reporting mechanism described above.
 First stage meta-events are aggregated and processed by time tracker 440 when determining if in subsequent time periods, whether a threshold level of repeat matches of similar interesting events occur (decision block 580). For example, if five first stage meta-events occurred in a first ten minute time period, time tracker 440 determines if five more first stage meta-events occur in a subsequent ten minute time period. If the threshold is met (i.e., five meta-events occur in ten minutes), then a second stage meta-event is generated (step 590). After generating a second stage meta-event, the process described above repeats, by accepting events at a rules engine 18 (step 520). Additional second stage meta-events can occur, however, if no second stage meta-events occur or stop occurring in subsequent time periods, a third stage meta-event is generated (step 599). The generation of a third stage meta-event signifies that a security attack on the system has ended. It also measures how long an attack was, and the attack's magnitude in terms of the number of network computers attacked. Both second and third stage meta-events can be reported as actions performed by action engine 450. By reporting second stage meta-events, the detection of repetitive behavior attacks and enterprise wide attacks is improved. Process 500 can occur within the process shown in FIG. 3, above. Specifically, process 500 describes how event data is processed through rules engine 18, as shown at step 42 of FIG. 3.
 Thus, a computer-based system for capturing correlating and reporting security events from heterogeneous sources, including a correlation engine with support for time-based rules has been described. In the foregoing description, the various examples and embodiments were meant to be illustrative of the present invention and not restrictive in terms of their scope. Accordingly, the invention should be measured only in terms of the claims, which follow.
Patent applications by Debabrata Dash, Sunnyvale, CA US
Patent applications by Hugh S. Njemanze, Los Altos, CA US
Patent applications by Shijie Wang, San Jose, CA US
Patent applications in class POLICY
Patent applications in all subclasses POLICY