IL221174A - System and method of rule engine configuration - Google Patents

Description

221,174/2 SYSTEM AND METHOD OF RULE ENGINE CONFIGURATION RELATED APPLICATIONS The present application is being filed concurrently with a related application titled "System and Method of High Volume Rule Engine".

FIELD OF THE DISCLOSURE The present disclosure relates generally to rule engines, and particularly to real-time rule-engines required to operate with high throughputs.

BACKGROUND OF THE DISCLOSURE Packet based communication networks include nodes which perform various tasks on the packets passing on the network. These tasks include, for example, firewall access control, traffic routing, QoS (Quality of Service) implementation and traffic probing. As different packets are handled differently, network nodes generally employ a rule engine which is configured with rules for handling packets . Each rule identifies a group of packets and indicates the handling to be applied to that group of packets. Packets received by the node are compared to the rules until a match is found and then the handling indicated by the matching rule is applied to the packet and generally to all subsequent packets belonging to the same session (referred to herein also as a packet connection) .

When a rule engine needs to apply a large number of rules, its performance may suffer due to the time needed in order to check all the rules, the amount of memory required and the increasing number of memory accesses, which are relatively slow operations. 1011-1066 US patent 7,139, 837 to Parekh et al. describes a rule engine which traverses a mesh having path nodes and path edges arranged in a tree part and graph part. The rule engine manages session entries for packets, such that the rule checking performed for a first packet of a session can be used to speed up the handling of further packets belonging to the session .

US patent 6,857,018 to Jiang describes using a multiple dimension spatial indexing and mapping to speed up rule lookup in a table.

US patent 8,005,945 to Cohen et al. describes adjusting the rules to segregate their ranges and minimize the number of ΊG LI 1 G S * A paper titled: "Deep Packet Inspection using Parallel Bloom Filters", by Sarang Dharmapurikar, Praveen Krishnamurthy Todd Sproull and John Lockwood, Hot Interconnects 11, IEEE Computer Society, pp . 52-61, Jan. 2004, suggests using Bloom filters to identify packets that potentially include predefined strings in their payload and then using an independent process to eliminate false positives.

A paper titled: "Fast Pattern-Matching Techniques for Packet Filtering, by Alok S. Tongaonkar, Master of Science in Computer Science, Stony Brook University, May 2004, describes applying techniques for pattern matching in packet filtering systems .

A paper titled .assification for Core Routers Is there an Alternative to CAMs?", by Florin Baboescu, Sumeet Singh, (Seorge Varghese, IEEE Infocom, 2003, suggests filtering packets using a two dimensional filter on the IP source and through the received results 1011-1066 US patent publication 2011/0102157 to Koma describes using a Bloom filter to filter out packe that should not be sent to a receiver.

US patent 2010/0195653 to Jacobson describes packet routing using a longest-prefix-match lookup engine which may be implemented using a TCAM, tree structure or Bloom filter. US patent publication 2009/0182867 to Milliken et al. describes a method of identifying malicious packets by comparing results of applying a hash function to packets with previously generated hash results of the malicious packets.

SUMMARY OF THE DISCLOSURE An embodiment that is described herein provides a rule engine which is designed to check packets for a large number of rules at high speeds.

There is therefore provided in accordance with am embodiment of the present invention, a rule engine controller, including an input interface adapted to receive rules relating to a plurality of fields of filtered packets and a rule manager processor configured to receive a batch of rules through the input interface, define one or more anchor sets of fields of the filtered packets, such that each of the rules in the batch of received rules includes conditions relating to the fields of at least one of the anchor sets, assign each of the rules to one of the anchor sets and configure the rules assigned to each anchor set into a respective filtering unit, such that the filtering unit begins the filtering of packets for the rules assigned to the anchor set with the fields of the anchor set.

Optionally, the rule manager processor is configured to define the anchor sets, such that at least one of the anchor sets includes a plurality of fields. Optionally, the rule manager processor is configured to define the anchor sets, 1011-1066 such that at least two of the anchor sets include a mutual field included in both the anchor sets. Optionally, the rule manager processor is configured to evaluate a plurality of different possible one or more anchor sets for the batch of received rules and to select one of the possible sets which is evaluated to achieve a fastest filtering of packets for the received batch of rules. Optionally, the rule manager processor is configured to define the anchor sets, such that preference is given to including in the sets, fields having an expected higher distribution of values in the filtered packets relative to other fields.

Optionally, the rui r processor is configured to calculate estimated probabilities of occurrence of false positive matches for rules for various fields and defining the anchor sets, based on the estimated probabilities. Optionally, the rule manager processor is configured to assign the rules to the anchor sets, such that each rule assigned to an anchor set includes equality conditions on the fields of the anchor set, which equality conditions are connected by AND operators.

Optionally, the rule manager processor is configured to assign the rules to the anchor sets, such that each rule assigned to only a single anchor set.

Optionally, the rule manager processor is configured to convert rules including an OR operator into a plurality of rules not including an OR operator, before defining the one or more anchor sets.

Optionally, the rule manager processor is configured to determine for each rule an order of checking of additional conditions of the rule not relating to the fields of the anchor set to which the rule was assigned.

Optionally, the respective filtering units perform the filtering by accessing a hash table and wherein the rule 1011-1066 manager configures the hash table with a rule by indicating in an entry of the hash table corresponding to values required by the rule for the fields of the anchor set, a possible match of the rule. Optionally, the respective filtering units perform the filtering by accessing a TCAM, a TRIE and/or an Extended Grid of Tries with Path Compression.

Optionally, the rule manager processor configures the rules into the respective filtering unit, while the packet filtering processing unit is filtering packets. Optionally, the rule manager processor configures the rules into a memory of the respective filtering unit, while the packet filtering processing unit is filtering packets, without locking the memory unit for access by the packet filtering processing unit. Optionally, the rule manager is adapted to continuously receive rules throughout the operation of the filtering processing units and to configure the rules into the respective filtering unit in real time.

Optionally, the rule manager is adapted to receive rules of different priorities and to configure the rules into their respective filtering unit within time frames determined according to their priorities.

Optionally, the rule manager is adapted to reevaluate the defined anchor sets in real time for at least some of the received rules. Optionally, the rule manager does not reevaluate the defined anchor sets for at least some of the received rules. Optionally, the rule manager is adapted to reevaluate the defined anchor sets periodically, according to the rules currently configured into the filtering units.

Optionally, the rule engine includes a filtering processor implementing the respective filtering units of the anchor sets and wherein the filtering processing unit is configured to pe irm only ;al time tasks iting rate of packet 1011-1066 filtering. Optionally, the rule manager is configured with management information of data structures used by the filtering processor, which are not accessible by the packet filtering processor.

There is further provided in accordance with an embodiment of the present invention, a method of configuring a packet filter for screening data packets for matching rules, including receiving a batch of rules by a rule managerprocessor, selecting, by the rule manager processor, a plurality of rule types each corresponding to a set of fields of the filtered packets, such that each of the rules in the batch belongs to at least one of the rule types in that the rule includes conditions on each of the fields in the set of fields corresponding to the rule type, selecting, by the rule manager, for each rule type a structure of at least one lookup key matching the set of fields corresponding to the rule type, receiving, by the rule manager processor, a plurality of rules, determining, by the rule manager, for each received rule a corresponding rule type, configuring the rules assigned to each rule type into a respective filtering unit, such that the filtering unit begins the filtering of packets for the rules assigned to the rule type with the fields of the rule type.

Optionally, receiving the plurality of rules comprises receiving the rules in the batch of rules. Optionally, determining for each rule a corresponding rule type comprises determining for each rule and rule type a match score and selecting for each rule the rule type with a best matching score.

The present disclosure will be more fully understood from the following detailed description of the embodiments thereof, taken tocfether 'with the drawings in which: 1011-1066 Fig. 1 is a block diagram that schematically illustrates a rule engine, in accordance with an embodiment of the present disclosure; Fig. 2 is a schematic illustration of data elements used by packet filter, in accordance with an embodiment of the disclosure; Fig. 3 is a flowchart of acts performed by a packet filter, in accordance with an embodiment of the disclosure; Fig. 4 is a flowchart of acts performed by rule manager in compiling rules, in accordance with am embodiment of the disclosure; and Fig. 5 illustrates a rule handled by a rule manager, in accordance 'with an embodiment of the disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS An aspect of some embodiments relates to a rule engine configured with at least one hash table which summarizes the rules managed by the engine. The rule engine receives rules and automatically adjusts the at least one hash table in order to relate to added rules and/or in order to remove cancelled rules. The adjustment is optionally performed within less than an hour, less than 10 minutes or even less than 1 minute from receiving the rule. In some embodiments, the adjustment is performed while the rule engine is filtering packets, without stopping the filtering.

Optionally, the rules are grouped into a plurality of rule types and for each rule type the rule engine performs one or more accesses to at least one hash table, to determine whether any of the rules of that type match the packet. In some embodiments, the rule engine selects the rule types responsive to a set of rules provided to the rule engine, possibly even responsive to the set of rules with which the 1011-1066 rule engine is currently configured. By automatically selecting the rule types, the rule engine adapts its operation to the specific rules it is currently handling, thus allowing the user a wide range of possible rules, while not spending resources on checking rule types not currently used.

In operation, the rule engine applies a hash function to one or more groups of fields of received packets and accesses a hash table using a resulting hash key. The results of accessing the hash table determine the rules that might be applied to the packets, possibly after further tests. The use of a hash table which summarizes the rules allows implementation of a rule engine which can handle large numbers of rules, above 10,000, above 100,000 and even above a million rules, with a very fast operation. The processing power required to scan a packet is substantially constant, and does not substantially depend on the number of rules configured to the rule engine . Scanning a packet for matching rules of a specific rule type involves performing the same tasks and requires the same amount of time, regardless of the number of rules In some embodiments, the at least one hash table comprises at least one Bloom filter. The term Bloom filter refers herein to a hash table in which entries provide a negative (e.g., null) answer for at least some accesses to the hash table, without checking a key of the entry to verify that the entry actually corresponds to a key used to access the entry. In some embodiments, the hash table of the Bloom filter does not include keys in the entries at all. In other embodiments, the hash table includes keys in the entries but the key is accessed only if the entry has a value which indicates a possibility existence more rulei corresponding tc ir-„o v « 1011-1066 The hash table is optionally designed to be sufficiently large such that most entries, or even at least 90%, 99% or 99.9% of the entries, return a negati e answer, according to an estimate of the largest number of rules expected to be used. The size of the hash table is naturally, also a function of the available memory of the rule engine.

In some embodiments, for each rule type, only a single hash table access is performed. Alternatively, for one or more of the rule types, possibly even for all the rule types, a series of accesses to one or more hash tables are performed, each access in the sequence being performed only if the previous access in the sequence did not indicate that no corresponding rules are available. Optionally, the stages in the sequence do not receive any information from previous stages, except for the knowledge that the present stage was reached, which indicates that all the conditions leading to the current stage were met. Each entry is optionally configured based on the conditions already met in previous stages.

In some embodiments, the series of accesses includes one or more elimination stages in which non-matching rules are excluded without checking whether the hash key has a match in the accessed hash table entry, and one or more subsequent affirmation stages. In the affirmation stages, the hash key used to accessed the table is compared to a corresponding key in the entry and the entry is used only if the keys match. Optionally, the entry received in one of the accesses of the sequence indicates the packet fields to be used in performing the next access of the sequence. The indication may be provided in a Bloom filter hash table and/or in a hash table in which key matching is affirmed before providing a result. 1011-1066 In some embodiments, a key used in accessing the hash table is formed of one or more fields of the packet, possibly at least three fields or even at least five fields of the packet. Alternatively or additionally, a key used in accessing the hash table may include the result of applying one or more operators to a field or fields of the packet. Using this alternative, rules corresponding, for example, to inequalities, value ranges and/or a plurality of conditions in the alternative, may be implemented. Optionally, the key used in accessing the hash table is configurable and is changed according to the set of rules currently handled by the rule engine. In some embodiments, different keys are used in accessing the hash table in different stages of checking a packet.

Optionally, the rule engine is formed of two separate components. A first component manages the rules and prepares them for implementation and a second component performs real time filtering with minimal writing or no writing at all. The two components optionally share a memory unit in a manner allowing them both to access the memory without locking. Optionally, an add-and-delete update method is used to update a hash table in the memory. In implementing a rule change, the first component optionally puts entries required for a new version of the rule in memory locations not accessible by the second component. When everything is ready for utilization of the new version of the rule, a pointer directing to an old version of the rule is changed so that it points to the new version of the rule and then the data entries of the old version are cancelled.

An aspect of some embodiments relates to a rule engine configured to check received packets against all rules with 1011-1066 which it is configured and to indicate all the rules that the packet matches and not just one of the matching rules.

System overview Fig. 1 is a block diagram that schematically illustrates a rule engine 20, in accordance with an embodiment of the present disclosure. Rule engine 20 includes a rule manager 22 and a packet filter 24. Rule manager 22 receives rules through an input interface 32, compiles the rules into an internal format of rule engine 20 and provides the compiled rules through an output interface 34 to packet filter 24. The compiled rules are optionally stored in a memory 26 shared by rule manager 22 and packet filter 24, In some embodiments, rule manager 22 directly stores the compiled rules in memory 26.

Packet filter 24 receives packets through one or more input interfaces 42 and determines for each packet whether there is a rule which pertains to the packet. Optionally, packet filter 24 is configured to check packets for matching rules by performing one or more lookups in a hash table. Optionally, each packet is checked for one or more rule types corresponding to different sets of fields of the packet, forming a lookup key into the hash table.

In some embodiments, rule manager 22 analyzes a set of rules with which rule engine 20 is configured, to determine a set of rule types (and corresponding lookup keys) for which each packet or stream of packets is checked. In addition, rule manager 22 receives rules to be configured into packet filter 24, determines a rule type that the rule will be associated with and accordingly configures the hash table and other data structures in memory 26, as detailed hereinbelow.

As a short example, filter manager 22 recei es the following set of rule conditions: 1011-1066 Source IP = 1.1.1.67 & VLAN = 5 Source IP = 2.2.1.67 & Best Port 25 Destinat ion IP = 3.3,3.3 & VLAN 64 and source port = 80 Source IP = 2.2.1.2 & Dest Port 26 and protocol = 62 Dest port = 56 & VLAN = 72 Filter manager 22 selects the fo lowing two rule types 1) VLAN 2 ) Source IP & Dest Port which cover the entire set of rules, as each rule can be assigned to at least one of the rule types.

Filter manager then assigns each rule to a rule type, configures the hash table with the portions of the rule covered by the rule type and configures memory 2 6 with the additional conditions that need to be checked for those packets that match the portions of the rule matched by the rule type. For example, the first rule is assigned to the rule type of VLAN. The comparison of the VLAN field of the packet will be compared to the value "5" in a hash table lookup, as discussed hereinbelow. Packets for which a match was found will be forwarded for further handling in which Source-IP will be compared to the value 1.1.1.67.

Packets for 'which rules are found are handled by packet filter 24 according to the rules, while packets for which rules were not found are handled according to a default, for example are forwarded through a default output interface 44 , Packet filter 24 is generally designed to handle a large number of packets, for example more than 1 million packets a second, more than 10 million packets a second, or even more than 15 million packets a second.

Rule engine 20 may be used for various tasks, such as routing, classification (e.g., as described in US patent publication 2004/0095936 to 0'Neill et al .) and network 1011-1066 security (e.g., as part of a firewall). For example, the rules may identify packets that are to be copied and forwarded through an additional output interface 46, in addition to their being forwarded through output interface 44.

Packet filter 24 optionally performs all real time tasks of rule engine 20, such that rule manager 22 has no real time tasks and cam spend sufficient time and memory resources in order to achieve an optimal compilation of the rules. In some embodiments, rule manager 22 may perform the compilation with the aid of a human. In some embodiments, rule manager 22 manages hash table and/or other data structure management information, such as information on the rule or rules corresponding to each hash table entry, in a memory unit separate from memory 26, reducing the amount of information that needs to be stored in memory 26.

Rule manager 22 and/or packet filter 24 are implemented in hardware, software, firmware or combinations thereof. Rule manager 22 typically comprises one or more processors 50, memory units and required software, but may also comprise firmware or hardware. In some embodiments, rule manager 22 is implemented by a general purpose computer. The software may be downloaded to the processor in electronic form, over a network, for example. Alternatively or additionally, the software may be held on tangible, non-transitory storage media, such as optical, magnetic, or electronic memory media. Packet filter 24 is optionally implemented by a network processor 48, although other types of processors, hardware, firmware or software based may be used. Packet filter 24 optionally includes one or more cache units 28.

Packet filter Fig. lematic iliustration of data elements rseci by packei in acsordance wit! an embodiment of the i3 1011-1066 disclosure. Packet filter 24 manages a rule type list 202 which lists types of rules to which packets (represented in Fig. 2 by a packet 220} need to be compared. For each rule type, rule engine 20 manages a type-record 204 which defines the rule type and indicates the specific steps to be used for the rule type, as discussed herein below. Each type-record 204 optionally also refers to a key structure 206, which defines the fields 208 of packets 220, to be used in constructing a key 210 used in determining whether the packet meets any of the rules of the rule type. A first hash function Hashl() is optionally applied to key 210 to generate a hash key 212. A second hash function Hash2() is applied to hash key 212 in accessing a record in a hash table 214. The records of hash table 214 indicate whether packets of the specific key provided have a corresponding rule and what further handling should be applied to the packet.

Alternatively or additionally to accessing hash table 214, hash key 212 and the second hash function Hash2() are used to access a record in a bloom filter 216, which indicates whether packets corresponding to the key 210 may have a corresponding rule. Further elements shown in Fig. 2 are discussed hereinbelow along with the discussion of the flowchart of Fig. 3.

Fig. 3 is a flowchart of acts performed by packet filter 24, in accordance with an embodiment of the disclosure. For each packet 220 received (302), packet filter 24 optionally determines (330) whether the packet belongs to a packetconnection for which a rule determination was recently performed. The determination (330) is optionally performed by searching a table 230 of recently handled connections. In some embodiments, the determination is performed using any of the methods described in US patent application 13/358,482, titled 1011-1066 System and Method for Flow Table Management, which is assigned to the assignee of the present application and was filed Janu ary 25, 2012.

If (330) the packet does not belong to a recently handled connection, packet filter 24 performs for each rule type (304A, 304B, 304C) with which it is configured, one or more (318) lookup stages (308) including preparing (306) a lookup key 210 and accessing (310) a hash table 214 and/or a Bloom filter 216. The number of lookup stages (308) and their details, such as which table is to be accessed in each stage (308), are optionally set by the contents of the corresponding type-record 204, The stages may be predetermined or may be dependent on the results of previous stages as discussed nereinbelow .

In lookup stages (308) 'which are last in a series or are a single stage for a rule type, the table entry accessed, if (309) there is a possibility of a matching rule, indicates the further handling (312) of the packet, for example by a pointer to a rule record 272 in a rule database 270. The further handling (312) in the rule record 272 may indicate the rule to be applied (314) to the packet and/or may indicate further tests (316) that need to be applied to the packet to determine whether it matches a rule. Lookup stages (308) followed by an additional stage optionally determine whether to continue to the next stage. In some embodiments, the result of an earlier stage in the series may also affect the structure of the key 210 used in the following stage, in addition to or instead of the kev structure 206.

Lookup stages ((308) are performed for all the rule types with which rule engine 20 is configured and if (309) no match is found, the packet is ham led (322) according to a default handling . 1011-1066 In some embodiments, after handling (312) a packet according to instructions in a corresponding table entry, the packet is not tested for compliance with other rules, as rule manager 22 is configured to have each packet correspond to at most a single rule. In other embodiments, packet filter 24 checks for each packet all the types of rules, even if a match is found. In still other embodiments, the instructions in the corresponding table entry indicate whether to test the packet for matching other types of rules.

If (330) a packet belongs to a recently handled packet connection, the packet is optionally handled (332) according to the results of the previous determination, as indicated in table 230, without performing lookup stages (308) . In some embodiments, packet filter 24 checks before the handling (332), in a rematch act (335), whether recent changes in the rules configured in packet filter 24 may require applying to the packet a different handling than that prescribed by table 230. Rematch act (335) optionally determines the rule types for which changes occurred, if any, and lookup stages (308) are performed only for those rule types. The recent changes are optionally also checked to determine whether a rule indicated for the packet connection by table 230 changed, in which case the handling (332) of that rule is not applied based on the entry in table 230. If the rule needs to be applied to the packet despite the change, it will be identified in the lookup stage 308 corresponding to its rule type .

In some embodiments, table 230 indicates f each racket connection recently handled, the most recent time at which the rules corresponding to the packet connection were determined . A rule update list 232 optionally lists rule types that were updated and their corresponding update times. Packet filter 24 1011-1066 optionally checks, in rematch act (335), for packets handled (332) according to listings in table 230, whether new rules were added since the rules corresponding to the packet connection were most recently determined. If there are one or more rule types which were updated since the most recent rule determination for the connection, packet filter 24 performs lookup stages (308) for the packet for these rule types. Table 230 is optionally periodically reviewed to remove old entries.

In some embodiments, packet filter 24 manages rul 10z oro'erec ccording to the time who typ was last updated with a new rule. In handling received packets, packet filter 24 determines from table 230 the time at which the handling for the connection was determined. Packet filter 24 then goes through ordered list 202 in the order from the most recently changed rule type and for each rule type record 204 determines whether there was a change after the handling of the connection was most recently determined. If there was a change, a lookup stage is performed for that rule type and the determination of whether there was a change after the handling of the connection was most recently determined is repeated for the next rule type in order until a rule type changed before the handling of the session of the packet was most recently determined, at which time the handling of the packet is terminated.

Optionally, only some types of rules are taken into consideration in setting the time at which a rule type was updated. In some embodiments, rule engine 20 is configured to handle rules of different types, such as user rules which are generally configured by humar users and have a low update rate and dynamic rules which are generated automatically and have a high update rate . Optionally, only user rules are taken into consideration in setting the most recent update of a rule 1011-1066 type. Dynamic rules generally relate to recently created connections and the handling of previously established connections is not expected to change due to an update of a dynamic rule.

In some embodiments, a special rule type is defined for dynamic rules without relation to the user rules. That is, different rule types are defined for dynamic rules and for user rules and even if a dynamic rule type and a user rule type relate to the same fields they are not merged. While these embodiments may increase the number of rule types for which each packet need to be scanned, they simplify the operation of rule engine 20, Alternatively or additionally to checking for rules updated since the determination performed for the connection of the packet, packet filter 24 checks in the rematch act 335 for packets belonging to connections established before a predetermined time. For example, packet filter 24 may identify packets whose connection was established before about a second and lookup stages 308 may be performed for the packet in order to confirm stability of the previous determination for the connection. The lookup stages performed in confirming stability may relate to all rule types or may be limited only to rule types including dynamic rules. In some embodiments, table 230 lists for each connection whether it belongs to a dynamic connection and the lookup stages performed for stability confirmation are performed only for connections identified as dynamic. The use of separate rule types for dynamic rules allows performing the stability confirmation lookup only for rule types relating to dynamic rules.

Alternatively to using rule update list 232, whenever a rule is updated table 230 is emptied. Further alternatively, whenever a rule is updated, table 230 is reviewed and packet 1011-1066 connections matching the updated rule are removed from table 230.

In some embodiments of the invention, table 230 additionally indicates for each connection a time at which the entry for the connection is to expire, regardless of whether any rules changed. Each time the table is accessed, packet filter 24 verifies that the accessed entry of the table is still valid. Optionally, the expire time is determined from rule record 272 and/or from an entry in hash table 214. For example, when configuring rules, an expiry time may be associated with the rule, according to an expected life of the rule.

Lookup stages In some embodiments, lookup stages 308 for each rule type include at least one preliminary lookup stage in which access is made to a Bloom filter 216 and at least one later stage in which access is made to a hash table 214, The Bloom filter 216 is used for coarse filtering in order to speedily eliminate packets which clearly do not match any rule of the rule type and the hash table 214 is used for a fine tuning. It is noted that further fine tuning may be performed in the further tests (316) indicated by the rule record 272.

This option is optionally used when the percentage of packets expected to meet rules is very low, such that the Bloom filter 216 is expected to reject a large percentage of packets, optionally at least 90%, 99% or even at least 99.9%. In some embodiments, only a single lookup' stage that accesses a Bloom filter 216 is used, and the rest of the lookup stages 308 access hash tables. In other embodiments, a plurality of accesses to a single Bloom filter 216 or to a plurality of different B1oom fi11ers, are used. For examp1e, for a specific rule type, rule manager 22 may determine that three fields 208 1011-1066 should be used in an elimination stage. The three fields 208 may be checked in a single elimination stage including a single access to Bloom filter 216 or in three separate accesses to the B1oom fi11er 216.

Optionally, each lookup stage 308 in a sequence of stages for a single rule type uses a different key 210 formed of different fields. In some embodiments, the keys 210 are totally different and do not include any common fields. Alternatively, two or more stages 308 use different keys 210, but the keys 210 have one or more fields in common. Further alternatively, two consecutive lookup stages 308, a first stage 308 accessing a Bloom filter 216 and a second stage 308 accessing a hash table 214, use the same key 210.

In some embodiments, a single access to a hash table 214 is performed in two steps. In a first step, the accessed entry is checked in a manner similar to that performed in a Bloom filter, without checking whether the accessed entry matches the hash key 212 used in accessing the entry. Only if it is determined that the entry corresponds to one or more rules is the key checked for a match. Thus, the advantage of elimination by a Bloom filter is achieved without adding an additional hash lookup stage (308). In one embodiment, only a single lookup stage 308 of this type is used for each rule type and a Bloom filter 216 separate from hash table 214 is not used at all. This reduces the memory requirements of packet filter 24 and/or the number of memory accesses performed by packet filter 24.

In some embodiments, all rule types involve the same number of lookup stages (308), for simplicity of operation of packet filter 24, In accordance with one embodiment, only a single lookup stage is used for all rule types and all additional tests are performed in the further handling (312) 1011-1066 indicated in the rule record 272. Alternatively, different rule types may correspond to different numbers of lookup stages (308) so as to better accommodate the rules corresponding to the rule type.

Rule types and keys The rule types are optionally defined according to the fields 208 of the packet to which they pertain. Each rule type pertains to a specific set of fields of the packets and these fields are included in the key 210 prepared for the rule type or keys 210 when the rule type corresponds to a plurality of lookup stages (308), For example, a first rule type may include fields which depend on IP 5 tuples (e.g., source and destination address, source and destination port and protocol), a second rule type depends on VLAN and destination address fields and a third rule type depends on a URL field. In some embodiments, lookup key 210 is formed only of one or more fields 208 of the packet 220, In other embodiments, the lookup' key 210 of one or more rule types and/or one or more lookup stages (308) includes one or more result fields 218 in addition to, or instead of fields 208 taken directly from the packets 220. Each result field 218 includes the result of applying one or more operators 250 to one or more fields 208. The use of result fields 218 allows checking inequalities and being within ranges, as well as complex conditions such as a plurality of equalities and/or inequalities connected by OR or XOR operators.

Alternatively or additionally, lookup key 210 includes one or more external fields, such as a time related field (e.g., day of week, day of month, time of day, indication of day or night).The rule types generally relate to layer—3 or layer-4 fields of the packets, but may alternatively or additionally relate to layer-2 fields and/or to higher layer 1011-1066 fields. The fields used may include, for example, source and destination addresses and ports, application classifications of the session of the packet, '/LANs, URLs, subnets of source and destination and tunneling parameters.

Typica1 ru1es may inc1ude, for examp1e: 1) Source IP or Destination is equal X 2) Destination port 25 (SMTP) 3) Source IP = X and Destination port = 80 (Neb traffic of specific person) 4) Source IP = X and PORT = 80 and URL = g ail.com 5) Source IP = X and application is NOT a point to point (P2P) application 6) Source or destination subnet = X.Y.Z.00 and PORT = 25 (source or destination) and VLAN = Y In some embodiments, the key 210 has the same length for all the rule types, for example at least ;ast tes or even an least 32 bytes. In one embodiment, the key 210 has a length of 32 bytes. Alternatively, the key 210 has a different length for each rule type, depending on the specifics of the rule type.

It is noted that rule engine 20 may be configured with rules ting to substa trally any of the fields >f the packets. Optionally, rule engine is configured with rules pertaining to more than 20 different packet fields, more than 30 different fields, more than 40 different fields or even more than 50 different fields. Generally, each rule relates to no more than 5 or 10 fields, but in some cases a single rule may pertain to more than 20 or even more than 30 different fields .

Hash table The hash function Hashl() used in generating hash key 212 may be anv suitable hash function known in the art, such as a 1011-1066 CRC-32 hash function or a function which breaks key 210 into segments (e.g., byte segments) and combines the segments using a XOR operator and shift operations.

In an example embodiment, the first hash function Hashl() converts key 210 into a number including between 6-18 bytes, for example 12 bytes. Optionally, hash key 212 includes in addition to the hashed value of key 210, one or more fields which identify the rule type and/or the fields of the key structure of the rule type which are currently used. For example, hash key 212 may include 24 bytes: 12 bytes resulting from a hashing of key 210, 8 bytes which identify the fields being used and another 4 bytes which identify the rule type. A second hash function Hash2() is optionally used in accessing hash table 214 or Bloom filter 216 with hash key 212. The second hash function Hash2() results in a table-access number having a number of bits matching the size of the hash table, for example about 24 bits.

Optionally, instead of holding key 210 in the hash table to verify that the record reached in a lookup is the intended record, key 212 is held in the record. Using key 212 has the advantage that it is generally smaller than key 210 and requires less memory and also that it has a fixed size. In some embodiments, key 212 is large enough such that the chances of a mistaken match of the key 212 of a packet to the key 212 of a record of the hash table are negligible. Alternatively or additionally, a further match verification is performed in the further tests (316) after reaching rule record 272.

In some embodiments, the second hash key used to access bloom filter 216 is calculated directly from key 210, without calculating the first hash key 212. Optionally, only if accessing hash table 214 along with a comparison to a hash key 1011-1066 of the accessed record in the table is required, first hash key 212 is calculated.

In some embodiments, hash table 214 comprises an array 280 of pointers. Each entry 282 in array 280 is optionally a four byte pointer field which either has a null value, indicating there are no corresponding rules or includes a pointer to a linked list 286. The term pointer is used herein to refer to any data unit which indicates a memory location and includes absolute address values and relative offsets.

The linked list 286 includes one or more records 284 which each include a corresponding hash key for verification that the packet matches the record, and further handling instructions . The further handling instructions optionally include a pointer to a rule record 272 or information directing to a next lookup stage 308. Optionally, the number of entries 282 in array 280 of hash table 214 is relatively large, such that at least 90%, 95% or even 99% of the entries 282 are expected to have null values.

Bloom filter Bloom filter 216 optionally includes a Boolean table which indicates the result of the lookup stage 308. If the Boolean table location is negative (e.g., null), the possibility of a match is excluded. If, on the other hand, a match is found, packet filter 24 proceeds to the next lookup stage (308).

Each entry of the Boolean table is optionally represented by a small number of bits (e.g., a single bit), such that the entire Boolean table can be held in a small memory area and the number of entries in Bloom filter 216 may be made large, increasing the percentage of rules which are excluded in an early stage. In addition, using a small memory area for the Bloom filter 216 may increase the chances of accessed entries 1011-1066 being found in cache 28, thus reducing the number of memory accesses. In some embodiments, packet filter 24 is configured to keep the entries of Bloom filter 216 permanently in cache memory 28 or at least to give priority in using cache 28 to the Bloom filter. This could substantially reduce the number of memory accesses, which are generally time-consuming, required by packet filter 24.

Alternatively, to a Boolean table, Bloom filter 216 may include in each entry a multi possibility value, for example a four bit or 8 bit integer, which includes additional information for cases in which the packet is not excluded. For example, the value 0 may be used to indicate that the packet does not have any matching rules, while the other values indicate the fields 208 to be used in determining whether the packet has a possible matching rule. In some embodiments, the entry value in Bloom filter 216 indicates a sub-group of the fields indicated in the key structure 206 corresponding to the rule type, to be used in the next lookup stage 308. Alternatively or additionally, the value from the Bloom filter 216 indicates fields 208 to be used, beyond those included in the key structure. Optionally, in these embodiments, instead of using the key structure 206 in generating the key 210 for the next lookup stage 308, the key 210 is generated based on the value in the entry of the Bloom filter 216.

For example, the key 210 used in accessing the Bloom filter 216 may be based only on the destination IP address. The entry in the Bloom filter 216 indicates which additional fields 208 are to be used in the key 210 used in the next lookup stage 308. For example, the code "01" may indicate using also the destination port, the code "02" using also the VLAN, the code "03" using both the destination port and the VLAN, the code "04" using an encapsulated IP address, etc. 1011-1066 The number of entries in the table of the Bloom filter 216 is optionally selected so that at least a predetermined percentage of the entries (e.g., at least 80%, at least 95% or even at least 99.9%), are null. Alternatively or additionally, the number of entries in the table of the Bloom filter 216 is optionally selected such that at least a predetermined percentage (e.g., at least 90%, at least 95% or even at least 99.9%) of the received packets that do not correspond to rules are excluded by the lookup in the Bloom filter 216.

In some embodiments, the Bloom filter 216 is accessed using only a single hash function (i.e., k=l). Alternatively, the bloom filter 216 is accessed with a pluralit lash functions, for example at least five or even at least 10. The number k of hash functions used is optionally selected based on the configuration parameters of packet filter 24, such as the number of entries in Bloom filter 216 and the time required for each access to the Bloom filter 216. If the accesses to the Bloom filter 216 are relatively slow, a low value of k, optionally k=l, is preferred. If, however, the accesses are relatively fast and the number of entries is relatively small, a higher value of k is selected.

In another embodiment, the values in the entries of the Bloom filter comprise a number of bits, but in accessing an entry only a sub-group of the bits, possibly only a single bit, is used. This allows, for example, easily changing the results from Bloom filter 216 over time and/or other current conditions. Optionally, in accessing Bloom filter 216, the exact bit or bits to be looked at is determined according to the time of day and/or according to a time counter 260 of packet filter 24. For example, each entry may have eight bits, seven bits corresponding to the days of the week and an eighth bit for nicfht hours. This allows setting different rules for o 1011-1066 different days of the week and/or hours of the day. In another example, different bits are used to represent different hours of the day.

It is noted that in some embodiments, packet filter 24 sequentially goes through lookup stages (308) for the different rule types, as illustrated by Fig. 3. In other embodiments, packet filter 24 may handle a plurality of rule types in parallel, for example managing a plurality of contexts by a single processor or by a plurality of processors which operate in parallel.

Rule compilation Fig. 4 is a flowchart of acts performed by rule manager 22 in compiling rules, in accordance with an embodiment of the disclosure .

Reference is also made to Fig. 5, which illustrates a rule 500, handled by rule manager 22, in accordance with an embodiment of the disclosure. Each rule 500 comprises a condition 502 and a task 504 to be performed on packets meeting the condition 502. The condition 502 comprises one or more terms connected by the logical AND operator 508. The terms may be simple terms 510 (marked also 510Ά, 510B) or complex terms 512 formed of a plurality of simple terms 510 connected by logical terms such as OR or XOR. Each simple term 510 may be an equality term 510A or a non-equality term 510B, which may include, for example, an inequality (e.g., >, , <=, >=) , range, mask, or searching for a pattern in a field 208.

In the method of Fig. 4, rule manager 22 receives (402) a set of rules which are to be enforced by rule engine 20 and accordingly defines (404) a set of rule types. The definition includes for each rule type, the fields 208 and operators 250 used in the key (or keys) 210 of the rule type. 1011-1066 Thereafter, for each rule, rule manager 22 selects (406) a rule type which best matches the rule. Rule manager 22 determines (407) based on the matching rule type, which terms (510, 512) of the rule are handled, at least partially, by lookup stages 308 and which are handled entirely by further tests 316.

For the terms handled by lookup stages 308, rule manager 22 optionally determines (408) the entries of hash table 214 and/or Bloom filter 216 which will be accessed by packet filter 24 in the process of handling the selected rule type, for packets that may match the rule. For each determined entry, rule manager 22 determines (410) the content that needs to be placed in the entry (e.g., entry 282 and record 284 of hash table 214) in order to direct packet filter 24 to a next lookup stage 308 or to a rule record 272 generated for the rule in rule database 270. Depending on the rule type, the entry content may merely include a match indication or may indicate fields 208 and/or operators 250 to be used in a next lookup stage 308. For the last (or only) lookup stage 308 in a sequence, the entry indicates the rule record 272 corresponding to the rule.

The required content of the rule record 272 corresponding to the rule is determined (412). The rule record 272 includes any further tests (316) that need to be performed and the handling that needs to be applied (314) to packets matching the rule. The further tests 316 include terms not handled by lookup stages 308 and/or fine tuning for terms handled only approximately by a lookup stage 308. For example, if a rule requires checking that X >= 52 and the lookup stage uses an operator 250 of X >= 50, the further tests include a verification that X >= 52. The complexity of further tests 316 is generally not an issue, since the number of packets that 1011-1066 reach the stage of the further tests is generally very small, such that the processing time spent on these packets is cons1dered neg11g1b1e.

In some embodiments, the further tests 316 prescribed by one or more rule records 272 direct back to one or more lookup' stages 308. Thus, a complex test may be performed between two lookup stages 308. For example, if many rules having one or more common attributes compare a field 208 to different values the rules having one or more attributes and direct them to a rule record 272 which directs the complex test. The rule record 272 in suei case does not include handling instructions for all the rule;; matching its condition. Generally, packets referred back to perform an additional lookup stage 308 will return to a different rule record 272 where the handling instructions are located.

After rule manager 22 determines required changes in the database of packet filter 24, rule manager 22 updates (414) the database in memory 26.

In some embodiments, the method of Fig. 4, including the defining of rule types (404), is performed periodically. When rule manager 22 is configured with new rules, the rules are accumulated until a next time the method of Fig. 4 is performed. Alternatively, between periodic defining of rule types, rule manager 22 receives rules and immediately, or periodically at relatively short intervals, updates the database for the rule, based on the current rule tpes. In still other embodiments, the defining (404) of rule types is performed only once, at installation. Another possibility used in some embodiments is that during selection (406) of rule types best matching newly received rules, rule manager 22 generally uses the current rule types, but may initiate 1011-1066 changing of one or more rule types when such a change is determined to be useful. For example, rule manager 22 determines cases in which the difference between a new rule and the closes rule type is too large and 111 tes changing of the rule types is such cases. Another possibility is that the definitions of the rule types are revisited whenever a new rule that does n match any of the current rule types is received.

The changing of the rule types may be limited to defining a new rule type or changing a single existing rule type so that it matches the rule. It is noted that when a rule type is changed, all the rules matching that rule type are optionally revisited to make sure they match the changed definition of the rule type. Alternatively, the changing of the rule types includes initiating a complete process of selection from scratch of the rule types to be used and accordingly a redefinition of the compilations of the rules.

Defining rule type set Referring in detail to defining (404) a set of rule types, in some embodiments, a predetermined set of rule types is used. Alternatively, the set of rule types is generated responsively to the received set of rules. In one embodiment, the defined set of rule types includes a separate rule type for each distinct of fields related by one or more rules . Alternatively, in order to reduce the number of rule types, one or more rule types in the defined set includes a limited set of fields serving as a common denominator of a plurality of different sets of fields related to by rules in the rule set. For example, for a rule set including one or more rules relating only to an IP source field, one or more rules relating to the IP source field and port source field and one or more rules relating to the IP source field and a 1011-1066 VLAN field, a single rule type may be defined for the IP source, which rule type is intended to cover rules of any set of fields including the IP source field.

Optionally, rule manager 22 determines groups of conditions which are shared by large sub-sets of rules in the rule set and defines rule types for these groups of conditions . In some embodiments, rule manager 22 estimates, for each condition term 510 or 512, the percentage of packets that can be filtered out by the term. Particularly, rule manager 22 optionally estimates the probability of occurrence of false positive matches for rules, for various fields and defines rule sets, which are expected to achieve lowest false positive match rates . In some embodiments, fields having an expected higher distribution of values are preferred in being included in the rule sets.

The estimation is optionally based on pre- configured statistics for the various possible fields related to in the terms. Alternatively, the estimation is based on a current set of rules being considered. Optionally, the estimation is based on the particular values to which the _ds are comp; _n the terms. For example, low port values are generally very common and therefore their filtering value is low, while high port values are relatively rare and their filtering value is nigh. The estimation of the percentage of packets filtered by terms relating to a port field is optionally based on the number or percentage of rules that relate to high port values.

Rule types are optionally defined for sets of terms that are expected to filter out a large percentage of the packets. It is noted that equality terms 510A can separately filter out each of the possible values of a field 208. In contrast, other terms, which involve operators 250, perform the filtering only for a single value included in the operator. For example, an 1011-1066 inequality term 510B relating to a field 208 of a value X, relates to a specific value, such as X > 5. In order to compare to another value, e.g., X > 20, using the hash table lookup, another operator 250 and corresponding field in key 210 are required.

The rule types are optionally determined in a manner which minimizes the percentage of packets being false positives, that is packets that are directed in the lookup stages (308) to rule database 270 but do not correspond to a rule. This is optionally achieved by having the rule types cover the rules relatively closely.

On the other hand, the rule types are optionally determined in a manner which minimizes the processing time that needs to be spent on filtering each packet, for example by minimizing the number of operators 250, the number of memory accesses performed by the rule type and/or the number of ru1e types.

Optionally, the generation of the rule types takes into account the possible values of the different fields 208, for example preferring rules relating to fields having a larger number of possible values and/or a relatively low frequency for any specific value. For example, rules relating to an IP address field are optionally preferred over rules relating to TCP port, as many packets have the same port value. Optionally, rule manager is configured with the possible values of each field 208 and/or with estimated frequencies of the va1ues of the fie1ds.

In some embodiments, rule manager 22 defines rule types only based on equality terms 510A, and all other terms are handled in further tests 316, In other embodiments, when a very popular non-equality term 510B or 512, which can add substantially to reducing the number of false positives, is 1011-1066 identified, an operator 250 for the term is defined and the term is included in one or more of the rule types.

Rule types optionally include only a single operator 250 for a field 208. Alternatively, a rule type may include a plurality of operators for a single field 208, for example comparing the field to different values. In some embodiments, the plurality of operators may even include contradicting conditions. For example, a rule type may include a first operator 250 for X < 20 and a second operator 250 for X > 30, for the same field 208 X. The inclusion of the plurality of operators in the same rule type, reduces the number of rule types required.

Optionally, when an operator 250 is defined for an inequality for a field 208 that is compared in different rules to different values, a single value or several values are selected to represent a larger number of values in the rules. For example, if a field 208 (X} is compared with a "smaller than or equal” operator (<=) in different rules to the values 55, 57, 59, 60 and 89, two operators 250 may be defined: X <~ 60 and X <== 89. Rules requiring a comparison X <= 57 are translated to the X <= 60 operator 250 and the verification that X is smaller than 57 is performed in the corresponding record 272 in the further tests (316).

In some embodiments, the defining of rule types begins with a beginning set of rule types, for example including between 1-4 rule types. The rule types in the beginning set may be predetermined, or may be selected based on the current set of rules of rule engine 20. Optionally, the rule types in the beginning set are single term rule types, for the most prevailing terms. Alternatively or additionally, the beginning set of rule types includes a minimal set. of single term rule types, which correspond to all rules currently handled by rule 1011-1066 engine 20, such that each rule relates to at least one field of the set of single term rule types. The minimal set is optionally selected by selecting the fields 208 not yet included in the set, which relate to the largest set of rules not yet corresponding to a rule type in the set. Alternatively or additionally, a plurality of sets are contemplated and the set with the smallest number of rule types is selected.

Optionally, after defining the beginning set, an iterative process is performed in which a current set, initialized with the beginning set, is updated. Optionally, in each iteration, each of the rules is classified to one of the rule types in the current set and a difference between the rule and the rule type is evaluated. For rule types having a large accumulated difference, rule manager 22 determines an additional rule type which could remove a substantial number of rules from the rule type with the large accumulated difference, in a manner which minimizes the sum of the accumulated difference of both the old and new rule types. The new rule type is added to the current set if the cost of adding the additional rule type in terms of processing power of packet filter 24 is found to be smaller than the cost of the accumulated difference. The new rule type may be generated, for example, by adding a term to the old rule type and/or by removing a term. In other embodiments, a new rule type is defined when a rule is found to be farther than its closest rule type by at least a predetermined extent.

Optionally, the number of rule types is limited to a maximal number, for example 10 or 15. Alternatively, the number of rule types is not limited. Optionally, the addition of each rule type is required to achieve a total gain in processing resources. In determining whether to add a rule type, rule manager 22 optionally estimates the percentage of 1011-1066 packets which will be rejected earlier (e.g., before reaching a record 272) due to the addition of the rule type.

While in some embodiments the selection of rule types is entirely automatic, in other embodiments the selection is performed partially or entirely by a human.

Rule manager 22 optionally determines the number of lookup stages to be used for each rule type. Optionally, rule manager 22 estimates the processing power required for each number of lookup stages and selects the most efficient number of stages. In embodiments in which all the rule types are required to have the same number of lookup stages, a minimization over all the rule types is optionally performed. In embodiments in 'which a maximal number of allowed stages is defined, only sequences including the allowed number of stages is considered.

Handling rules As to selecting (406) a rule type which best matches the rule, rule manager 22 optionally compares the rule to each rule type and determines the rule type for which the percentage of false positives is lowest. Optionally, a rule is not assigned to a rule type that relates to a field 208, not related to by the rule. Alternatively, if the possible number of values of the field 208 included in the rule type but not in the rule is relatively small and the advantage from matching the rule type for the rule is relatively high, the rule type is used. For example, when results of operators 250 are Boolean, a rule type may relate to several operators 250 although rules matching the rule type only relate to a sub group, possibly only one or none, of the operator results.

As to determining (10) the content that needs to be placed in the entry, when the rule type has only a single lookup stage 308, the entry directs to the corresponding 1011-1066 record 272. When a plurality of lookup stages are used, the non-last stages optionally return a Boolean answer as to whether a match is found. Alternatively, the entry of one or more lookup stages 308 may indicate the fields 208 to be used in the key 210 of the next stage. A key structure 206 is optionally defined for each lookup stage of the rule type.

As an example, if rule manager 22 is configured with several rules relating to source IP address (IPsrc)? source port. (Port src) and possibly another field, the rule type may 10 correspond to two lookup stages 308. In a first 3.Cføf cl lookup is performed based on IPsrc anc4 Portsrc, and rrd results of the lookup direct to a rule record 272, direct to another lookup stage or provide a negative answer, depending on the values of IPsrc and Portsrc . For example, given the fo110w ing ru1e condi11ons: 89 d) lPsrc = 444 AND Portsrc= 111 AND Portd£._ Z'J the hash table entry corresponding to the IPsrc anc4 Portsrc values of a, would lead to a further lookup based on IP st, and the hash table entry corresponding to the IPsrc anc4 Portsrc values of b, would lead to a further lookup based on VLAN. The hash table entry corresponding to the IPq, and o Portsrc values of c and d, would lead directly to respective corresponding rule records 272 and other values would provide a negative answer indicating there are no matching rules. It is noted that the rule record 272 of c merely states the handling of the packet, as any packet reaching this record 272 30 matches the rule. In contrast, the rule record 272 of d 1011-1066 includes a verification instruction, which verifies that the packet meets Portgst < 291 before moving on to the handling of the ru1e record 272, As to determining (412) rule record 272, when more than one rule is directed to the same rule record 272, the further tests (316) prescribed by the rule determine which of the rules is met, if any. In such case, the portion of the record 272 indicating the handling that needs to be applied (314) to packets matching the rule may include several different handling methods, for the different matching rules.

Database u date The operation of rule manager 22 provides for each new or changed rule, required updates to hash table 214 and/or Bloom filter 216. In addition, for at least some rules, a record 272 needs to be generated or updated. Also, changes to the rule types require changing type-records 204 and/or key structures 206.

The updating (414) for a batch of a plurality of new rules received together may be performed together, after determining the required changes for all the new rules. Alternatively, after determining the required updates for a rule, the updates are performed immediately and then the updates required for a following new rule are determined.

In updating (414) the database of packet filter 24 for rules being added, rule manager 22 optionally performs the updating in a manner 'which allows packet filter 24 to continue operating during the updating. Optionally, packet filter 24 does not 'write to hash table 214 and/or Biloom filter 216, such that a write collision between rule manager 22 and packet filter 24 is not possible.

In adding a new rule, rule manager 22 optionally first generates record 272 of the rule and records 284 in hash table 1011-1066 214. These data elements are not yet accessible by packet filter 24 as there is no pointer entry 282 leading to them. If a rule hash table record 284 for the rule already exists (e.g., due to a similar rule previously configured into packet filter 24), rule manager 22 optionally generates a new record 284 and then changes the pointer entry 282 to lead to the new record 284 and deletes the old record 284. Similarly, in some embodiments, any time a change to a record 272 is required, rule manager 22 prepares a new record in a memory location not accessible by packet filter 24 because it is not on any linked list 286. Then, in an atomic operation, e.g., a short memory write, the new record 272 is linked to from an entry of Hash table 214, in a process that does not allow collisions.

Changing to hash table entries 284 and entries of Bloom filter 216 are optionally performed in a manner which reduces the chances of a collision. Optionally, the entries are updated from the last lookup stage 308 to the first lookup stage.

When changing a rule type, new records may be required for all the rules of the rule type. Optionally, new records are generated for the rules and then the pointers are updated sequentially from the old records which are still in use, to the new records. After all the relevant pointers were changed, the old records are deleted.

If the change in the rule type is performed because of a new rule, the new rule may be first configured into packet filter 24 based on an old rule type, even if inefficient and then the change of the rule type is organized. This allows immediate implementation of the new rule without requiring waiting for changes in a rule type, which may be timely. Alternati ely or additionally, packet filter 24 is configured to test packets for a limited pool of rules using prior art 1011-1066 methods and rules waiting for changes in a rule type are included in the limited pool.

Updating (414) packet filter 24 for a rule, optionally includes listing the change in rule update list 232. Erasing a rule from packet filter 24 optionally includes erasing the rule from rule update list 232. Optionally, rule manager 22 does not change existing rules in packet filter 24, due to the complexity of such an update. Instead, a new rule is updated into the database of packet filter 24 and the old rule is removed. In some embodiments, rule manager 22 manages an internal database, listing for each rule how it was implemented in memory 26, When the rule needs to be removed, rule manager 22 can use the information in the internal database to quickly cancel the rule.

When a rule is removed and the rule is the last rule of a rule type, rule manager 22 optionally cancels the rule type by removing its type-record 204.

Optionally, packet filter 24 is configured to check for all the rules using hash table lookups. Alternatively, when rule engine 20 is configured in addition to a large number of rules handled by lookup tables, with a small number of other rules having different formats from the rest of the rules, the small number of rules may be implemented using other filtering methods, without table lookups. In such cases, for each packet, packet filter 24 performs table lookups for one or more packet types and also checks the other rules using other filtering methods.

Rule manager 22 may relate to all rules in a similar manner. Alternatively, rule manager 22 may handle different rules differently. For example, rule manager 22 may be configured to handle different rules with different priorities. Some rules may be considered of high priority 1011-1066 meaning they must be configured into packet filter 24 within a short period, for example less than 1 second, while other rules may require configuration within less than a minute or only less than 10 minutes, allowing rule manager 22 to configure a plurality of rules together. In some embodiments, rule manager 22 handles separately human-specified rules and automatically generated rules. Optionally, each is assigned a different set of one or more rule types.

It is noted that in some embodiments the access to memory 26 is expected to be the most time consuming act of packet filter 24. Accordingly, various memory access optimization methods known in the art may be used to speed up the operation of packet filter 24. For example, various values accessed in the memory may be set to sizes which are easy to access (e.g., 32 bytes).

In the above description, the contents of the hash table entries have a single value which directs the further handling of the packet. In some embodiments, however, the result in the hash entry comprises a plurality of values corresponding to different conditions, such as different time.Tj Of Ό y Or different times from configuration of the rule. In these embodiments, in each lookup, the value of the entry corresponding to the current conditions is used. This allows configuring the rule engine with rules applying only to specific conditions.

The term "real time" refers herein to tasks being performed shortly after instructions to perform the task are received, for example within less than 10 minutes, less than 10 seconds or even less than 50 milliseconds.

While the above description relates primarily to finding matching records for lookup keys using Hash tables, in other embodiments of the invention other data structures are used

Claims (22)

221,174/2 CLAIMS

1. A rule engine controller, comprising: an input interface adapted to receive rules relating to a plurality of fields of filtered packets; and a rule manager processor configured to: receive a batch of rules through the input interface, define a plurality of anchor sets of fields of the filtered packets, such that each of the rules in the batch of received rules includes conditions relating to the fields of at least one of the plurality of anchor sets , assign each of the rules to one of the anchor sets; and configure the rules assigned to each anchor set into a respective filtering unit, such that the filtering unit begins the filtering of packets for the rules assigned to the anchor set with the fields of the anchor set.

2. The controller of claim 1, wherein the rule manager processor is configured to define the anchor sets, such that at least one of the anchor sets includes a plurality of fields .

3. The controller of claim 1, wherein the rule manager processor is configured to define the anchor sets, such that at least two of the anchor sets include a mutual field included in both the anchor sets.

4. The controller of claim 1, wherein the rule manager processor is configured to evaluate a plurality of different possible anchor sets for the batch of received rules and to 221,174/2 select one of the possible sets which is evaluated to achieve a fastest filtering of packets for the received batch of rules .

5. The controller of claim 1, wherein the rule manager processor is configured to define the anchor sets, such that preference is given to including in the sets, fields having an expected higher distribution of values in the filtered packets relative to other fields.

6. The controller of claim 1, wherein the rule manager processor is configured to calculate estimated probabilities of occurrence of false positive matches for rules for various fields and defining the anchor sets, based on the estimated probabilities .

7. The controller of claim 1, wherein the rule manager processor is configured to assign the rules to the anchor sets, such that each rule assigned to an anchor set includes equality conditions on the fields of the anchor set, which equality conditions are connected by AND operators.

8. The controller of claim 1, wherein the rule manager processor is configured to assign the rules to the anchor sets, such that each rule assigned to only a single anchor set .

9. The controller of claim 1, wherein the rule manager processor is configured to convert rules including an OR operator into a plurality of rules not including an OR operator, before defining the plurality of anchor sets. 221,174/2

10. The controller of claim 1, wherein the rule manager processor is configured to determine for each rule an order of checking of additional conditions of the rule not relating to the fields of the anchor set to which the rule was assigned.

11. The controller of claim 1, wherein the respective filtering units perform the filtering by accessing a hash table and wherein the rule manager configures the hash table with a rule by indicating in an entry of the hash table corresponding to values required by the rule for the fields of the anchor set, a possible match of the rule.

12. The controller of claim 1, wherein the respective filtering units perform the filtering by accessing a TCAM.

13. The controller of claim 1, wherein the respective filtering units perform the filtering by accessing a TRIE or an Extended Grid of Tries with Path Compression.

14. The controller of claim 1, wherein the rule manager processor configures the rules into the respective filtering unit, while the packet filtering processing unit is filtering packets .

15. The controller of claim 14, wherein the rule manager processor configures the rules into a memory of the respective filtering unit, while the packet filtering processing unit is filtering packets, without locking the memory unit for access by the packet filtering processing unit.

16. The controller of claim 1, wherein the rule manager is adapted to continuously receive rules throughout the operation 221,174/2 of the filtering processing units and to configure the rules into the respective filtering unit in real time.

17. The controller of claim 16, wherein the rule manager is adapted to receive rules of different priorities and to configure the rules into their respective filtering unit within time frames determined according to their priorities.

18. The controller of claim 16, wherein the rule manager is adapted to reevaluate the defined anchor sets in real time for at least some of the received rules.

19. The controller of claim 18 wherein the rule manager does not reevaluate the defined anchor sets for at least some of the received rules.

20. The controller of claim 16, wherein the rule manager is adapted to reevaluate the defined anchor sets periodically, according to the rules currently configured into the filtering units .

21. The rule engine of any of claims 1-20, comprising a filtering processor implementing the respective filtering units of the anchor sets and wherein the filtering processing unit is configured to perform only real time tasks operating at the rate of packet filtering.

22. The rule engine of claim 21, wherein the rule manager is configured with management information of data structures used by the filtering processor, which are not accessible by the packet filtering processor.