PPM provides thresholding mechanisms that can be used to provide a basic level of latency control for snort. It does not provide a hard and fast latency guarantee but should in effect provide a good average latency control. Both rules and packets can be checked for latency. The action taken upon detection of excessive latency is configurable. The following sections describe configuration, sample output, and some implementation details worth noting.
To use PPM, you must build with the –enable-ppm or the –enable-sourcefire option to configure.
PPM is configured as follows:
config ppm: max-pkt-time
config ppm: max-rule-time
Packets and rules can be configured separately, as above, or together in just one config ppm statement. Packet and rule monitoring is independent, so one or both or neither may be enabled.
fastpath-expensive-packets - enables stopping further inspection of a packet if the max time is exceeded - default is off
pkt-log [log] [alert] - enables logging packet event if packet exceeds max-pkt-time - default is no logging - if no option is given for ‘pkt-log’, ‘pkt-log log’ is implied - the log option enables output to syslog or console depending upon snort configuration
debug-pkts - enables per packet timing stats to be printed after each packet - default is off
suspend-expensive-rules - enables suspending rule inspection if the max rule time is exceeded - default is off
rule-log [log] [alert] - enables event logging output for rules - default is no logging - one or both of the options ‘log’ and ‘alert’ must be used with ‘rule-log’ - the log option enables output to syslog or console depending upon snort configuration
The following enables packet tracking:
config ppm: max-pkt-time 100
The following enables rule tracking: config ppm: max-rule-time 50, threshold 5
If fastpath-expensive-packets or suspend-expensive-rules is not used, then no action is taken other than to increment the count of the number of packets that should be fastpath’d or the rules that should be suspended. A summary of this information is printed out when snort exits.
The following suspends rules and aborts packet inspection. These rules were used to generate the sample output that follows.
config ppm: max-pkt-time 50, fastpath-expensive-packets, pkt-log, \ debug-pkt
config ppm: max-rule-time 50, threshold 5, suspend-expensive-rules, \ suspend-timeout 300, rule-log log alert
Packet Performance Monitor Config: ticks per usec : 1600 ticks max packet time : 50 usecs packet action : fastpath-expensive-packets packet logging : log debug-pkts : disabled
Rule Performance Monitor Config: ticks per usec : 1600 ticks max rule time : 50 usecs rule action : suspend-expensive-rules rule threshold : 5 suspend timeout : 300 secs rule logging : alert log
… PPM: Process-BeginPkt caplen=60 PPM: Pkt Used= 8.15385 usecs PPM: Process-EndPkt
PPM: Process-BeginPkt caplen=342 PPM: Pkt Used= 65.3659 usecs PPM: Process-EndPkt
PPM: Pkt-Event Pkt used=56.0438 usecs, 0 rules, 1 nc-rules tested, packet fastpathed (10.4.12.224:80 -> 10.4.14.108:54321). PPM: Process-BeginPkt caplen=60 PPM: Pkt Used= 8.394 usecs PPM: Process-EndPkt
PPM: Process-BeginPkt caplen=60 PPM: Pkt Used= 8.21764 usecs PPM: Process-EndPkt …
Packet Performance Summary: max packet time : 50 usecs packet events : 1 avg pkt time : 0.633125 usecs Rule Performance Summary: max rule time : 50 usecs rule events : 0 avg nc-rule time : 0.2675 usecs
Enforcement of packet and rule processing times is done after processing each rule. Latency control is not enforced after each preprocessor.
This implementation is software based and does not use an interrupt driven timing mechanism and is therefore subject to the granularity of the software based timing tests. Due to the granularity of the timing measurements any individual packet may exceed the user specified packet or rule processing time limit. Therefore this implementation cannot implement a precise latency guarantee with strict timing guarantees. Hence the reason this is considered a best effort approach.
Since this implementation depends on hardware based high performance frequency counters, latency thresholding is presently only available on Intel and PPC platforms.
Time checks are made based on the total system time, not processor usage by Snort. This was a conscious design decision because when a system is loaded, the latency for a packet is based on the total system time, not just the processor time the Snort application receives. Therefore, it is recommended that you tune your thresholding to operate optimally when your system is under load.