Bug #2100: af_packet: High latency - Suricata - Open Information Security Foundation

Actions

Copy link

Bug #2100

closed

af_packet: High latency

Added by Igor Novgorodov over 7 years ago. Updated about 6 years ago.

Status:

Closed

Priority:

Normal

Assignee:

Eric Leblond

Target version:

4.1rc2

Affected Versions:

Effort:

Difficulty:

Label:

Description

Same configuration as in https://redmine.openinfosecfoundation.org/issues/2099

Network topology is simple:

[host1 eno50] <-> [eno50 host2(suricata) eno49] <-> [eno50 host3]

host1# ip addr add 192.168.99.1/24 dev eno50
host3# ip addr add 192.168.99.2/24 dev eno50

host3# ping 192.168.99.1 -c 1000 -q -f 
PING 192.168.99.1 (192.168.99.1) 56(84) bytes of data.

--- 192.168.99.1 ping statistics ---
1000 packets transmitted, 1000 received, 0% packet loss, time 19977ms
rtt min/avg/max/mdev = 23.308/26.272/39.973/0.457 ms, pipe 3, ipg/ewma 19.997/26.274 ms

When using NFQUEUE and routing through Suricata the latency is fine (sub-millisecond)

Actions

Copy link

Updated by Andreas Herz over 7 years ago

Assignee set to OISF Dev
Target version set to TBD

Actions

Copy link

Updated by Igor Novgorodov over 7 years ago

Testing with latest git confirms that problem still persists, latency is still about 20ms.

Actions

Copy link

Updated by Eric Leblond over 7 years ago

Are you testing with tpacket_v3 ? If yes can you test without it (so with v2) ?

Actions

Copy link

Updated by Igor Novgorodov over 7 years ago

Yep, it was TPACKET_V3.
I've read somewhere that V3 is experimental, but didn't pay much attention to that.
With TPACKET_V2 it's fine - 0.2ms, thanks!

Actions

Copy link

Updated by Eric Leblond over 7 years ago

It is not really experimental regarding to IPS. It will never work correctly: tpacket_v3 is using a block concept that contains a group of packets and deliver block by block so this induce a latency..

Actions

Copy link

Updated by Victor Julien over 7 years ago

Eric should we add a big fat warning or even outright refuse to work in IPS mode with AFPv3?

Actions

Copy link

Updated by Igor Novgorodov over 7 years ago

Yes, the CRIT log message during startup would be very nice and some mention in the docs, so others would not guess the cause of latency.
Sadly, but V2's performance is much worse. I was able to achieve the 0% drop on V2 only with 35% lower PPS than with V3.

Actions

Copy link