Hi,

 

 

For testing purposes, I want to measure how much traffic a single queue
nprobe instance can handle. Unfortunately, I'm seeing a lot of packetloss
with it. I would expect to see 100% cpu for the nprobe process if it cannot
keep up, but the strange part is nprobe consumes almost no cpu (3.6%).
Pfcount works fine (0% packetloss, 3.7% cpu usage).

 

What am I doing wrong?

 

 

 

The full details:

 

The hardware is a HP DL380G7 (dual sixcore X5670, 12GB ram, Intel dual port
82599EB nic) with RHEL 6.4 x86_64).

I installed the rpm packages of pfring (pfring-5.5.3-6049.x86_64) and nProbe
(nProbe-6.11.130311-3252.x86_64).

 

The pf_ring/DNA drivers are loaded as follows:

insmod /usr/local/pfring/kernel/pf_ring.ko

insmod /usr/local/pfring/drivers/DNA/ixgbe.ko RSS=0,0,0,0 MQ=0,0,0,0
LRO=0,0,0,0 FCoE=0,0,0,0

ethtool -K dna1 lro off gso off gro off

ip li set dna1 up

 

 

If I run 'pfcount -i dna1' I get no packetloss:

 

=========================

Absolute Stats: [8881013 pkts rcvd][8881013 pkts filtered][0 pkts dropped]

Total Pkts=8881013/Dropped=0.0 %

8'881'013 pkts - 7'647'642'728 bytes [193'049.62 pkt/sec - 1'329.91
Mbit/sec]

=========================

Actual Stats: 198364 pkts [1'000.08 ms][198'346.74 pps/1.39 Gbps]

=========================

 

 

And the proc dna stats file (in /proc/net/pf_ring/stats/) gives:

 

Duration: 0:00:00:55

Packets:  10411066

Dropped:  0

Filtered: 10411066

Bytes:    8861158308

 

Cpu utilisation of pfcount is 3.6%.

 

 

 

However when I run 'nprobe -b1 -w 16777216 -e0 -i dna1' I get:

 

15/Mar/2013 16:45:51 [nprobe.c:2004] Average traffic: [14.633 K pps][12
Mb/sec]

15/Mar/2013 16:45:51 [nprobe.c:2011] Current traffic: [14.283 K pps][12
Mb/sec]

15/Mar/2013 16:45:51 [nprobe.c:2017] Current flow export rate: [1349.4
flows/sec]

15/Mar/2013 16:45:51 [nprobe.c:2020] Flow drops: [export queue too
long=0][too many flows=0]

15/Mar/2013 16:45:51 [nprobe.c:2024] Export Queue: 0/16777216 [0.0 %]

15/Mar/2013 16:45:51 [nprobe.c:2029] Flow Buckets:
[active=33407][allocated=33407][toBeExported=0]

15/Mar/2013 16:45:51 [cache.c:744] Redis Cache [0 total/0.0 get/sec][0
total/0.0 set/sec]

15/Mar/2013 16:45:51 [cache.c:1226] LRUCache L7Cache [find: 0 operations/0.0
find/sec][cache miss 0/0.0 %][add: 0 operations/0.0 add/sec][tot: 0]

15/Mar/2013 16:45:51 [cache.c:1226] LRUCache FlowUserCache [find: 0
operations/0.0 find/sec][cache miss 0/0.0 %][add: 0 operations/0.0
add/sec][tot: 0]

15/Mar/2013 16:45:51 [nprobe.c:1871] Processed packets: 1316989 (max bucket
search: 3)

15/Mar/2013 16:45:51 [nprobe.c:1854] Fragment queue lenght: 2

15/Mar/2013 16:45:51 [nprobe.c:1880] Flow export stats: [262157752
bytes/350134 pkts][75054 flows/0 pkts sent]

15/Mar/2013 16:45:51 [nprobe.c:1890] Flow drop stats:   [0 bytes/0 pkts][0
flows]

15/Mar/2013 16:45:51 [nprobe.c:1895] Total flow stats:  [262157752
bytes/350134 pkts][75054 flows/0 pkts sent]

15/Mar/2013 16:45:51 [pro/pf_ring.c:80] Packet stats (PF_RING):
1316989/14692243 [1115.6 %] pkts rcvd/dropped

15/Mar/2013 16:45:52 [engine.c:1921]
[maxBucketSearch=1][thread_id=0][idx=10508315][packet_hash=1470126107]

15/Mar/2013 16:45:54 [engine.c:1921]
[maxBucketSearch=2][thread_id=0][idx=10508233][packet_hash=1470126025]

 

The /proc/net/pf_ring/stats/ file contains:

Duration: 0:00:01:30

Packets:  1763196

Dropped:  19789008

 

The cpu utilisation of the nprobe process is 3.6%

 

If nprobe cpu utilisation would be a problem, how comes it only uses 3.6%?

 

 

 

Kind regards,

Dirk Janssens

 

 

 

_______________________________________________
Ntop-misc mailing list
[email protected]
http://listgateway.unipi.it/mailman/listinfo/ntop-misc

Reply via email to