Hi,

I hope nprobe is also discussed here because I have a problem with it.

I'm testing nprobe for complete traffic statistics for a cable provider.
Currently we have about 1Gbit/s of traffic to scan but will be divided
over 4 nprobe servers in it's final implementation.

Anyway, with initial iperf tests and nprobe everything seemed okay but
with the first tests on live traffic the nprobe process was allocating
memory like a madman. About 1MB per second. After there wasn't any
memory left, the server crashed. (The server had 2500MB of memory)
Initially I blamed weird dual Xeon Compaq setup and/or/even RedHat but
back in our Office I set up a new nprobe server (dual Pentium Pro-200
with 64MB of memory) based on Slackware 9.1 and captured on the live
nprobe server a pcap stream of 91MB (1.000.000 packets) and copied it to
the new machine.

Replaying the captured stream with tcpreplay I saw the nprobe memory
footprint rise to 21MB. This means the memory increase was cause by the
amount of different flows. (I'm talking here about 1500 exported flows
per second)
Playing the stream again DIDN'T increase the allocated nprobe memory
size. I would figure nprobe keeps memory allocated for each unique flow
even if the flow has already expired.

Is this the normal behaviour of nprobe or do I have a situation which
hasn't tested before?

Any help would be welcome.

Thanks,

Danny
_______________________________________________
Ntop-dev mailing list
[EMAIL PROTECTED]
http://listgateway.unipi.it/mailman/listinfo/ntop-dev

Reply via email to