We don't see any drops, the frag packets are simply "missing".  We noticed
this with 16-byte as well.

Using BreakingPoint 10g traffic generator we can send packets with
different fragment attack methods.  Anything less than 24-byte
fragments does not get reassembled completely.  24-byte and above
reassemble fine.  We noticed this when  packet captures showed some
fragments missing in Wireshark when looking at the pcaps.  The
fragments that are there seem to be assembled in proper sequence.

If we disable defrag in pfring.conf or remove pf_ring from the
picture, all fragmented packets arrive.

We tested using:  CentOS 6.4 - kernel 2.6.32-358.14.1 X_64.
PF_RING Versions:  5.6.0 and 5.6.1
Traffic speeds: 200Mb/s  - 600Mb/s
VMWare ESXi
Intel 10G NIC
Driver for 10G CentOS - VMXNet3 and the NIC is NOT bridged directly to the VM.
Driver for 1G system - e1000 pfring enabled.
Defrag is enabled via pfring.conf and verified.
Transparent mode - 0  ( for these tests)
Test method - frag - 8-byte and 8-byte out-of-order  ( 16 byte has the
same issues )

Sifting through the code we see the ip_defrag is a call to the CentOS
kernels' internal defrag routine.  So this begs the question of kernel
"tuning".  Although I don't find anything directly effecting that
issue.

So my question is, has anyone else noticed this behavior and was there
a way to correct it?

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

Reply via email to