>  Another question : is the netflow caching managed by hardware or by        
>  software ? And so does netflow impact the CPU load ?                       

Yes, netflow impacts the CPU. Many Cisco devices do netflow in
software, which can cause heavy CPU loads. Netflow load varies
depending on what kind of supervisor you have in the cat6500, what
kinds of traffic loads are crossing the 6500, and what kinds of
netflow are enabled (full flow, sampled, etc.). 

In my experience, the worst case for generating high CPU loads due to
netflow has been scanning attacks. Dpending on the attack software
every packet can be seen as a new flow, generating high rates of
netflow export.

One of my favorite test tools for lab tests is to find a copy of the
old attack program called "stream.c". This program generates packets
from random source addresses with random ports, causing netflow to
regard each packet as a new flow. Even though this program isn't a
very efficient packet generator, it can still drive Cisco CPUs to high
loads for netflow processing.

I've noticed that the CPUs on Sup1as and Sup2s can be driven to very
high loads (> 90%) due to heavy netflow export under scanning
attacks. 

The Sup720, on the other hand, has a dedicated chunk of TCAM for
netflow operations. This has the effect of capping the max CPU load
apparently due to the dedicated hardware effectively putting a limit
on the max netflow export rate. Lab tests and real world experience
shows that the max CPU rates caused by netflow on the Sup720 for
traffic generated by scanning attacks ranges around 55-60% load.

-Charles

Charles E. Spurgeon / UTnet
UT Austin ITS / Networking
[EMAIL PROTECTED] / 512.475.9265
_______________________________________________
Flow-tools mailing list
[EMAIL PROTECTED]
http://mailman.splintered.net/mailman/listinfo/flow-tools

Reply via email to