Hello Mike,

I will say I have been spending a little time looking up ideas.  It does 
seem like most people who use commodity hardware end up distributing the 
load which is something I have not been thinking too much about at first. 
 I was using an older but beefy dual quad core dell power edge running 
ubuntu server, over 36 gig of ram, velociraptor 1 TB hard drive but have 
recently shifted to using a spare LANforge-Fire Traffic Generator because 
there was an obvious issue with the system bus speed.  The Fire is a 64-bit 
Fedora, One Quad-Core Intel E3, 3+Ghz, 1 PICE x16 slot, 1 10G fibre, at 
least 8GB of RAM, I think a 150 GB standard hard drive.  I would give exact 
specs but I am not at work at the moment. 

I will definitely look at pdflush, load balancing...especially on the 
disks, and tuning write cache.  The more I think about it my biggest 
problem seems like it would be writing to disk.  My disk might even be a 
solid state, I forgot since we ordered it several months ago.  Anyhow, 
thanks again for your ideas.  If you have any more suggestions or things to 
research please let me know.

Dan



On Saturday, August 31, 2013 5:26:27 PM UTC-4, Mike Reeves wrote:
>
> Dan,
>
> There are many factors that go into PCAP capturing at high speeds. In our 
> tests we have been able to get up to about 4Gbps before we saw loss. Since 
> we load balance we were able to run 2 netsniff processes and were able to 
> capture and write between 7 and 8 Gbits per second. This was a beefy box 
> though with external SAS storage balanced over 2 controllers (12Gbps) 
> writing to 30 spinning disks at 4TB each. I would extend your tests beyond 
> 1 minute and run them longer so you can look at things like pdflushing etc. 
> I had to tweak the write cache to get better performance out of the disk so 
> it would write more consistently. Can you specify the specs of the box you 
> are running this on? 
>
> Thanks
>
> Mike
>
>
>
> On Sat, Aug 31, 2013 at 4:15 PM, Daniel Martin 
> <[email protected]<javascript:>
> > wrote:
>
>> Hello,
>>
>> I have been trying to Stress Test our Systems with 10Gbps Network 
>> Traffic.  I have it all going through a Cicso Catalyst Switch and I have 
>> SPAN setup to send all traffic to my netsniff-ng machine.  I am very 
>> confident the system physically should be able to handle the traffic.  I 
>> have the most recent Sniff-NG software compiled and running.  All I am 
>> doing is trying to capture all 10Gbps traffic and also write it to a pcap 
>> file.  So it might be possible the writing to a file could be the weak 
>> link, I am not sure?  I am doing 1 minute tests and I am getting about 89% 
>> packet loss. Perhaps other people out there have ideas or suggestions?
>>
>> Thanks,
>>
>> Dan
>>
>>
>>  -- 
>> You received this message because you are subscribed to the Google Groups 
>> "netsniff-ng" group.
>> To unsubscribe from this group and stop receiving emails from it, send an 
>> email to [email protected] <javascript:>.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>
>

-- 
You received this message because you are subscribed to the Google Groups 
"netsniff-ng" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to