RE: Iperf Performance


Thanks for the response. I really appreciate it.
Even though I agree on your point that there might be a Network
bottleneck some were down the line, If 2 TCP streams can transfer more
packets and we achieve a much higher throughput
(130Mbytes/sec(2streams)compared to 90Mbytes/sec(1 stream) here)what's
limiting Iperf not to achieve the same when the whole CPU is there for
it to us? 

Btw, Netperf was giving us about 130Mbytes/sec for 1 TCP stream but
takes almost 100% CPU. 

Is there some tweaking that we could do to increase the throughput of 1
TCP stream to the practical extent the hardware can support?

Thanks for the help,
Samson 

-----Original Message-----
From: owner-iperf-users --at-- dast.nlanr.net
[mailto:owner-iperf-users --at-- dast.nlanr.net] On Behalf Of Marc Herbert
Sent: Thursday, August 24, 2006 10:30 AM
To: iperf-users --at-- dast.nlanr.net
Subject: Re: Iperf Performance

On Wed, 23 Aug 2006, Sahadevan, Samson joseph dev wrote:

> When I run Iperf with an I/O buffer size of about 64 bytes (-l 64) and
1
> TCP stream, the CPU remains almost idle (99%-100%) and I get a BW of
> about 90Mbytes/sec (1GB link).

> Is there a way to use the CPU more efficient thereby driving more
> throughput?

The CPU being idle likely means the bottleneck is elsewhere. For
instance you may have a switch or a NIC which has reached its maximum
packet/seconds rate. In this case the CPU can not help.

You can easily verify if the bottleneck (whatever it is) is
packets/second limited by slowly increasing the packet size and
observing the maximum packets/second you get along the way.

Packets/second bottlenecks are typically found in hardware IMHO.

Keep us posted.



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index