Networker

Re: [Networker] Performance Tuning backup

2007-07-26 12:05:19
Subject: Re: [Networker] Performance Tuning backup
From: Davina Treiber <Davina.Treiber AT PEEVRO.CO DOT UK>
To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
Date: Thu, 26 Jul 2007 17:01:51 +0100
Reed, Ted G [IT] wrote:
You have an assumption in your math section that is not valid. You assume that GigE can perform at 125M/sec. And theoretically, that's correct. The same way, theoretically, a 100Mb NIC can do 12.5M/sec. However, the reality is that the overhead on a 100Mb means 8-9M/sec is realistic throughput for client data. And while you'd think that'd mean that a realistic throughput for GigE then must be 80-90M/sec (GigE=10*100Mb), the fact is that 40-50M/sec is all you can expect to see out of a GigE card, out of the box, without some level of performance tuning. With extensive tuning, I've seen people get up to 70M/sec out of a GigE...maybe the occasional burst higher. But in the cases I've dealt with directly, we've come to the conclusion that an out-of-box GigE is the equivilent of 5*100Mb, or half what the specs lead you to believe.
My tests in years gone by bear your figures out. Additionally:
* GigE on fibre is quicker than GigE on copper.
* Solaris gives higher numbers than Windows, perhaps a more efficient TCP/IP stack. * When I installed Linux on the same Intel hardware I used to test Windows throughput I found that the throughput was the same but with significantly less CPU overhead.

To sign off this list, send email to listserv AT listserv.temple DOT edu and type 
"signoff networker" in the body of the email. Please write to networker-request 
AT listserv.temple DOT edu if you have any problems with this list. You can access the 
archives at http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER