[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-390
Subject:    Re: FW: OSA-Express3 10 Gb, Vswitch, and SLES10 Linux Throughput
From:       Rob van der Heij <rvdheij () velocitysoftware ! com>
Date:       2009-10-27 7:15:11
Message-ID: 6227a1ef0910270015r336ca28bl61b4233f7f59c1d1 () mail ! gmail ! com
[Download RAW message or body]

Don't get me wrong. For most situations VSWITCH is a perfect solution,
and it certainly was at the moment when it was first introduced. But
if you have servers that fully utilize a resource for a longer period,
there is little to share with others. That's the point where should
measure whether trying to share makes sense.
There rarely is a one single solution that works best for all. If you
configure your system based on the numbers from someone else, you may
not have an optimal solution.
In this particular case there were quite a few aspects that made it
interesting to measure.

On Mon, Oct 26, 2009 at 9:44 PM, Harder, Pieter
<pieter.harder@brabantwater.nl> wrote:

> QETH_OPTIONS="buffer_count=128"

This was one of those tiny things. If you do your math right, you will
find that with high network bandwidth it does not take very long to
fill those buffers. Beyond that, the virtual machine needs to be
dispatched so that it can empty the buffers and prepare new ones. So
at that point the latency for dispatching the virtual machine affects
your throughput.
I know Pieter and I talked about the idea to set up multiple devices
so the OSA could fill more buffers without needing the virtual machine
dispatched. But I'm not sure we ever measured that properly.

Rob
--
Rob van der Heij
Velocity Software
http://www.velocitysoftware.com/

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to LISTSERV@VM.MARIST.EDU with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
[prev in list] [next in list] [prev in thread] [next in thread]