<HTML><HEAD>
<META http-equiv=Content-Type content="text/html; charset=windows-1252">
<META content="MSHTML 6.00.2900.5626" name=GENERATOR></HEAD>
<BODY style="MARGIN: 4px 4px 1px; FONT: 10pt Tahoma">
<DIV>Sorry for the aside on this, but this info might be helpful for anyone who is using SLES10 SP2 and testing performance using /dev/zero.</DIV>
<DIV> </DIV>
<DIV>I have noticed extreme drops in output from /dev/zero on any SP2 kernel. This has actually been confirmed by Novell support... Evidently the SLES programmers in Germany changed some of the properties of /dev/zero to fix "other memory issues". According to them, /dev/zero is no longer a good measurement of throughput. Supposedly, the programmers are looking into restoring /dev/zero output performance levels, but their comment was "no one uses /dev/zero for REAL performance testing these days".</DIV>
<DIV> </DIV>
<DIV>Here is a good example:</DIV>
<DIV>On a HP DL365 with 32GB of RAM, the following test: ( dd if=/dev/zero of=/dev/null bs=1G count=1 ) produced an output of 28GB/sec on SP1 and 700MB/sec on SP2. I also noticed performance differences on output to raid hardware, but don't have those numbers. </DIV>
<DIV><BR>Andrew Armstrong</DIV>
<DIV> </DIV>
<DIV><BR>>>> On 10/20/2008 at 6:33 AM, in message <20081020103350.GA10572@soda.linbit>, Lars Ellenberg <lars.ellenberg@linbit.com> wrote:<BR></DIV>
<DIV style="PADDING-LEFT: 7px; MARGIN: 0px 0px 0px 15px; BORDER-LEFT: #050505 1px solid; BACKGROUND-COLOR: #f3f3f3">On Sat, Oct 18, 2008 at 09:18:33PM +0200, Petersen, Joerg wrote:<BR>> Well, my experience is that 8.0.7 is the fastest DRDB...<BR>> Later Versions just got slower!<BR>> Try: no-disk-flushes<BR>> and: no-md-flushes<BR>> if you are using DRBD >= 8.0.13<BR>> Not completely same to 8.0.7 but near to it...<BR><BR>to recommend a known buggy version for imagined performance reasons is<BR>not particular HA.<BR><BR>> Hi Lars,<BR>> <BR>> I did these:<BR>> <BR>> node1:/opt # hdparm -t /dev/drbd0<BR><BR>> Do these tests make any sence?<BR><BR>if you care for write performance,<BR>why are you doing read benchmarks?<BR><BR>-- <BR>: Lars Ellenberg<BR>: LINBIT | Your Way to High Availability<BR>: DRBD/HA support and consulting <A href="http://www.linbit.com">http://www.linbit.com</A><BR><BR>DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.<BR>__<BR>please don't Cc me, but send to list -- I'm subscribed<BR>_______________________________________________<BR>drbd-user mailing list<BR>drbd-user@lists.linbit.com<BR><A href="http://lists.linbit.com/mailman/listinfo/drbd">http://lists.linbit.com/mailman/listinfo/drbd</A>-user<BR></DIV></BODY></HTML>