This is not really good news because these servers do not really have quieter
moments with lower load. They are SIP proxies running the registration for our
VoIP customers' phones, where each phone registers itself with the proxy every
2 minutes, 24x7, unless the phone is switched off.
I will investigate other possibilities, like off-loading the logs (which eat
90% of the disk bandwidth) to some other box that does not really care if the
disks are slow for a while.
For a future PERC, having a controller with two batteries running staggered
charge cycles would be a nice feature, like the one found on the entry-level
EMC SAN's.
Thanksfor the tips anyway,
Robert
________________________________
De : [email protected]
[mailto:[email protected]] De la part de James Kunze
Envoyé : vendredi, 17. juillet 2009 22:04
À : [email protected]
Objet : RE: Battery learn-cycle = horrible write performance
I tried forcing write back with omconfig on 2950 servers where I a) had
more than reasonable confidence in ups, and b) didn't care anyway because the
data was replicated. I couldn't get it to actually work. And while delaying
the learn seems to work, forcing a learn at a reasonable time does not reset
the clock - in other words, the scheduled learn cycle occurs anyway. So we
have resorted to scrips that warn of impending learn cycles and delay to
quieter times.
From: [email protected]
[mailto:[email protected]] On Behalf Of
[email protected]
Sent: Friday, July 17, 2009 7:48 AM
To: [email protected]; [email protected]
Subject: RE: Battery learn-cycle = horrible write performance
You're probably better off delaying the battery learn till a low load
time. The reason I wouldn't recommend switching it to forced write back is that
the battery isn't protecting the controller cache for a sufficient period when
it is in learn mode. Therefore if the server lost power you would be much more
prone to data loss. This is the reason for the behavior.
From: [email protected]
[mailto:[email protected]] On Behalf Of Robert von
Bismarck
Sent: Friday, July 17, 2009 8:35 AM
To: linux-poweredge-Lists
Subject: Battery learn-cycle = horrible write performance
Hi guys,
We have a PE1950III with a PERC6 and two 15k 74Gb disks in RAID 1. This
server runs a telephony app that is pretty verbose and hits the disk pretty
hard (5Mb/s sustained). During the battery learn-cycle the performance drops,
and the load on the server goes through the roof, until the controller is
satisfied that the battery is ok. In the logs we see that the controller sets
the cache to write-through for this operation. This is a pretty sensible thing
to do, but it usually happens late at night, and our duty tech gets woken up
because it triggers our monitoring system.
Going through the OMSA documentation, we found that we could set the
cache to "Force Write-Back" with omconfig
omconfig storage vdisk action=changepolicy writepolicy=fwb controller=0
vdisk=0 (untested command)
Would enabling this over an extended period be bad, or would it be
better to enable it before the learn-cycle and disable it right after it's done
?
Thanks for any hints,
Robert von Bismarck
Senior Systems Engineer
VTX SERVICES SA
Une société du groupe VTX Telecom
================================================================
Tél : 021 721 11 11 - Mobile : 079 541 37 28 - VoIP VTX 062 503 79 05
Av. de Lavaux 101 - 1009 Pully
http://www.vtx.ch <blocked::http://www.vtx.ch/> - robert.von
[email protected] <blocked::mailto:[email protected]>
----------------------------------------------------------------
VTX, votre partenaire telecom proche de vous !
================================================================
_______________________________________________
Linux-PowerEdge mailing list
[email protected]
https://lists.us.dell.com/mailman/listinfo/linux-poweredge
Please read the FAQ at http://lists.us.dell.com/faq
