I have to agree with Jeff M here. The histogram shown in nfsstat might show unaligned IO, but that doesn't mean misaligned IO. You should expect significant false positives with this tool when any kind of database is involved. The reason is the logging, which produces partial writes, but they virtually never cause problems because it's a sequential overwrite of a file. The partial writes only exist for a split-second before the next write fills out the block.
If MBRscan shows it's aligned, then everything should be just fine.
From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of Jeff Mohler Sent: Monday, June 16, 2014 9:00 PM To: Will.Burchell@skanska.co.uk Cc: toasters@teaparty.net Subject: Re: High CPU VM misalignment confusion
The support guy tells me he seems misalignment when he runs nfsstat -d but MBRSCAN shows these are aligned. What is going on here? --- Easy, support isn't PS.
He's using the output of a tool to provide knowledge, without understanding what the tool is measuring.
MBRSCAN is correct. Nfstsat is observing a signature, without the context of what's creating it, and that would be DB log writes.
"You've got to learn WHY things work on a Starship." -Captain Kirk
On Mon, Jun 16, 2014 at 11:49 AM, <Will.Burchell@skanska.co.ukmailto:Will.Burchell@skanska.co.uk> wrote: Hello. I am hoping you can guide me in the right direction
We have been experiencing very high CPU load on a 7-mode HA pair of 3270 controllers run 8.1.3P2
We have worked with netapp support on these issues and they note our workload is very high on one controller (where we run our VMware setup from)
We also have so called "bad practice" where we are running our exchange ISCSI LUNs on SATA with logs and dbs on the same aggregate (currently separating this out as I type)
I have been told by support we have VMDK misalignment, however I spent a long time a few months ago resolving this firstly by using the VSC tool to confirm the problem and then fixing it with a combination of MBRALIGN and VMware converter as a V2V process
The support guy tells me he seems misalignment when he runs nfsstat -d but MBRSCAN shows these are aligned. What is going on here?
Trying to reduce our CPU and IO burden but getting conflicting information.
Finally I think we should look to upgrade to 8.1.4P2 to remove some bugs? We would consider 8.2.x but I don't think we can as we run Exchange 2010 (using SME 6.x etc)
Thanks in advance
William
_______________________________________________ Toasters mailing list Toasters@teaparty.netmailto:Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters
-- --- Gustatus Similis Pullus