Thanks for all the advice - thanks to Romeo Theriaults awesome tool
I'll run some stats over time and see how things look.
I'll probably give Logicmonitor a quick look and see how it compares to DFM.
Cheers,
Raj.
Raj,
I'll share knowledge I've received specific to NFS environments (and
this may be specific to VMware over NFS - sorry I can't confirm). The
way I understand it is that clients start experiencing negative
performance after 20ms disk latency. I do not know if that same latency
applies to iSCSI traffic.
As it relates to the disks making up your aggregates, I can offer the
following numbers as they relate to disk IOPS, which is also somewhat
related to the original request in this thread. Numbers below are the
IOPS that disk can sustain before experiencing a >= 20ms delay.
FC@10KRPM: 120IOPS
FC@15KRPM: 220IOPS
SATA: 40IOPS
I had looked on NOW for this type info but did not come up w/anything
quickly - perhaps you could reach out to your SE.
-/-
Kevin Parker
919.521.8413
http://theparkerz.com
Blackle.com - Saving energy one search at a time.
-----Original Message-----
From: owner-toasters@mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Raj Patel
Sent: Monday, October 05, 2009 12:59 AM
To: Romeo Theriault
Cc: Maxwell Reid; vera.lee@avagotech.com; Milazzo Giacomo;
toasters@mathworks.com
Subject: Re: High CPU utilizations and file I/O
As a matter of interest how much latency is to much (I know, how many
angels can you fit on a pinhead) ?
We've had a few DBA types come and comment on long SQL queue lengths
associated with disk i/o - the SQL luns are connected via iSCSI.
As far as the end user is concerned performance seems fine for things
like SharePoint and CRM. Even Exchange. However a few apps with high
i/o (mail archiving to database and SCOM which seems to thrash the
database) do seem to suffer a little in terms of slow response time.
Do people have tips for optimising iSCSI performance (targets are ESX
servers, Exchange on physical hardware and SQL on a mix of physical
and virtual servers) ?
What key counters should I keep an eye on ?
Fibre Channel is still prohibitively expensive plus 10GbE seems very
promising and more pervasive than a few years ago.
Cheers,
Raj.
On 10/3/09, Romeo Theriault romeotheriault@gmail.com wrote:
I think everyone would be interested in the updated version!
Great! Ok, I'll put up the version that can also filter on aggregates
on
Monday. I hadn't put it up earlier mainly because it takes a bit more
setup
to get working. I used the netapp perl api to access the aggregate
information so it also requires a user account with the appropriate
api
permissions and the associated netapp perl api modules. I just haven't
gotten around to figuring out if I can legally bundle the netapp perl
modules and just generally putting it in a easily usable package for
the
general public. I'll try to figure out the details on monday and see
what I
can get out, with or without the netapp modules.
--
Romeo Theriault
System Administrator
Information Technology Services