Without going too crazy into technical detail (statit counters), the
easiest way has been to look at sysstat output: if disk reads are
significantly higher than net out or tape backup (depending on the
workload - ie, no reconstructs or reallocates running), then you could
probably use minra.
That type of fingerprint signifies that the disks have more reads than
necessary. Some are ok but too many will cause performance problems.
Beyond this, you can look at the statit and wafl_susp counters to
determine things about disk fragmentation and I/O and readahead cache
(respectively).
Glenn
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Peter D. Gray
Sent: Thursday, August 31, 2006 1:59 AM
To: toasters(a)mathworks.com
Subject: when to use minra ?
Does anybody have recipes or suggestions on how
to work out if turning on minra will improve
or decrease performance? Are there counters
you can query in ONTAP that are helpful?
Regards,
pdg
--
See mail headers for contact information.
sts more data from the same file,
the data is already in the cache so the response is faster.
However, if your file reading pattern is completely random, then
minra off may result in extra disk reads that are wasted because
the client never requests the extra data read from the disks.
Your reading pattern may not be as random as you think, however,
and it is very unusual for minra on to be helpful.
I have seen one situation where minra on was a win. On a FAS960c clustered
pair and we had FC LUNs. The clients were Solaris boxes running Veritas.
For performance we used Veritas to stripe a pair of LUNs, one LUN from each
filer. We built a Veritas filesystem on the striped LUNs. As you can
imagine, this thoroughly scrambled up file data because it was striped over
two filers, somewhat defeating the Netapp read ahead strategy. We were
storing email on this striped filesystem and so we had a very heavy read
load. With minra off disk reads in KB/sec were about 10% to 20% higher
than FC writes, indicating that the extra reads were probably going to
waste. I turned minra on and the disk reads dropped to a little less than
the FC writes.
We have since switched from FC LUNS to NFS mounted volumes and turned
minra back off.
Steve Losen scl(a)virginia.edu phone: 434-924-0640
University of Virginia ITC Unix Support