The question of adequacy is an interesting one. a 2GB ARC is fine for
many things, but remember that it's a read and write cache.
AND - if you are using ZFS root, then it's also the cache for your OS disks.
I'm not saying ZFS if your problem - but I have certainly found that
more memory is better for workloads where the *read* cache is of
assistance... (and having enough in reserve for the write is of course
beneficial).
In a recent T-series deployment, we decided on 4GB - and that was
enough. We tuned it back to 2GB, and the whole box ran like crap. (also
of interest was that it was not simple to observe that it was the ZFS
tuning that had lead to this!)
So - Start bigger and see where it leads.
Cheers!
Nathan.
Gary Andresen wrote:
Thanks.
So 4 Gig is still recommended for zfs then. I thought 2Gig was adequate
now days.
I give it a tweak.
Regards
Gary
On 4/12/2010 9:59 PM, Octave Orgeron wrote:
Hi,
If you are using ZFS as the file system in the control domain, I would
allocate atleast 4GB of memory to allow enough space for the ZFS ARC.
There is also a tunable to control the size of the ARC, take a look at
the ZFS guide on solarisinternals.com. Another possible area is the
network settings in the link aggregation on the server or switch side.
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Octave J. Orgeron
Solaris Virtualization Architect and Consultant
Web: http://unixconsole.blogspot.com
E-Mail: [email protected]
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
----- Original Message ----
From: Gary Andresen<[email protected]>
To: [email protected]
Sent: Mon, April 12, 2010 9:32:41 PM
Subject: [ldoms-discuss] ldm list -l primary takes ~30 secs to return
Still trying to track down whats happening at a customer site, but
right now on a T5440 box with 2 CPUs 32 gig of memory and setting the
control domain
to 1 core (8 threads) and 2g of memory, 1 MAU with Solaris 10 U8 with
latest patch cluster,Latest Firmware 139446-10, boot disk is a zfs
pool mirrored.
Network is a aggregate of nxge0 and 4 I believe and vsw0 was created by
ldm add-vsw net-dev=arrg1 primary-vsw0 primary
unplumbed aggr1
plumbed vsw0 in it's place.
All seemed to be going well but;
Running 'ldm list -l primary' takes up to 20-30 secs to printout data.
No errors that they can see (dmesg, /var/adm/messages).
Missing patch? Ideas?
Gary
_______________________________________________
ldoms-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
_______________________________________________
ldoms-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
_______________________________________________
ldoms-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
