From: Swider, Bucky (jswider@concordefs.com)
Date: Thu May 22 2003 - 09:23:15 EDT
Hi All,
First off, where's a good place to ask non time-critical questions? I can't
get to the newsgroups listed in the FAQ, This is the best group of people
to reach, and I hate to abuse the system with non time-critical queries such
as this one....
Anyhow, we're going through some intense debates in deploying our EMC SAN.
The main questions revolve around MetaDevice size. (For those not
EMC-fluent, a MetaDevice is a virtual disk created by 'x' number of
hypervolumes, with a hyper representing a physical volume, when laid out
correctly. EG, an 8-way meta made up of 8-GB hypers would present to
solaris as a single 64GB LUN, although on the back end it is striped over 8
different physical devices, and 4 different disk controllers).
One school of thought is to keep the metadevices smaller, i.e. 32GB, and
glue them together on the host with Volume Manager- the theory being that
you have more device queues to parallellize across. The other school of
thought is to make the metadevice as large as you need; e.g. a 200GB
metadevice presented as a single LUN to solaris, with the theory that using
Volume Manager to glue luns together is ultimately a bad thing, as sysadmins
have a tendency to eventually overlook the physical backend and create a
Volume with componenets on the same physical.
Anhyow, what do I need to worry about as far as device queues? With large
LUNS, could I bottleneck on a single device queue in SD, in the HBA driver
(emulex), or anywhere else? Which IO queues do I look at in iostat to
determine thresholds? Are there any good docs on this subject? I've found
lots of info about IO from a filesystem perpective, but scant little about
the device level. Thanks!
Bucky Swider
Concord EFS
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:26:27 EDT