logical volumes & striping

[Dave Olson]

In <9105101949.AA20619 at centrs1.CENTRCN.UMontreal.CA> chouinar at centrcn.umontreal.ca (Luc Chouinard) writes:

| We have 6 scsi disks on two controlers, and we are
| considering using a striped logical volume for one
| of our partitions.
| 
| This logical volume would be spread across four disks, two
| on each controler.
| 
| We estimate that about 50 users will access this partition
| at the same time.
| 
| We currently believe that striping is always a good thing
| to do.  It distributes uniformely the disk access throughout
| the controlers.
| 
| We have just been advised by a knowledgeable person that
| this would not be true in all cases, specifically ours.
| And the explanation that was given to us makes sense.
| 
| We would appreciate your opinions on the matter of using
| striped volumes, according to the different environments.
| More specifically, we would like to know if there is an
| overhead associated with accessing a single block in a
| striped logical volume, versus a standard partition.


There is of course SOME overhead in using logical volumes.
It is quite low (lost in the noise for almost all the tests
we benchmarked).  Since you are running on 2 controllers, you
should see a net gain for almost all cases.  About the only
time IRIX will read a single block from a file that is > 1
block long is when doing lots of 'random' i/o, and during
directory searches if the directories are small.

I'd be interested in the knowledgable person's arguments,
and whether they are based on measurements on SGI or other
implementations of striped disks.
--

	Dave Olson

Life would be so much easier if we could just look at the source code.