SUMMARY 2 -- Disksuite Question

From: Bruce Purcell <bpurcell_at_csuhayward.edu>
Date: Fri Mar 21 2003 - 12:36:19 EST
The answers continue to roll in with pretty the same advice -- I forgot
to mention that most are also suggesting switching from RAID0+1 to
RAID1+0, so I will look at this as well. The following answer is quite
detailed and the best overall summary and provides some really good
explanation. It took Blair Rampling a whole hour to respond (damn, you
guys are fast) with it, so I summarized too quickly. Here's his
response:
 
It is possible to assign slices on the same disk to different
metadevices using different RAID levels but with this configuration you
will lose on I/O performance because you will have two different I/O
patterns trying to access the same read/write head. The advantage of the
SAME (Stripe and Mirror Everything) methodology for Oracle is seen
mostly if the disks are dedicated to the RAID1+0 array. Remember that in
a 1+0 during a multi-block read or write, each disk will be accessed in
sequence. If the RAID5 volumes are doing an I/O at the time, the process
doing the I/O for oracle will block until the read or write for the
RAID5 array is done. This makes it hard to say if the performance
increase picked up by adding the four spindles to the array (from 8
disks with no RAID5 to 12 with the RAID5 slices) will overcome the
performance hit from I/O contention. This all depends on how busy the
image storage metadevice is. If it was up to me, I would separate the
two.

As for a RAID5, you can put as many disks as you want into a RAID5
group. You add write performance with each additional spindle you add,
similar to a RAID1+0. The trade-off with RAID5 is that while the disk
space savings over 1+0 are larger, the performance increase per spindle
is smaller.

 
Bruce Purcell
Enrollment Services
Director, PM/TS
bpurcell@csuhayward.edu
510 885 4494
510 885 4595 (fax)
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Fri Mar 21 12:39:24 2003

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:43:07 EST