Re: Matching storage stripe size with ASM allocation unit?
- From: Yong Huang <yong321@xxxxxxxxx>
- To: oracle-l@xxxxxxxxxxxxx
- Date: Tue, 27 May 2008 09:27:20 -0700 (PDT)
Thanks, Alex and Gopal. We use external storage as most shops do. I was looking
for an explanation why for best performance the storage team needs to use a
stripe size of 1M for Oracle ASM, assuming they're extremely collaborative and
1M is technically feasible. Nitin and others' ASM book has one paragraph on
p.76 that sounds relevant:
"the best stripe size [contiguous space on 1 disk, not "stripe width" in their
terminology which means stripe size times number of striped disks] for reading
is 1MB. However, few storage arrays support that large a stripe. For efficient
writing, it is desirable to have a single ASM extent span all the disks in the
RAID 5 set; for example, with four data and one parity, a stripe size of 256K
would mean that a full 1MB ASM extent write would hit all disks, and no extra
reads would be required for calculating parity. Note that for this to work, the
correct alignment is critical."
I thought the ASM allocation unit was only meaningful during file or ASM disk
creation or resizing. It looks like it does play a role in regular I/O
operations. Also, the book implies that if storage array supports, 1M stripe
size (i.e. 4MB stripe width in case of 4+1 RAID 5) is better than 256K stripe
size; we fall back to 256K only because "few storage arrays support" 1M stripe
size. The 1M stripe size implied here is exactly what the Oracle consultant
told me.
Yong Huang
--
//www.freelists.org/webpage/oracle-l
Other related posts:
