Srik1920
2 Iron

HOW RANDOM AND SEQUENTIAL IO'S AFFECT RAID PERFORMANCE

Hi friends,

can you please explain me how random and sequentail IOs going to affect Raid performance.I read in ISM document that

Raid 3 is good for large sequential IOs but I could not figure out how, because no furthure explaination given about that point.Please

explain me what are sequential and random IOs and how they are going to affect RAID 3 and RAID 5 performances.

Tags (3)
0 Kudos
3 Replies
Highlighted
Quincy561
4 Tellurium

Re: HOW RANDOM AND SEQUENTIAL IO'S AFFECT RAID PERFORMANCE

Symmetrix does not support RAID3.  Many years ago we had something called "parity raid" which was very much like RAID3, but that is no longer supported on any of the modern Symmetrix systems

Sequential writes can be optimized if all the write data for a raid stripe is in cache before destage.  In this case we can calculate the parity without doing any reads of old data or old parity.  If you only write one block on RAID5, the Symmetrix must first read the old data and old parity before writing new data and new parity (4 disks IOs)

Srik1920
2 Iron

Re: HOW RANDOM AND SEQUENTIAL IO'S AFFECT RAID PERFORMANCE

Hi Quincy,

Thanks for reply.I too know symmetrix does not support RAID 3.Short but nice,I understood what u wrote.

0 Kudos
RRR
5 Osmium

Re: HOW RANDOM AND SEQUENTIAL IO'S AFFECT RAID PERFORMANCE

it doesn't really read all the data before writing the new stripe to disk.

What happens is this:

D1 D2 D3 D4 D5 D6 D7 P

data needs to be written to D2

D2 is read as well as P (2 read I/Os)

the new D2 is compared to the old D2 and with the old P the new P is calculated.

the new D2 as well as the new P is written to disk (2 write I/Os)

So this gives you a total of 4 I/Os. No matter what width of a RAID5 stripe, the penalty is always 4, because only the affected data "disk" and the parity is touched and none of the other "members" of the stripe.

So a 3+1 as well as a 7+1 have a write penalty of 4 when using RAID5.