RAID10+10Krpm vs RAID5+15Krpm

[Aaron Couts]

I'm setting up a SCSI drive array, and I was wondering, which is more
significant -- RAID level or drive speed? I'm considering the following
setups:
RAID 10 with 10K rpm drives
RAID 5 with 15K rpm drives

Obviously RAID 10 will always have the advantage of greater redundancy,
but I've never had two drives fail at the same time, and with RAID 5
I could probably afford a hotspare.

Anyone been able to compare the speeds of these two setups? My machine
is a heavily loaded database server with roughly twice as much read
activity as write activity.

Thanks,
Aaron

 

[J. Clarke]

I'm setting up a SCSI drive array, and I was wondering, which is more
significant -- RAID level or drive speed? I'm considering the following
setups:
RAID 10 with 10K rpm drives
RAID 5 with 15K rpm drives

Obviously RAID 10 will always have the advantage of greater redundancy,
but I've never had two drives fail at the same time, and with RAID 5
I could probably afford a hotspare.

Anyone been able to compare the speeds of these two setups? My machine
is a heavily loaded database server with roughly twice as much read
activity as write activity.

As a general rule RAID 5 is at a disadvantage for writes because of the
parity calculation that has be be performed for every sector written. For
reads the only way to be _sure_ is to do a test with samples of the actual
data and usage pattern.

 

[Marc de Vries]

I'm setting up a SCSI drive array, and I was wondering, which is more
significant -- RAID level or drive speed? I'm considering the following
setups:
RAID 10 with 10K rpm drives
RAID 5 with 15K rpm drives

Obviously RAID 10 will always have the advantage of greater redundancy,
but I've never had two drives fail at the same time, and with RAID 5
I could probably afford a hotspare.

Anyone been able to compare the speeds of these two setups? My machine
is a heavily loaded database server with roughly twice as much read
activity as write activity.

For enterprise servers a RAID10 array with lots of spindles is usually
assumed to be the fastest solution for databases.
Of course the question remains whether that is more than the extra
performance of the 15k drives.

Still you have quite a lot of write activity, so I think that Raid10
would be the fastest.

But are the logdrives included in that write activity? In that case
you might want to use a seperate disk or seperate RAID1 volume for the
logdrives.
If you database then mainly has read activity you can use Raid5 for
the database itself.

Also you should tru to maximize the numbers of harddisks in your
array. A database usually has mostly random reads and writes, and
performance of that will be increased if you use more harddisks in the
array. So if you have the room in your server/enclosure, use a large
number of smaller disks instead of a smaller number of large disks.

You will find this article interesting:
http://www1.us.dell.com/content/topics/global.aspx/power/en/ps3q99_raid?c=us&l=en&s=esg

Also a little test here from someone who tested raid5 and raid10 on
MySQL: http://www.johnleach.co.uk/documents/benchmarks/raidperf.html

Marc

 

[Aaron Couts]

But are the logdrives included in that write activity? In that case
you might want to use a seperate disk or seperate RAID1 volume for the
logdrives.
If you database then mainly has read activity you can use Raid5 for
the database itself.

Thanks for your very helpful response and links. I think we're going to
do both -- a RAID10 for the logfiles and some of the data, and a RAID5
for indexes. We have everything on a single a RAID10 now.

-Aaron