Re: RAID Strip size

[Brent Baisley]

Actually, you want to try to match the stripe size to your data size. 
The ideal would be to have a stripe size equal to the size of a record 
in your database. This way the disk needs only one read or write for 
each database record. You really don't want to fragment a record. A 
large stripe size is good if you have large files, like graphic files, 
that you read in their entirety. It's bad if you are reading small 
amounts of data, like in a database.

For instance, if you set a stripe size of 128K and you need to read 100 
records that are not in the same disk sector, the disk ends up 
retrieving over 12MB of data. If your typical record only contains 2K 
of data, that's only 200K of data you need out of the 12MB that the 
disk retrieved. That's a big waste.
On the flip side, a stripe size too small will fragment your records 
and cause excessive disk access.
On the other hand, if you are typically doing full table scans, then 
you are reading most of a large file and a large stripe size would be 
good.

You really need to know your data and how it is accessed in order to 
set an optimal stripe size. Even then, you need to benchmark to see if 
what you set if correct, especially since different tables have 
different data sizes and access patterns.

When in doubt, it's usually best to leave the stripe size at the 
typical default of 4K. The reason for this is that most operating 
systems track RAM in 4K increments, so there is a one-to-one relation 
between disk sector size and RAM "sector" size. At least that's what I 
learned a few years ago in an IBM class, perhaps RAM is tracked 
differently now with the extremely large RAM configurations that are 
now possible.

On Dec 3, 2003, at 4:05 PM, trevor%tribenetwork.com wrote:

Greetings Mysqlians,



            Please comment on the validity of my logic:



In setting the RAID(10/2disks) strip size everything I read says you 
must
benchmark your particular system.  Since that is not an option, my 
current
logic is to have a large strip size (1024) with the reasoning that 
fewer
writes/reads (yet longer writes) will be better in a database which 
has a
large amount of disk access.  The disk cache size is 1GB on our disk 
device
but that is not quite enough to hold all the tables which get
accessed(written to and read from) frequently.  I figure setting a 
large
stripe size is a conservative approach allowing for better scalability.



Many Thanks,



Trevor


--
Brent Baisley
Systems Architect
Landover Associates, Inc.
Search & Advisory Services for Advanced Technology Environments
p: 212.759.6400/800.759.0577
--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql
To unsubscribe:    http://lists.mysql.com/[EMAIL PROTECTED]