raid 10 performance vs raid 5 1

[chenthorn]

I have four new dell power edges each with dual xeon 1.8gz
processors and 2gb of ram. I have set up the first three
servers with a raid 5 data stripe, and the fourth I did a
raid 10. I did some benchmarking, and I do not seem to
notice any performance difference between the four
servers.
Should I see a difference in performance between servers
when the only difference is the raid solution? I am doing
very heavy inserting and updating of multiple multi-
million row tables on these servers, and I am not getting
much better performance than my dual 450 xeon server.
any advice would be greatly appreciated!!!
thanks!

 

[MeanGreen]

A lot of times more POWER does not dictate better performance. You probably need to optimize your SQL to get better performance. Many times a business thinks that by throwing more Processing Power that they will get better performance from their database, but many times optimization or possibly a redesign of the database would return much better results. As for Raid 5 versus Raid 10, there really is very little overhead related to the two since the mirroring that comes with Raid 10 is background processed most times.

Hope this helps.

 

[SQLSister]

Things you probably need to consider to improve performance are indexes (these can slow down inserts and updates just as they speed up selects, especially clustered indexes on multimillion record tables on a text field.)

If you are calling your SQL from ASP code instead of from stored procedures this will slow things down.

You should also make sure your indexes and transaction logs are on separate drives with separate drive controllers from the main data, this will crease the multitasking the systems can do.

If alot of those millions of records are historical data never to be changed again, consider moving them to a data mart or archive tables (depending on the frequency and type of access to the data needed)

Consider denormalizing your data somewhat to reduce the number of joins needed to perform common queries , but make sure you set of ways to make sure that automation, not user input supplies the denormalized values in order to maintain dat aintegrity).

Get rid of any queries using cursors that can possibly be gotten rid. Cursors are extremely slow and in most cases can be replaced with set oriented code that runs much faster.

Consider updating your statistics or defragging your indexes.