question

DaniSQL avatar image
DaniSQL asked

What counters should I baseline before moving to a SAN?

for the last three weeks I was moving some of my servers off of direct attached storage(DAS) to a SAN. I have had success with all of them and enjoyed performance boost and flexiblity so far. However one of the servers suddenly start taking 2-3 times for maintenance jobs(backup and integrity check). This happened after they did SAN related maintenance on the backup array which theoretically should only affect the backup process. However integrity check is also taking very long time. The only thing i have on this instance running is maintenance jobs since this is a new server and the databases are getting upgraded to support an application upgrade so I dont have application/database performance baseline to compare it with....No configuration change is made regarding SQL SErver. What do you think should I check for troubleshooting(or tell the SAN team to check)? (Additional info about the environment: This particular instance is a win 2008 R2, SQL 2008 R2 Ent, 24GB RAM and all DATA(RAID5) log/tempdb/(RAID1) and backup are on separate arrays and the SAN is EMC CLARiiON) Also, most importantly, since I still have other servers which are currently in production that I have to move to the SAN, what counter should I check before moving as a baseline to compare performance? Thanks.
storagesan
10 |1200 characters needed characters left characters exceeded

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

1 Answer

· Write an Answer
Tim avatar image
Tim answered
I/O per second. You need to check the through put of the disks. You need to make sure that you get equal or better iops on the new SAN. You can have the state of the art SAN but if the disk are not configured properly you are toast. How many spindles do your luns have? How is the cache configured on the SAN? There are tools out there SQL IO SIM that can generate loads for you to test. The documentation online is very helpful for benchmarking the systems. Also [Brent Ozar]( http://www.brentozar.com) has some great material on his blog.
3 comments
10 |1200 characters needed characters left characters exceeded

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.

Thanks @TRAD. Idk how many spindls does the LUN has. I did align the disks with 1024k myself and it seems the disks configured properly cause they are working just fine with the other servers and they were working fine on this servers to until a week ago. Any specific things that I should check on components like the HBA? In the future I will check the disk through put as you said.
0 Likes 0 ·
You may want to ask the SAN admin if anything else is sharing luns on the array. The SAN is a shared resource, someone else could be impacting the overall speed of your disk. Also the fabric you are using, just because it is fiber attached doesn't mean you are getting speed of light connectivity. If someone else is saturating the switch your through put is affected.
0 Likes 0 ·
@trad I have used the open source [IOmeter][1] to benchmark the specif LUN that I am having a performance problem with and compared it with another LUN on a different server which is carved up from the same underlying array of disks and i was able to convince the sysadmins something was horribly wrong by showing the big difference in IO/sec, MB/s and response time . Upon investigation we found out that one of the two fiber cables is loosely attached to the switch and not working properly which makes the other fiber the only path to handle all the traffic and that was the reason we were taking a performance hit. Now its fixed and all is well now. (But i still didnt figure out why they were not getting alerts) Thanks anyway. [1]: http://www.iometer.org/
0 Likes 0 ·

Write an Answer

Hint: Notify or tag a user in this post by typing @username.

Up to 2 attachments (including images) can be used with a maximum of 512.0 KiB each and 1.0 MiB total.