Performance hit when moved to 550GB drive 1

[vnad]

We are on a EMC Symmetrix and have been running great for over a year. We were running low on space, so we added a 550GB drive. When we copied our SQL Server database over to the drive, we lost about 40% in performance from the drive we were on (250GB) drive. We are using only one HBA (LP8000) to see the two drives. We saw the performance hit when we moved the database, not when the server was able to see the new 500GB drive.

We are running Advanced server 2000 with a SQL Server 2000 database. We were told that the new volume is exactly the same as the old drive we were on.

Any ideas?

 

[Chacalinc]

did they config the symmetrix with metavols? are they using striping or not? I think they made a mistake with the metavols config... you must see the disk drive config v/s metavols config (symm back-end).

Cheers.

 

[vnad]

thanks for the quick reply

They are using striping for sure and I think they are using metavols. What are the common mistakes made when configuring this?

 

[Chacalinc]

not using all DA buses, or using the same physical disk drives for the same stripe set (metavol), or creating metavols with no striping, or using striping with element size bigger than 2 cyls ... etc.

I would make a hardware map (disk drives v/s hypervolumes), with the old metavol and the new one and make a comparation. May be it better to have 2 metavols and not only one, so SQL server can use 2 drive letters, not just one...

Best regards.

 

[vnad]

Are hypervolumes only used when the disks are setup to do parity? We are currently on mirrored disks.

 

[Chacalinc]

No, hypervolumes are the partitions inside a disk drive. 1 disk drive could have from 2 to many hypervolumes (microcode dependant) and it can be protected either mirror or RAID S.

Example, a 18 GB disk drive can not be configured with only one volume of 18 GB, it must be configured with minimum 2 hypervolumes of 18.414 cylinders (or 9 GB).

 

[vnad]

Found out it was that the metavolume was on the same set of disks twice/three times. The original disk we were on, did not "loop" back on the same disks. Would this cause a large performance decrease? What is the typical size for a hypervolume?

 

[Chacalinc]

1. Would this cause a large performance decrease?

yes, keep in mind when you create stripe set, it is in order to spread the load in several disk drives, so you get better performance. If you create metavol using the same disk drive the symm box must to do seeks in one sector and in several differents at the same time in the same disk drive, so for example: in order to read a file (or block) of say 2560 bytes (512 x 5 sector) and you have an stripe size of 512 bytes you -ussualy- would read the file spreaded in 5 disk drives, if you have a metavolume, conformed of 5 hypervolumes but the hyper are in the same disk drive.. you actually must read the same disk drive 5 times!!

2. What is the typical size for a hypervolume?

It depends of requirements and disk drive size. There are no rule of thumbs.