Selection of Storage array and its configuration.

[Ninad_1]

Hi,

Are there any steps/methods in selecting a particular storage array and then determining what and how much components like cache etc required ? How is usually a storage array selected to give good performance. Always the DBAs complain saying that the database shows high IO waits and only method is to have a good storage subsystem - how do we generally approach ?
Currently I have a system running using XP256 array - I want to determine what would be a good solution giving better performance. Please let me know if I need to post any data.

Thanks,
Ninad

P.S. This thread has been moved to General to Storage Area Networks (SAN) (Enterprise). - Hp Forum moderator

[Ivan Ferreira]

Your DBA should inform you about the TPs, you can also capture information about the current disk I/O. Then, select a storage based on the specifications that match your I/O load.

Fibre arrays provides better performance that SCSI arrays. RAID 0+1 are the best candidate for database (availaiblity and performance). The more disks you have in an array, the better the performance.

Virtualization technology also increase the performance of the arrays.

I don't know the specificic characteristics of this array, but you may improve performance just modifying your current disk layout.

[Ninad_1]

Ivan,

Yes I understood the general concept - need TPs and disk IO statistics. But can anyone tell me
1. what would be the steps involved - like say use iostat to check the bps and sps for all the disks , add the figures to get total data throughput and no of IOs etc etc.
2. Then how to use these values to select a new storage array - what do I need to check ? which metrics of new arrays ? how many disks of what speed ? how much cache ? etc etc.

Above are the only things I could think logically - Please help me in understanding the procedure/process that goes into the selection - step by step to select a storage configuration to give comparable/better performance than existing.

Thanks a lot,
Ninad

[Alzhy · Accepted solution]

Ninad,

Peace!

High I/O waits on a performing Database server is not generally bad IF you see no queing of each individual luns via "sar -d".

IF you have highIO waits and you see no queuing on the individual components luns/disks making up your database storage volumes (raw or filesystem based) - then the problem is NOT with the storage but how storage is accessed by the Volume Manager, FIlesystem (if used) or Application (SGA sizing, placement of files, tuning). This can most often be remediated by close coordination (and literacy) beteen the admin the dba. You need to look at possibly spreading I/O to more luns and relocating hot storage spots so its sprad out. It may also mean tweaking SGA so the instance has more buffer cache, etc...

You ask - for sic. "steps/methods in selecting a particular storage array and then determining what and how much components like cache etc required" ?

There is unfortunately no set process as its a "hit and miss" or "start with this config" kind of thing. You can however consult various resources on the web, the vendors and compare their published performance metrics, connectivity, scalability, feature sets, etc.. Or if you can pay, ask for "independent" assessments from such storage information vendor like the Evaluator Group.