SQL-Traditional RGs, big metalun, or combo for all luns?
On our CX700 we just freed up 25 146g 10k FC drives and need to lay out SQL luns. How about 24 drives in 4x4 1/0 and carve luns for database, tempdb, and logs on these 24 drives? I think this may be a bad idea for tempdb and logs but I'm not sure and don't have time to benchmark before the weekend. Is it ok to have tempdb and logs on same spindles in a large meta? Or will this be better laid out with tempdb and log on totally separate disks for response time?
This helped us fine tuning the sql, not just on the array side of things. We did setup the tempdb on it's own RG (2 * 146gb FC 1/0). This gave a performance increase. Our developers have not been weaned of cusors so they where dumping a lot of user objects in the tempdb.Our MDF's are on a shared r5 RG, with the LDF's on a shared 1/0 RG.
Re: SQL-Traditional RGs, big metalun, or combo for all luns?
I don't know what the temdb does, but I'd say: first try to figure out how many IOps your app needs and configure your disks accordingly. RAID10 is the best but I suppose when you're not reaching the limits of the disks anyway and need the space, you could do with RAID5 for the db.
- How many IOps do you need ? - How much space do you need ?