Hello JHunter, thanks for this quick reply.
The spacerequirements are about 300gb for primary db and 150gb for secondary db, at the moment but we're expectinga 50% growth in the next 6months. My thoughts are just to allocate all available space according to the spacedistribution
As you suggest regarding the tempDB on it's own is a good idea (also best practice)... but I've many theoretically suggestions not to follow that 'best practice' - also why the database and indexes should be placed on the same lun(s)
Let me try to explain my thinkings regarding db and indexes mixed up. This only applies to random reads and no writes (the tempDB has probably another setup, which I don't yet have had time to think through)
Lets concentrate onto datafiles and indexfiles...
If we configure 2 luns each with 4 disks RAID 10, you've around 300 IOPS available at each lun. We then do a query against a table with a clustered index placed at primary and a nonclustered index at an index filegroup, lets call the filegroup FG_INDEX. The query runs and uses the nonclustered index (this index is about 10% of the total tablespace) and reads the data located on primary filegroup. It cannot read any faster than 300IOPS.
But what if we mix both the index and datafiles, would we then get 1.95 x 300 IOPS? Ofcause this is theoretically, but then lets say 20% overhead = 1.75 x 300 IOPS, which should be 75% faster than using dedicated LUNs for datafiles and indexfiles.
The primary goal would allways be to distribute the workload to all disks - right?
And is'nt the whole idea using SAN to let the HW distribute IO's even between the disks?
The issue you mention regarding separate shelfs, is actually something I've been wondering alot about. I took the liberty to create another thread, specifically for this issue (Is 'Best practice' really that old)
Why should the OS disks be dedicated?.. it costs a lot of money to dedicate 5 disks for this, do you know how much the OS requires in space btw?