February 16, 2006 at 2:13 am
we have had our main production system in a single filegroup throughout its life
we have tuned it to hell, it's pretty quick for our purposes, however we are making some key changes and we think this requires some reconfiguaration of the filegroups and disk systems - so I hope i'm in the right forum because we are doing this for performance enhancements down the road...
the database is NOT large, about 50Gb, and we estimate growth over the next period to be a further 100Gb, we have some very large tables in terms of rows, but they are small in terms of data kept (usually 40ish fields of which 10 may be codes in small VCHAR(12)s, and the rest INTs)
we are proposing to install a new raid 0/1 set based on two 320GB SATA drives - this will result in 1 set,
will we see any performance gain if we partition this raid set into 4 disk-partitions and use each for filegroups
p1: 100Gb - main read table (60 million rows, grows 2 million per quarter), + countTable (360 million small rows, not partitioned, but higly indexed - nb this currently reads back from SPs at under 0 secs for a count, so we're to worried 8) )
p2: 100GB - indexs for main read table and count table
p3: 80GB - main look up (20 million records), minor look ups (5 tables, 200 rows)
p4: 40GB - Indexes for look ups
will this work?
have I asked the right question?
February 17, 2006 at 9:59 am
It's hard to say without performing actual tests, but I suspect that the separate partitions won't help nearly as much as more physical disks. Assuming your disk controller(s) are up to the task, more spindles results in more parallelism. I would think four or six smaller disks in two or three RAID 1 mirrors would do better than a single pair.
Viewing 2 posts - 1 through 1 (of 1 total)
You must be logged in to reply to this topic. Login to reply