8x600gb raid 10 mdf array
4x600gb raid 10 ldf array
8x600gb raid 10 mdf array
2x600gb raid 1 ldf array
2x600gb raid 1 tempdb mdf/ldf array
READ/WRITE STATS ON CURRENT SAN
writes= 343k (as in 343,000)
RAID 1 as opposed to current RAID10 is a question mark as our current SAN uses RAID10 for everything (mdf raid 10 , ldf raid 10).
As can be seen above: DB.2 is very write intensive in both its log file and mdf I would presume. DB.s is very read intensive but this maybe left on current san.
I have been looking studies that benchmark using the same test RAID1 and RAID10.
From what i understand based on a score of 1 to 5 (http://www.pcguide.com/ref/hdd/perf/raid/levels/comp-c.html) :
RAID1 reads = 2 out of 5
RAID1 writes = 3 out of 5
RAID10 reads = 4.5 out of 5
RAID10 writes = 3.5 out of 5
If this is to be believed and performance monitor shows the ldf array with no reads but consistent writes then putting the LDF on a RAID1 array should have little performance impact making RAID1 for log files acceptable. Fine, great.
Looking at a RAID1 array for TEMPDB however we can tell by the read write stats (above) that the read write values are heavy on both counts.
Therefore moving tempDB MDF from raid10 array1 and the ldf away from raid10 array2 to merge on a raid 1 single array on a new san then becomes a question of the how well tempDB READS perform on RAID1 (as writes would not be a problem as literature suggest raid1 and raid10 writes speeds are similar).
In short under proposal 3 my TEMPDB reads may suffer but would it be better or worse than contended reads when sharing disk space with a write intensive database like db.2 ???
This is the final question in need to answer.
Now bearing in mind on the current san db.s currently hosts 315m reads which is hosted on the same drives (mdf/ldf array) as db.2 and its relative 341m writes.
In this context relative stats from tempdb 69m reads and 39m writes should not have nearly as big an impact as db.s reads could possibly have. And it is perfectly possible that the drop from raid10 to raid 1 will have an impact on the write performance of tempdb.
So again i refer to this article regarding the isolation of tempdb. http://blogs.msdn.com/b/cindygross/archive/2009/11/20/compilation-of-sql-server-tempdb-io-best-practices.aspx
Q: How can i gauge of the drop from RAID10 to RAID1 but isolating tempDB on its own array is better than sharing a RAID10 array with a database that has 315m writes but little reads ? (assuming db.s remains on current san)
tempdb 69m reads on raid1 isolated
tempdb 69m reads + db.2 36m reads on a shared raid 10 array.