Microsoft SQL Server Q&A
SAN and Raid Levels
Last Post 21 Nov 2011 07:16 AM by Duane Bentley. 3 Replies.
16 Aug 2011 10:04 AM
With SANS today and the data caching available, is there any need or purpose for being specific on RAID levels for database drives. I'm compelled to spec out RAID levels, but the SAN manager keeps telling me it is a waste of time and resources due to data caching. Should I not worry about spec'ing RAID?
16 Aug 2011 11:55 AM
Maybe still help for local disk activities like paging.
16 Aug 2011 04:07 PM
The caching is not an answer to everything and it is highly over rated on most SANs due to the wide number of hosts requesting data. The read cache on a SAN is almost useless for SQL Server and the write cache is what can be helpful. The problem is there usually isn't enough devoted to writes. But as with everything it depends. The key is that SAN's disk for disk will never out perform direct attached storage. If the SAN is properly configured to support your type and quanity of I/O in addition to all the other I/O from the other hosts then yes it can do the job. But the more demanding your workload the less likely one big raid 5 group will perform as needed even with the cache. So it depends on your workload. With high volumes of writes it still is recommended to have a RAID 10 for the tran logs that is not shared with otehr LUNS. The bottom line is that an answer like that from the SAN admin tells me he either doesn't care or he doesn't know what is really going on. Again not to say it wont do just fine but it all depends on the configuration and load and that is a lot of factors. Did he even ask what the load would be? Probably not.
21 Nov 2011 07:16 AM
We had the same issue with our SAN Admin. Initially they set us up on a 4 drive RAID 5. Performance was absolutely dismal at best. Then after complaining enough we got setup on a RAID 10. Performance was magnified by 10 to 25 times.
The last reply was on the money, it does matter and the more spindles per LUN the better your performance no matter which RAID level is used. I do agree on direct attached storage it is faster. All that I have seen and experienced with SAN's they are highly over-rated and to me the performance does not justify the cost. You pay a whole lot for very little. I have had DAS 14 disk RAID 5 out perform SAN 4 and 6 disk RAID 10.
But if it were up to me I would have the following (of course depending on the sizes of your DB's and companies internal infrastructure)
Below based on Direct Attached Storage:
RAID 10 14 disks 7 disk for each stripe set mirrored Active DB or tables for read / write tables partitioned out
RAID 10 14 disks 7 disk for each stripe set mirrored Transaction logging and Temp DB (depending on how heavy your temp db usage is)
RAID 5 14 disks for Read only, archive and reporting (reads are quite high and only writes should be from replication)
RAID 5 14 disks for backups
RAID 10 would give you highest overall performance with good redundancy
RAID 5 would give you the highest capacity over all and really decent read performance
Microsoft SQL Server Q&A
Acceptable Use Policy