Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

How do I determine the optimal stripe size for RAID 10 for mail stores and the transaction logs for Exchange 2003?

I am planning to have 2 RAID 10 arrays for my exchange setup. For the data store, 10x 399 GB disks and for the transaction logs 4x 399 GB disks. My stripe size options are from 2^2 to 2^7 KB in increments of powers of 2. The file system will be NTFS.

share|improve this question

2 Answers 2

up vote 2 down vote accepted

Microsoft gives you a lot of good information here: http://technet.microsoft.com/en-us/library/bb125079.aspx

Off the top of my head, I believe you want 32KB stripes and you'll want the partitions aligned to the stripe boundaries. Don't take my word for it-- read the guide. I can be an idiot sometimes. (smile)

Edit: I said 64KB first. It should be 32KB. See-- I'm an idiot.

Also see: http://blogs.technet.com/evand/archive/2004/10/14/242127.aspx

An aside: Reading Evan Dodds posting again made me recall that I did see an adverse performance situation on an Exchange 2003 installation where the SMTP queues were put onto the same spindle as the transaction logs. I recall you saying that this box is going to be a back-end server, but it will, obviously, still be receiving incoming email from your front-end servers into SMTP queues. The situation I recall is much the same as Evan Dodds described-- the SMTP queues were on the transaction log spindle and the spindle ended up thrashing between trying to do sequential writes of the transation logs and random writes on the SMTP queues. Moving the SMTP queues to a spindle of their own ended up making the box a lot happier. (It got the disk queue length way down on the transaction log disks, and decreased the user-visible symptom-- RPC timeout popups galore!) Watch out for that.

share|improve this answer
    
What about having the SMTP queues on the database partition? –  Kyle Brandt Jun 3 '09 at 13:22
    
Evan Dodds article says that the access profile of those queues is closer to the database than anything else. We put them onto the same spindle with the operating system in the situation I described and things got better. I didn't try putting them on the database spindle (and since things were better, we left well enough alone). It's worth a shot, certainly. –  Evan Anderson Jun 3 '09 at 13:35

Also take a look at the following article on Disk Partition alignment for RAID volumes. While the article says it is for SQL, it is really for any database running on top of RAID with Windows Server 2003 or earlier. You can increase the performance of your setup SIGNIFICANTLY (increase throughput and IOPS by as much as 30% while reducing latency) by making the change (but it has to be when you first partition your volumes, before they have been formatted):

http://msdn.microsoft.com/en-us/library/dd758814.aspx

more here: http://blogs.msdn.com/jimmymay/archive/2009/05/08/disk-partition-alignment-sector-alignment-make-the-case-with-this-template.aspx

share|improve this answer
    
Yep. That's what I mean by having the partitions "aligned to the stripe boundaries". –  Evan Anderson Jun 4 '09 at 2:39

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.