SAN Strategy, SQL & Exchange, Dell MD3200

[Samalie]

Hi everyone,

For background, we are a SMB with <50 users.  All the users are on our Exchange server (Database currently approximately 100GB, WAY TOO LARGE, but I digress), and 25 of the users are on a SQL based database application.

I've bought a pair of Dell  R510's and a MD3200 SAN, which will be direct-attached to the R510's via 6Gbps Fibrechannel.  Both physical servers will have 4 connections to the SAN.

The SAN is 12 disks, all 300GB 15K drives.

Now, I know in terms of total IOPS that the MD3200 will handle my needs, but I want to optimize the creation of LUM(s) as best I can for performance, and quite frankly, I'm really just not sure.

I've thought about doing 3 4-disk RAID 10 LUMs, but that leaves me no hot spares.  I could go 2 5-disk RAID 5's with a hot spare per RAID, or just one big giant RAID 5 LUM with a single hot spare.

All of this will be used in a v5 VMWare environment.

Thoughts, opinions?  And my thanks in advance for your thoughts

Kevin

[Samalie]

Oh, and to add (sorry) - there will be many more VM's besides the SQL/Exchange spread accross the 2 physical servers, but their IOPS needs are minimal at best (AD, BES, File Services, WSUS, etc) so I'm not overly concerned about them.

Thanks again,

Kevin

[JohnADCO]

You know?   You will get canned answers.   But I have some experience with these MD arrays.

In my opinion.  For your type of load?    I swear they run best with as many spindles as you can get in one group, then doll out the multiple LUN's / Datastores you need from there.

Most are going to have you do multiple smaller raid sets,  splitting up log and data locations for your SQL / Exchange?

I say no way...   We run a similar, maybe a littler meaner load than you, and we ran it on MD3000i's for 5 years.   Now I have setup and maintain a couple of MD3220i's for some non-profits.   So there is my experience level, if your curious.

[Dave_Mishchenko · Accepted solution]

I'd put all the drives into a single array (minus what you allocate as a hotspare).   That way you get all the disks contributing equally to I/O.

Splitting I/O into data / logs isn't a bad thing, but practically it's hard to do and one tends to end up with some disks that are too busy while others are idle.

[murphyslaw1978b]

I agree with this approach.  It's called wide striping, and it's a common way these days to do mixed-load SAN environments.  The more platters you have, the more horsepower you have access to across ALL the VMs, not just certain ones.  If you find you have an application that is slow, then you may have to add more spindles, cache, or controllers. 

[Josh26]

Samalie wrote:

Hi everyone,

For background, we are a SMB with <50 users.  All the users are on our Exchange server (Database currently approximately 100GB, WAY TOO LARGE, but I digress), and 25 of the users are on a SQL based database application.

Kevin

I would actually consider that an incredibly small Exchange database for that number of users.

Regardless, this sounds like a fairly small environment for that level of SAN. The general rule is that more spindles = faster, so there's no reason to create more than one RAID array. I would be incredibly surprised if you weren't happy with the performance of a single RAID 5 array. If you're concerned, a single RAID 10 makes a lot more sense than 3-4 RAID 10 LUNs.

[Samalie]

Thank you very much for your responses...I appreciate the assistance

Kevin