Spindle count vs. Speed vs. Size

[ITJohn3]

I am preparing to purchase a SAN and continue to run into an issue regarding SFF SAS drives. To put it bluntly... they are limited as of right now when it comes to speed compared to size so the question then becomes "size vs. speed vs. spindle count".

Here's the break down (sorry for the length)...

Current SAN is a MSA1000 and uses ULTRA SCSI 320 Disks (mostly 15k). The SAN holds the data for 3 ESX hosts (10 guests on each including 4 Citrix Servers), Exchange 2007 (about 150 Mailboxes), and 2 production SQL servers. All these servers get hit pretty hard.

I am proposing to purchase a HP P2000 SAS SFF which has 24 slots in the main chassis as a replacement to our current SAN. The budget won't allow for purchasing an extension at this time. No here comes the dilemma, the SAS SFF 146 GB drive are the highest disk size that is 15 k in speed. The next size up is the 300 GB drive but only have a speed of 10K. Using our exchange server as an example, having 6x146 disks in a raid10 for our information stores (not logs) would give me 438 GB. The size and speed are great but I used 6 out of 24 slots for only 1 logical drive, not great seeing how many servers I have left. But if I change that to 4x300 10K drives in a raid 10 that would give me 600GB and also gain 2 slots but sacrifice access speed and spindle count. But compared to the old scsi ultra 320 drives would that make a different. I am not sure and that is where I am getting stuck. Would using 10k SFF (small form factor) SAS drives be comparable to using to using 15k large scsi drives? What's more important.. access speed or through put? Plus would reducing the spindle count by dropping to 4 from 6 drives (for example) truly make a difference? I am looking for thought and opinions.


John Sorensen
Network/Systems Admin

http://it-john.com

 

[needcoffee]

John,

To truly evaluate you would need to know the IOPS your 15k Ultra SCSI 320 can sustain vs the other drives you are looking at and how many Ultra SCSI drives are used for each LUN in the MSA1000. As long as you are giving each LUN on the P2000 the same or higher IOP and your RAID config is the same then you should be able to figure out what size drive is best fit.

As a general storage rule, more spindles per LUN will give you better perfomance.

needcoffee

 

[kmcferrin]

Good point, needcoffee.

As to his specific questions...

Sustained throughput is going to be a function of how quickly data passes under the read head. Spindle speed is a major factor there, and all else being equal more RPMs = more throughput. The size of the disk platter will have some affect on this, as bits written on the outside of the disk platters will pass under the read/write heads at a higher rate than those on the inner tracks. So the outer part of the disk is faster.

Comparing 2.5" to 3.5" drives you would expect that at the same spindle speed (15K RPM) the outer tracks on the 3.5" drive would provide higher throughput than the outer tracks on the 2.5" drive, which they do. But one major benefit of the 2.5" drives over 3.5" drives is that due to their smaller size you can cram more spindles into the same amount of space, meaning that even while they individually provide lower throughput that they can still provide higher throughput in aggregate.

It's also important to note three other changes.

Firstly, the smaller disk platters mean that the read/write heads don't have to travel as far to traverse the disk, which reduces worst case scenario seek time (and average seek time). This can make the disks more responsive and could allow them to push more IOPS.

Secondly, the 2.5" drives tend to have more cache onboard than the older 3.5" drives. This equates to faster response times and potentially burst throughput.

Finally, the SAS drives are now coming in SAS6G formats, as opposed to be limited to Ultra320 SCSI speeds. These drives have higher theoretical performance limits, and may result in even higher performance.

All that said, you can see that there are likely too many variables to account for to allow anyone to say definitively which way to go. All I can say is that generally speaking, an array using 2.5" SAS6G drives will have higher performance capabilities than a similar array using the old Ultra320 SCSI. But you really do need to get a good idea of the IOPS numbers that you need provide to your servers before making a decision on which platform can provide that level of performance.

Especially when it comes to virtualization, far too many people end up buying SANs based on disk capacity rather than performance and end up incredibly dissatisfied with the results. If I had a nickel for every time a customer bought a cheap SATA iSCSI "SAN" to host their virtualized environment I could retire early. Don't get cheap, but what you really need up front (even if it means an expansion). Otherwise you'll end up frustrated and have to spend the money eventually anyway.

________________________________________
CompTIA A+, Network+, Server+, Security+
MCTS:Windows 7
MCTS:Hyper-V
MCTS:System Center Virtual Machine Manager
MCTS:Windows Server 2008 R2, Server Virtualization
MCSE:Security 2003
MCITP:Server Administrator
MCITP:Enterprise Administrator
Certified Quest vWorkspace Administrator