5

We are deciding whether to use SATA or SAS aggregates on our Filer for our VMWare datastore. We will be using NFS to connect the VMWare hosts to storage.

Doesn't seem to make sense to use SAS disks that have 6GB/s pipeline (15000 RPM) if it's going via NFS over gigabit network (1/8 GB/s). We are teaming the NIC cards but that's still 1/4 GB/s (in ideal conditions).

The SATA disks are 7200 RPM and, according to specs, 3GB/s.

Perhaps someone with more real-world experience could check my logic.

Additional information

  • Both SAS and SATA aggregates have the same amount of disks.
3
  • Something to think about - ZFS with Hybrid Storage Pools. You could then combine SATA with SSD at a lower cost than SAS with potentially better performance.
    – hookenz
    Mar 1, 2012 at 18:28
  • Wait, are you building a filer?
    – ewwhite
    Mar 1, 2012 at 19:13
  • Nope. NetApp filer already setup. Mar 1, 2012 at 21:48

5 Answers 5

10

The benefit of SAS is the number of IO/s they can do compared to SATA (or midline). SATA drives rotate more slowly (7200 RPM) and thus have a higher read latency. This is made worse by the fact that you'll have more VMs running per drive because of the high density of space. SAS drives that run at 10k or 15k RPM will have a much higher number of IO/s they can do per spindle.

You can put some loads on SATA, but I would use SAS for anything that can't afford slow response times.

5
  • Even if we're using NFS? Mar 1, 2012 at 18:08
  • 1
    @BeamingMel-Bin Even if.
    – sysadmin1138
    Mar 1, 2012 at 18:09
  • 1
    NFS is a way of letting your servers query the data on a disk. It's just a protocol, and doesn't have any effect on the underlying storage. So yes, even if.
    – Basil
    Mar 1, 2012 at 18:16
  • Okay. My thought was that, no matter how fast the storage it would be negligible because of the Gigabit. However, reading through the answers, I see how I missed some (now obvious heh) parts of the equation). Mar 1, 2012 at 18:18
  • Common misconception. Normal VMWare operations aren't gated by network speed, the bottleneck will be on the disk. Or it can be, at least.
    – Basil
    Mar 1, 2012 at 18:19
5

15K drives will support higher random I/O operations per second than 7.2K RPM drives.

15K drives supporting random I/O should only saturate GigE if there are enough of them to spread the load. You don't mention how many drives are going in here, so it's hard to say how far it'll scale for you.

A blended solution is actually not a bad plan. Some SAS, some SATA. Keep in mind that drive performance has more to do with drive count than drive size. For some workloads using SATA makes perfect sense, while other more databasey workloads really should go on the faster SAS disks.

Most people end up having to make a judgment call between filer size and filer performance. This compromise usually dictates the mix, if any between 15K and 7.2K RPM drives.

1
  • Both SAS and SATA aggregates have the same amount of disks (14 right now). Mar 1, 2012 at 18:12
5

You should have an option for nearline/midline SAS disks as well. They are roughly equivalent to the SATA drives mechanically, but uses the SAS protocol. That should be the choice if you're concerned about capacity...

But with regard to performance, storage is rarely about maximum throughput speed. It's more often centered around IOPS random operations. So even though you're serving NFS over gigE, the VM traffic should always benefit from faster disks.

1
  • Good to know. Glad I asked even though my initial thought was wrong. Mar 1, 2012 at 18:15
4
  1. SAS handles varying concurrent high loads better than SATA because its queue management is far more efficient than SATA's NCQ.

  2. Most if not all SAS disks are designed for a 100% duty cycle, whereas few SATA disks are, most only offering 30%. This could massively affect your MTBF/reliability.

  3. I'm personally no fan of NFS for working with vSphere, your mileage may vary but we couldn't live with its limitations and performance.

  4. vSphere doesn't support LACP so your teaming is unlikely to work as you imagine.

2
  • Off-topic but: What limitations are you hinting at? Mar 1, 2012 at 19:57
  • In particular you can't have RDMs on NFS, so multi-host MS clustering won't work - same for anything else that needs RDMs.
    – Chopper3
    Mar 1, 2012 at 20:43
-1

Another cheap option is to take 7.2k rpm SATA disks and only use 1/4 of the capacity (thus matching the more expensive SAS capacity). This will make up for the lower RPMs.

I would obviously do this with an array that you can easily hot swap on since you are not using enterprise drives with higher MTBF. I also assume you don't need hot swap on your raid controller cards.

6
  • -1 Using whatever portion of the disk makes no difference. The simple fact is that on a faster disk it takes less time for the platter to rotate under the head. The shortened head stroke will make almost no difference on a modern disk with a voice coil actuator.
    – Chris S
    Jul 11, 2012 at 19:20
  • zdnet.com/blog/ou/how-higher-rpm-hard-drives-rip-you-off/322 It does make a difference.
    – rbeede
    Jul 12, 2012 at 1:53
  • Of course price wise now a days you're better off with SSD over mechanical SAS if you really need the heavy random I/O workload performance.
    – rbeede
    Jul 12, 2012 at 2:03
  • Also if you short stroke your SAS drive by a bit you could also jump over any benefit of short stroking a SATA drive. It comes down to the work load and how much you want to pay for performance.
    – rbeede
    Jul 12, 2012 at 2:08
  • Yes, a 6 year old article written for "enthusiasts" with absolutely no actual test results has convinced me that storage administrators and manufacturers have been ripping us off for years.
    – Chris S
    Jul 12, 2012 at 4:16

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .