I do IT infrastructure support primarily for small businesses (up to ~50 users). The past few years I have been doing some server virtualization for situations that don't require a second physical machine, or for consolidation when it comes time to replace old physical servers.

90% of these environments are something like a Windows SBS server and another 1 or 2 Windows servers, usually to run some line-of-business application, accounting software or whatever else doesn't play well with SBS. Most of the businesses are something like 10-25 users.

My go-to server configuration for these businesses is usually:

  • Dell T610
  • Dual procs
  • 12-48GB ram
  • PERC 6/i or H700 RAID controller
  • RAID 1 with 2x 15k 146GB SAS drives
  • RAID 5 with 3x 15k 450GB SAS drives (maybe 600GB drives depending on needs)

I put the host OS on the RAID 1 (ESXi or MS Hyper-V) and the VMs on the RAID 5. I like the T610 because it has 8 SAS slots, so I can add another RAID 5 with 3 drives in the future if more space is needed.

My question is regarding the drive configuration. Is there another configuration that would be better than the RAID 1/RAID 5 config above for these situations? Like drop the RAID 1 and use those slots more efficiently for something else?

This config has seemed to serve me well, but it seems like I could be limiting my I/O performance if I add more than a couple VMs to the host (unless I add another RAID 5 to gain more spindles).

I know there are reams of questions similar to portions of this question...I've just spent a few hours reading ~50 of them, but I haven't been able to come to a conclusion on this.

Any input appreciated. Thanks in advance.

  • how are the drives connected into the controller ? do you use both channels ? – silviud Apr 20 '11 at 1:31

Personally I'd drop the mirrored drives and grab as many of the larger drives for the RAID5 as the budget allows. The array is almost always going to be faster than the mirror (with a good RAID HBA) and a well installed/configured ESXi/Hyper-V Server isn't going to need much for the base OS. Of course this is all very generic, but so is the Question..,.

Also, RAM is like money, can't have too much; and more cores, most software is licensed per CPU, not core. And if you need more disk performance, RAID10 is usually faster than RAID5 and larger HBA BBWC makes a big difference.

  • +1 for RAID 10, but you do lose a fair bit of space (the inevitable compromise). – gravyface Apr 20 '11 at 1:46
  • So you suggest dropping the RAID 1 and making 1 RAID 10 with 4 drives (for example)? Then I would use part of that RAID 10 for the base OS and the rest for disks for the VMs? The reason I put the base OS on separate spindles was for performance, but it sounds like that's a marginal gain versus having it all on 1 array? – user78940 Apr 20 '11 at 3:06
  • If the Host OS is going to be a full Windows deployment, separating the spindles might make sense (depending on what the host OS is doing). If the Host OS is just Hyper-V Server or ESXi it'll barely use the disks after it's booted, no real point in dedicating any spindles. I highly recommend using the latter method if the situation allows it, as you'll have the smoothest virtualization experience. – Chris S Apr 20 '11 at 12:29
  • @user78940 +1 RAID 10. Drop the OS drives, put everything on the RAID 10. RAID 1 is a waste. The hypervisor is running from memory (in the case of ESXi) so the only thing the drive is for is writing out config changes and logs. ESXi will also run on a USB flash drive, if you are hell bent on keeping it separate from your datastores. – JakeRobinson Apr 20 '11 at 14:33
  • @Chris S, @JakeRobinson - The host OS is just ESXi or Hyper-V (no full-blown OS). I'll skip the RAID 1. Is there anything that needs to be considered regarding the ESXi swap file without the RAID 1? Also, is there any benefit to running ESXi on a built-in SD card that's availabe on some of the Dell servers? If there would be some sort of performance gain, I'd do it, but if it exists just for manageability reasons I'd pass in this case and load the host OS on the same drives as the datastores. – user78940 Apr 20 '11 at 15:14

The short answer is that using a two-disk RAID1 with SAS drives just for the host OS is a waste.

The long answer is that it would be worthwhile to consider an alternate strategy.

I have deployed configurations like this using ESXi and direct attached storage, but hate it more and more. You lose a lot of flexibility and management control on the storage side. Even with the best controller, you won't be easily able to remotely manage the card. You'll be forced to shut down all your VMs and reboot just to check basic health.

My current strategy is to simplify the server: just go bare bones, minimal drive bays, and no fancy controller. Run the ESXi hypervisor from CF or USB or some SATA drive. ESXi rarely writes to its own volumes, so the media will last a long time. This hardware change will save a lot of costs; all you need is 1U, lots of RAM and CPU and that's it.

Then money you save on the server should be put into a SAN. QNAP has some pretty affordable stuff (SATA only) but there are lots of other options as well. They all have much better network management, hot adding capability, thin provisioning, reporting, and so forth.

Central storage is a driving concept behind ESXi. Especially if you have multiple physical servers you will save a lot of time on management, and have less downtime.

[I'm new here - sorry if I broke any rules]

  • I don't think you broke any rules but I also don't think the SAN approach is realistic for the very small companies the poster has indicates he supports. If these companies are small enough to have just 1 or 2 servers total, I can't see how the complexity of even a tiny SAN makes sense. – icky3000 Apr 20 '11 at 3:25
  • Tiny SANs don't need to be expensive, or complex. Infortrend, for example, offer fairly well-priced systems that don't require you to buy vendor-marked-up disks, and they perform pretty well. If there's zero chance the customer will never expand past the need for a single server then a SAN seems like overkill, otherwise I think it just makes sense. – Daniel Lawson Apr 20 '11 at 4:40
  • The likelyhood of them needing more than 1 physical host is very slim due to the company size. Also, cost is always a concern with businesses this small, so $8k-10k is about all they will stomach for a new server usually. Also, purchasing hardware that is 100% supported by the manufacturer (Dell) is a requirement so it's covered under their onsite hardware support plan, which usually means buying a Dell-branded hardware product. – user78940 Apr 20 '11 at 15:06
  • Dell onsite warranty adds a lot to the cost. Having the external mini-SAN saves a lot on the cost of the main hardware. And it will really reduce the maintenance costs, since storage upgrades and restructuring will be much easier and require less downtime then with traditional direct attached storage (DAS). I forgot to mention that using an iSCSI mini-SAN opens the door to better backup tools such as Veeam, which will generally not work at all with DAS. – Wim Kerkhoff Apr 22 '11 at 3:03

I would like to know more about the customer(s) you are installing at, specifically how often the server(s) are monitored and what the backup policies are like.

Personally I wouldn't stick data on a RAID5 array at a small business due to the frequent neglect that the server(s) may see. RAID5 only allows for a single drive failure and due to the tendency of drives to fail at close intervals combined with the fact that the server may not be monitored frequently for issues, this could lead to data loss. I can't count how many times I've seen a server lose two drives within a week after having no faults for 2+ years. Not good in an environment where no one is checking the "blinking red lights".

I would prefer RAID6 or possibly even RAID10 for their higher fault tolerance. Especially in lue of superior backup policies.

Having recently done a full scale SAN deployment at a small business I would tend to disagree with Wim Kerkhoff. You should only be looking for SAN storage if you need host clustering (using two or more VM hosts to provide fault tolerance at the host level). Correctly architect-ed SAN deployments are not cheap as they require additional switch ports, NICs, cabling, much more configuration/troubleshooting time, and often cause you to be locked into a vendor's marked up drive prices.

If you do go for a SAN, keep in mind that booting up 5+ VMs on a SATA based SAN will take AGES (we had to reevaluate and purchase some SAS drives after discovering just how slow SATA really was in a VM environment).

I agree with the idea of using internal flash storage for the hypervisor. Some of the higher end dell machines come with a RAID1 compactflash card setup which is very nice - not sure if that is an option on the server you are speccing. I would be a little scared running the hypervisor off an external harddrive or USB flash memory however.

As always, you need to evaluate the cost of the options vs the cost of the downtime for your specific application.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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