1

I want to set up a virtual server running windows server 2008 64-bit. I need a motherboard that supports dual processors and holds a lot of RAM. I also need 2 network adapters so I can use one for the host server(Win2K8) and the other for the virtual network.

Can anyone give me a list of what the best components would be for this project? I am really concerned about getting a good motherboard that isn't terribly expensive. Can anyone recommend any specific models.

I am also not too sure what type of hard drive configuration to use. What would be the best configuration to use for this application and what are some high quality drives? How important is the speed of the drives?

As far as processors go, what would anyone recommend? I know it's going to depend on the motherboad I get, but this could definately influence that decision. Can anyone recommend anything in particular?

7 Answers 7

5

"Good" and "Not Expensive" don't usually belong in the same sentence together...

That said, here's some general pointers:

1) If you're on the cheap, go for an Intel Core 2 processor. Dual or Quad Core, whichever, depending on how many VMs you want to run. Choose the ones with the biggest cache you can afford.

2) The other advantage of running a Core 2 is that RAM (DDR2) is exceptionally cheap. You need lots of it. Lots and lots. 8Gb would be a good number to start at. Note that because this is on the cheap, there will be no error checking (ECC), so it's not really that ideal.

3) Disk I/O is never going to be as good as it would be in a native solution, but it all comes down to what you want. Highly Redundant and Super Fast? RAID10. Massive storage with low cost? RAID5. By spending $200 on a motherboard you should get one with on-board RAID, but don't be suprised if it doesn't do RAID5.

4) Steer clear of massive disks. My most recent app server has 5x 1TB drives, and I had a disk fail the other week. 1TB is a lot of storage to risk. Again, go for big cache over capacity.

5) Most good motherboards will have dual NICs, however if they don't, any half-decent PCI nic should do. If budget permits, get a dual nic card and use redundancy.

For the record, the app server I recently deployed using desktop components is:

Core 2 Quad 2.8Ghz
8Gb DDR2 RAM
5x 1TB RAID-5 (SATA-150)
Adaptec RAID card

And it runs quite nicely. On the flip site, the last app server I deployed with server components is:

Dual Quad-Core Xeon 2.4Ghz
24Gb FB-DIMM ECC REG'd RAM
3x 74Gb SAS Disks
iSCSI remote disks

Budget permitting, I would choose the 2nd server any day.

4
  • I think point 5 is misguided. The risk of having a disk fail does not go down with smaller disks. However, the more disks you have, the more likely one of them is to fail. Thus for a given amount of storage capacity, using larger disks is possibly lower risk. Your storage redundancy plan needs to address the number of disks you can risk failing at one time. Thus, for a given amount of storage and a given RAID level (i.e. same number of failed disk support) you actually are protecting less data using more disks. Not to mention, RAID is no replacement for proper backups.
    – Burly
    Feb 16, 2010 at 15:11
  • I think you mean Point 4. Since I wrote this post I've had 3x 1TB drives fail, and none of my 500Gb's which were all about 6 months older, but each to their own :) Feb 16, 2010 at 20:44
  • Ahh, yeah, I do mean point 4. Your 3 or 4 drives (or even 400) are not a representative sample of the entire population of drives out there, so your situation, while unfortunate, is just anecdotal. At most, your situation may demonstrate a bad batch of hard drives or a firmware issue in the RAID card in dealing with large drives, etc - not a fundamental reduction in reliability of RAID with larger drives.
    – Burly
    Feb 20, 2010 at 2:53
  • @Burly No, failure rates for drives that have been available show that drives over 2TB were significantly more prone to failure than those under. No doubt this has or will change as they become a more established tech however there is always the bleeding edge. Also MTBF has not risen with in line with size so in a RAID rebuild larger drives are more prone to corruption.
    – JamesRyan
    Apr 23, 2014 at 16:34
2

Given the budget concerns, I would suggest "Last Season's Model" of server from the likes of Dell or HP.

By watching the various dealsite RSS feeds for "Small Business" related feeds, I have found that these standard servers cost less than roughly equivalent servers I have built myself, and they require a lot less time to install. My only complaint with the stock servers is the fan noise, which in a few, very limited situations, has been a problem. Also, waiting for a manufacturer's special can be annoying, but sometimes one has more time than budget.

Farseeker's suggestions for App servers seem pretty good.

1
  • Ex-Lease or Ex-Demo servers are usually perfecly fine and can be had at a fraction of the price of a brand new one. Jun 4, 2009 at 4:49
0

Hard drive speed is very important to I/O. If you google enough, you can find benchmarks that clearly prove that HD access times are critical in virtual environments.

If performance and budget are most important (and fault tolerance no so much), I would get 4 10K-15K RPM hard drives and create two pairs of RAID 1/0 (You can get more performance with RAID-0 but you should consider regular automated image backups in case of HD failure).

One RAID volume should hold the Host OS (and maybe some low-priority VM's) and the other RAID volume should hold the VM's.

As for the HD brands. As of this date, the best price/performance would be either a traditional 74GB WD Raptor (very inexpensive but fast and reliable drive) or a 150GB WD Velociraptor (slightly faster, bigger, costs more). The next level up from there would be Fujitsu SAS drives, which offer slightly better access times, but require a SAS RAID controller (which doesn't come cheap).

As far as good 'affordable' server boards: SuperMicro, TYAN, and Intel all make solid affordable boards.

0

You probably won't like this answer, but my feeling is that for something like this its best to just buy something off the shelf that the virtual server software you are using supports. I know price is a concern and this sounds like the most expensive way of doing things, but imho you'll end up paying the same or more over the lifetime of the server anyway, the only difference is that with a pre-built server you're paying more of the cost upfront rather than later on.

VMWare, for example, have an extensive HCL for ESX Server type products, and in terms of making sure all your components are on it, it's easier just to buy a server that is on their HCL rather than plough through all the components individually. The comment in another answer about buying 'last year's model' of server is a good one I think.

If you want to build your own, then besides checking HCLs, you need to ensure that the processors support virtualisation directly, and use as many disk spindles as possible to improve IO throughput.

0

Here is my current list of components for my next build. I have documented the reasoning for each option. This is a single processor machine (I7 920). For my purposes I do not think my Hyper-V workload will be CPU-bound.

0

I'm not sure what you really mean by a "lot" of memory, but I assume it's well over 4 GB.

The cheaper solution is to go with an i7 system; ASUS P6T motherboards have six slots, giving you up to 12 GB of memory. That's not dual-CPU, but it is 4-core with two hyperthreads per core, which might be enough for your needs. (Yes, the hyperthreading really works.)

The next step beyond that is something along the lines of a Supermicro X8DTU-F, which has twelve slots, for up to 24 GB of unbuffered memory or 96 GB of registered ECC memory. This will give you two CPUs, each with 4 cores and hyperthreading.

I am currently using systems with both of these motherboards, and I'm very happy with them.

0

Just buy a HP DL380 G6, they're the BMW 3-series of servers, well engineered slighty overpriced but hugely capable for 90% of usage cases, and can be ran into the ground over a decade or so.

That's why they're something like 40% of all servers sold worldwide, they're a no-brained unless you're on a very tight budget.

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