2

We’re planning on going virtual from this point forward and our plan will be 1 server, 1 task, so while we are a small company we're going to have a high number of servers for our size, but this should make it easier when one of them has an issue since it will no longer affect other software that is on the same server. We’ve never gone virtual in production before hopefully I can get some guidance & help for planning this.

We’re going to have the following servers running on it.

  1. Exchange 2007 (100 users, max 30 active at a time, not likely growing)

  2. Symantec Anti-Virus Server (Anti-Spam & Anti-Virus for Email, about 5,000 a month blocked)

  3. BlackBerry Server (10 users)

  4. Domain Controller

  5. External Website (low traffic)

  6. Proxy Server

  7. 2 or 3 other low use servers for testing (these and all others will be Windows)

The server is a Dell 2.5 GHz Xeon with 8 gigs RAM and a PERC6i controller using Dell Near Line 7200RPM SAS drives. Our host will be ESXi 4.0 due to it allowing memory over commitment.

First Question, can we do V2V from ESXi to Hyper-V in the future? We’d like to go with Hyper-V in the long run but not until it can do the memory over commitment. We want Hyper-V in the long run because then if we have an issue with the hypervisor or the guest OS, it’s all the same company to deal with for support. Not being able to V2V we may end up going to Hyper-V right away.

Second Question, what is the best HDD configuration?
I can have up to 4 drives in the server, but the budget is tight.

Option 1 - I was thinking initially RAID1 which I know isn’t the best, but leaves me 2 empty bays for later expansion with another RAID1. When I expand with the second RAID 1, I’ll split the heaviest use servers between the 2 RAIDs.

Option 2 – RAID 5 with 3 Drives, (I can likely get this in the budget) which gives me more spindles right away but I’ll likely have to blow away the array (or at least full backups of all the VM’s) if/when I add the 4th drive. Will this really give me better (or worse) performance then a RAID1?

Option 3 – Get 4 drives now. I know this is the best option, but I need to give a good case to basically double the cost of the HDD’s in the servers, and it’s not likely but I have to ask. If I do manage to get this, what’s the best RAID based on what I’m planning on doing with this? I cannot use 10k SAS drives, the price jumps too much, so the near line SAS is the best compromise we could go with between full SAS and SATA.

Thanks in advance all.

EDIT:So for those of you thinking this is the entire shop it's not. This is the current setup.

Server 1 - Dell 840 (2 gigs, RAID1 SAS5/iR)

  • File Server
  • Print Server
  • Domain Controller

Server 2 - HP something really old, but still running strong (new hard drives though)

  • Domain Controller
  • Symantec Anti-Virus Console for the Desktops
  • Local Backup Server

Server 3 - Dell 1850 (4 gigs, RAID1 PERC4i)

  • Exchange 2003
  • Symantec Anti-Virus/Spam
  • BlackBerry
  • External Website (as you can see this is basicly all the 'net stuff)

So the new server would come online and we'd migrate everything off of server 3 to the new one, splitting everything into it's own operating system as we go. Adding in another DC to give us a total of 3 DC's (2 existing GC's & add 1 for the FSMO) and a proxy server as it's been requested to track all websites visited (we have blocking already via OpenDNS, but need it recorded now too) So the new server with the exception of the DC & Proxy would be taking over server 3's role. While I will agree all this may push the hardware, I don't see this as being an all eggs in one basket from a single point of failure. Since the Proxy & DC would the last added and really are optional, if the server is loaded by the time the other stuff is running then they can be dropped from the plan, sorry I didn't mention that before.

When everything comes off server 3, it can be wiped, ESXi put on it, and it can take over a couple of the VM's such as the External Webserver & Proxy and hopefully add RRAS for the VPN users since with our current VPN router users can't use their AD account to authenticate so thats been a request as well.

I'll be asking for 2 more drives to do a RAID 10 for the new server.

4 Answers 4

5

Moving a VM from VMware to Hyper-V is straight forward.

You are pushing things to depend on the memory overcommit and 7.2K drives (in any configuration) with a heavy I/O use like Exchange.

That said, any use of striping (RAID 0) will help I/O. You could do all four drives as RAID 10. That RAID 1 for redundancy/safety. The RAID 0 part at least spreads the I/O out a little. One problem with that is RAID 1 being the highest consumer of space. That is, if you have 4 drives, each at size x, then RAID 1 (or 10) will net you 2x in space. With RAID 5, it would be 3x. Of course, if you want safety, you should use that fourth drive slot for a hot spare.

Good luck at putting all this into such a small package.

3
  • 1
    +1 for RAID10, only possible RAID level with this kind of limited I/O avaible.
    – pauska
    Nov 29, 2009 at 17:56
  • Spend the dollars now and go with RAID 10.. I have a RAID 1 Hyper-V box and it just does not perform likes its brother that has RAID 10. Nov 29, 2009 at 20:27
  • Thank you for a good answer. The RAID 10 for the performace boost makes sense and with that I should be able to justify 2 more drives. Since this would give us 1TB of space with the drives we are going to get, that is about 4x more then we are using now the hit on space isn't an issue.
    – user26147
    Nov 30, 2009 at 13:56
2

You might not have the $$ to get an MD-1000..but you are really living on the edge here. 100 users/30 active with spam filtering on the same set of spindles really does not give you a lot of capacity for peak usage.

I would suggest that you separate spam filtering and exchange. Spam filtering can be spread out on a couple of lower powered pizza boxes if you have them, or even do something like Grey listing before (http://hcpnet.free.fr/milter-greylist/) you send your mail to the Symantec suite.

If you are installing Exchange new you will find some new anti-spam features that were released in the last roll up. I have kicked Symantec to the curb completely in favor of grey listing and VIPRE from Sunbelt software. 20 users for 200 bucks for a kick tail antivirus/antispy solution that is light and easy to install.

I currently have 4 Core/Hyper-V servers and have had great success, but would not dream of putting my entire shop on just one. With Server 2008 R2 you can really visualize and protect yourself at the same time for about the same money.

1
  • I'll take a look Grey listing to see about taking some of the load off the spam engine.
    – user26147
    Nov 30, 2009 at 14:25
0

Frankly at this point I'd recommend going back to your budgeters and asking why they think saving a couple of grand on the machine that they are apparently putting their entire IT infrastructure on makes any business sense. But that's just me.

With 4 drive slots your only reasonable option is RAID 10, for the redundancy (because no production machine should be without it) and the speed (for I/O-happy Exchange). Unfortunately, that leaves you with two drives worth of total space, but I guess beggars can't be choosers.

Re: V2V between ESXi and Hyper-V, it's no problem.

4
  • I didn't know I had said that this was the entire IT infrastructure....oh wait I didn't
    – user26147
    Nov 30, 2009 at 13:48
  • Truth hurts? Seriously man, don't be insulted when people tell you that you aren't using enough hardware. You ask the question, you're going to get an honest answer.
    – phoebus
    Nov 30, 2009 at 15:51
  • Hardware for what I wanted to do with this, yes correct I even acknowledged this could be pushing it in my edit. By no means did I say my entire infrastructure on this one server. Thats why I made the comment, you made an assumption this was the entire network. With 2 more seperate DC's and a seperate file server I'd say it's about as broken down as you can get with 3 servers.
    – user26147
    Nov 30, 2009 at 22:08
  • I think you took a comment I made in jest WAY too seriously. The point I was making, via hyperbole, is that one would hope your budgeters will understand the issues with putting so much on one box. I didn't literally mean that this box had to be your one and only IT asset.
    – phoebus
    Nov 30, 2009 at 23:35
-1

all that on a single host, while the host is not at least an R900 or suchlike (since we're talking Dell here), is rather heavy. Only the Exchange VM is going to require all the spindles you've got, and the Symantec one is only going to add to that load.

Not to mention you're putting all your eggs in the same basket here.

I would suggest getting another host and a simple DAS to set up in between (Dell again - an MD1000 should be enough, although an MD3000 is preferable), and setting up a proper Virt-solution. My fave is RHEV (memory overcommit is there, btw.), but you can go with ESX or Hyper-v if you like

5
  • You're assuming this is our only server, it's not. I'm also not sure I understand how having a RAID 10 in an MD1000 is any different then a RAID 10 in the server itself. If I'm debating over purchasing 2 more drives due to budget, there is no way an MD1000 is even up for consideration. I might as well be asking for a flying monkey to deliver it too. Sorry for being scastic but really, what your suggesting isn't even an option.
    – user26147
    Nov 30, 2009 at 13:52
  • Just think why EVERY post in here was about you not having enough resources to pull off what you're trying to do. NL-SAS drives are really SATA disks that have a SAS interface, so the ATA-SCSI conversion overhead is spared, nothing more. if you have four of them, your Raid10 will not be very fast, and disk IO will e the main bottleneck. Two more drives will really add only one spindle to the equation, so instead of trying to push limits, think about either leaving the roles on separate, if rather weak, servers, or getting some more serious virtual hosts.
    – dyasny
    Nov 30, 2009 at 16:55
  • Ok you summed it all up nicely. Unless youve got $10,000 to get yourself a or big disk array with there is no way at all you can virtualize anything. So if you cant afford it, go home we dont want you playing in our sandbox. According to you even if I went with 15k SAS drives in the server itself it still wont be enough and there is not way this will work without that array.
    – user26147
    Nov 30, 2009 at 22:13
  • Yes I voted you down this time for suggesting the array. No all of us have massive IT budgets, and $10,000 for us would be massive.
    – user26147
    Nov 30, 2009 at 22:15
  • I'm afraid you simply do not get what I am saying. You can play with virtualization as much as you like, but putting ALL YOUR PRODUCTION SERVERS on a single host, with no possibility of failover, introduces a huge SPF. If the host goes down, and there are enough components there that might do that, you lose all the servers. If downtime in your case is cheap, it's your call. In my experience, it never is. Sigle host virtualization is perfect for testing and integration staging, as well as geoclustering if you have that set up, but in production in a local DC? I'd never be caught doing that.
    – dyasny
    Dec 1, 2009 at 10:29

You must log in to answer this question.

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