Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

Host Server:

  1. Dual Xeon E5530's
  2. 24 GB Ram
  3. 4x 1GB NIC's (operating at 100MB because of 100MB maanged switch)
  4. Server 2008 R2 Enterprise with Hyper-V Role + BackupExec 2012 backing up to NAS connected via iSCSI

VM's:

  • Domain Controller (also DNS server)
  • File Server
  • Database Server
  • Two App Servers

Current NIC Setup:

  • 1 Physical NIC dedicated to host OS so that BackupExec can do it's thing without choking out VM's.
  • 1 Physical NIC shared between APP servers. (NIC usage peaks and valleys through the day)
  • 1 Physical NIC shared between SQL Server and File Server (two biggest bandwidth hogs)
  • 1 Physical NIC dedicated to DC

Questions:

  1. Is dedicating one physical NIC to the DC/DNS overkill? I have about 20 users.
  2. Any tips about setting this whole thing up better?
  3. Are there any way to prioritize the different VM's sharing a NIC?
  4. I'm going to stack a 1GB switch on the 100MB one. 3 physical Servers, NAS's and that kind of thing will connect to the 1GB switch. Users will all be plugged into the 100MB switch. With the increased bandwidth am I safe putting more VM's on one physical NIC, or are there other factors to consider?

Thanks!

share|improve this question
add comment

3 Answers

up vote 7 down vote accepted
  1. Yes
  2. It's pretty common to bond all the VM NICs and share them between the machines. It's common to have two Management NIC interfaces that are bonded as well, but not used for any VM traffic. This leads to most VM servers have at least 6 NIC interfaces, and 8 to 10 is not uncommon.
  3. Yes, but not with what you've got now, and you're probably not interested in what solutions do exist.
  4. Without knowing the actual usage that's common on your servers I can't say with any certainty. But, that sounds reasonable, especially given the above. Do note that if you're 100Mb switch can have 1Gb uplinks of any kind I would highly recommend getting a couple. If not, bond at least a few 100Mb uplinks from that switch to the 1Gb switch.
share|improve this answer
    
I'm thinking of flagging your answer as a pre-emptive duplicate of mine. I think time-travel was involved? I've been playing a lot of Portal 2 lately. –  mfinni Aug 10 '12 at 19:08
    
Wow 1 minute apart! I switched the check from mfinni over to you. Thanks for the advice. The 100 MB switch is a PowerConnect 3548. Per the Dell Rep I can stack 1 GB switches. –  FullTimeCoderPartTimeSysAdmin Aug 13 '12 at 22:35
add comment
  1. Yes. Think about how much traffic is generated between 20 users and a domain controller. It's not much.
  2. I know more about VMware and less about Hyper-V, but the VMware way to do this is to bond/trunk all of those interfaces for guest traffic, and then VLAN the guest NICs as necessary.
  3. Are you actually facing any problems? From your description, it just sounds like you're prematurely optimizing. That said, if your switches support it, you could do QoS there to prioritize, I dunno, SQL traffic maybe?
  4. Without knowing what your traffic is like, I'm going to say "Probably almost definitely yes."
share|improve this answer
    
On #3, the Hyper-V switch is 10Gb, so the VMs will be able to saturate the uplink from that to the physical switch easily. This can lead to problems, but as you point out he's probably trying to optimize something that isn't actually a problem. –  Chris S Aug 10 '12 at 19:08
    
Re # 3 yes. Previously I had all 4 NICs shared with the physical host. This worked well until I attached a NAS via iSCSI with multipath and started pushing large (multi-TB) backups. DC was being choked out and there was a big clusterf*ck. (Or that's what it seemed like to me.) All this partitioning is in response to that. It sounds like I've went a little overboard in the other direction though. Thanks for your help!. –  FullTimeCoderPartTimeSysAdmin Aug 13 '12 at 22:28
    
Right - that's what I meant by "bond/trunk all of those interfaces for guest traffic" - that means just the ones for guest traffic. You do need to keep some bandwidth dedicated for host and storage. –  mfinni Aug 14 '12 at 0:02
add comment

Is it overkill? No, absolutely not. You're dealing with 100Mb ethernet here and the more NICs you can use the better.

I say, monitor performance, move things around if necessary, and then make it a goal to get to Gigabit.

Also... I'm assuming when you said "stack" that you meant you'd connect a gigabit switch to the existing switch by some kind of "stacking" mechanism, and not by daisy chaining, otherwise, you'll only be improving server to server connectivity, and making server to user worse.

share|improve this answer
1  
Spiff, you really think his single DC for 20 users will fill a 100 Mb link? –  mfinni Aug 10 '12 at 19:09
    
I do agree with the need to monitor (otherwise, you don't know what your problems are, you're only guessing), the desire to move to Gb, and the points about stacking. But I am guessing that, in the absence of some weird usage pattern, a dedicated 100 Mb link for his DC is indeed overkill. –  mfinni Aug 10 '12 at 19:22
    
No, I think those other servers will fill up the 100Mb link and response times from his DC will not be as timely. I'm sorry, an application can wait, processing of login scripts, DNS lookups, and LDAP requests should not be competing with other servers. –  SpacemanSpiff Aug 10 '12 at 19:31
1  
And personally, I'd be putting the App servers and SQL servers into the same virtual switch so server to server traffic does not even leave the host, unless Microsoft's virtual switch is just so poor that the traffic must traverse a physical link from one VM to another. –  SpacemanSpiff Aug 10 '12 at 19:37
    
Yes this was the fear. (That the other servers on the 100MB link would choke out the DC.) I know that a DC with 20 users and a dedicated 100MB link isn't an issue. If 100MB is way more than enough though (as others have suggested), I can try stacking one of the lower bandwidth servers on the same NIC and seeing what happens. –  FullTimeCoderPartTimeSysAdmin Aug 13 '12 at 23:02
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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