locked
CSV and Networking recommendation for HyperV Cluster RRS feed

  • Question

  • Hello All,

    I have a project where I am required to build a HyperV Cluster with 4 nodes (4 host servers), and I had some questions regarding the use of CSV and network cards. Below is a brief summay of the environment.

     

    1. All 4 servers are brand new HP DL380 G7 and are exactly identical.

    2. The shared storage is an HP P2000 G3 FC SAN.

    3. All the servers have HBA cardswith fiber cables that connect them with the P2000 SAN, through HP(brocade) SAN switches.

    4. Each server has 4 network cards as they come standard on HP G7 servers.

    5. Typical virtual servers being hosted in this environment will vary between 200-500 GB in size, while possibly only one approaching 1TB.

    6. LIVE Migration is a requirement and should be available for all the Virtual Servers.

    7. The OS on all host servers will be Windows 2008 R2 Enterprise.

    8. The environment will be managed by SCVMM 2008 R2.

    9. I should also add that the SAN storage has enough space to allow seperate LUN for each virtual server if needed.

     

    Now questions I have are as follows,

     

    1. Should i use CSV in this environment or not? If yes Why , and if no Why not?

    2. How should I configure the network cards available and for what traffic?

    3. Is it advisable to use teaming on the network cards? what are the pros & cons? 

     

    Thanks in advance.

    Hardstone.







    • Edited by Hard Stone Friday, April 27, 2012 7:47 PM
    Friday, April 27, 2012 7:42 PM

Answers

  • 1. For networks, you need at least 4 (VM communications, CSV/Live Migration, Cluster Management, Heartbeat).  You connection to disk from the hosts is via FCP, however how about VMs?  Will everything reside on VHDs?  Will you do passthroughs?  Will you connect iSCSI from the guest?

    2. For CSV/Live migration network ... The onboard NIC is probably light unless you're willing to accept performance degradation at times.  When a CSV is in redirected mode, all traffic goes across the network to the coordinator node and then down to disk via block.  Live Migration time is a funtion of network speed in part...

    3. Do use CSV.  It's a lot less painful than drive letters or mount points.  You can still have a CSV for each node, however your VM failover times will be much faster with CSV.

    4.  Configure networks?  See 1 above.

    5.  I'd get mor ports and team or get a 10G card (and switching infrastructure) for the CSV/Live Migration network.

    J

    • Marked as answer by Vincent Hu Monday, May 7, 2012 8:06 AM
    Friday, April 27, 2012 7:51 PM
  • Hi,

    1. To use CSV or not, you can check the following blog.

    Hyper-V: CSV not Required for Live Migration

    http://blogs.technet.com/windowsserverexperts/archive/2010/01/29/hyper-v-csv-not-required-for-live-migration.aspx

    2. For Hyper-V Failover Cluster network configuration, you can check the following guides.

    Hyper-V : How many network cards do I need?

    http://blogs.technet.com/b/gavinmcshera/archive/2011/03/27/3416313.aspx

    Designating a Preferred Network for Cluster Shared Volumes Communication

    http://technet.microsoft.com/en-us/library/ff182335(WS.10).aspx

    3. It depends. Someone use teaming without any issues, some other not. And you need to know that teaming is not supported by Microsoft, at this moment.

    • Marked as answer by Vincent Hu Monday, May 7, 2012 8:06 AM
    Tuesday, May 1, 2012 6:15 AM

All replies

  • 1. For networks, you need at least 4 (VM communications, CSV/Live Migration, Cluster Management, Heartbeat).  You connection to disk from the hosts is via FCP, however how about VMs?  Will everything reside on VHDs?  Will you do passthroughs?  Will you connect iSCSI from the guest?

    2. For CSV/Live migration network ... The onboard NIC is probably light unless you're willing to accept performance degradation at times.  When a CSV is in redirected mode, all traffic goes across the network to the coordinator node and then down to disk via block.  Live Migration time is a funtion of network speed in part...

    3. Do use CSV.  It's a lot less painful than drive letters or mount points.  You can still have a CSV for each node, however your VM failover times will be much faster with CSV.

    4.  Configure networks?  See 1 above.

    5.  I'd get mor ports and team or get a 10G card (and switching infrastructure) for the CSV/Live Migration network.

    J

    • Marked as answer by Vincent Hu Monday, May 7, 2012 8:06 AM
    Friday, April 27, 2012 7:51 PM
  • Thanks john  for your reply.  Regarding iscsi, no there is no iscsi planned for this environment as everything is on fiber channel.

    Or the network recommendations you gave, how many switches will be needed?

    And if I understand correctly you recommend teaming but only for the Csv/live migration network, why not for others such as management or VM comunications?

    Friday, April 27, 2012 8:15 PM
  • Teaming adds complexity in and of itself.  If you don't absolutely need it, don't use it.  On the CSV network, you'll need it.  If you have a 4 node cluster and each node as 2x 4Gbit FC connections to storage, when in redirected mode up to 2 x 4Gbps could swamp your CSV network.  When you're in redirected mode, all io from non-coordinator nodes goes over the network to the coordinator, then out FCP to the storage.  Now of course the coordinator only 2 FC HBAs and in or 4Gbps example you're bottlenecked there when redirected.  It's better to have the bottleneck at the HBAs on the coordinator than on your CSV network however.  The quadport NC364T NIC is 1Gbps per port.  That could be a problem.  You might be better off with the Emulex 10Gps NIC in the servers for the CSV network.  Of course, then you'd need 10Gps switches....

    Friday, April 27, 2012 9:32 PM
  • From all I've read the VM communications network is most heavily used and most critical as that network has all the virtual servers traffic which are the whole reason hyperv is implemented, and that is the network that matters to users and where most load will be.

    Whereas the management network is not that traffic heavy, and realistically how often is CSV going to go in redirected mode when everything is dual (HBAs etc.) ??  Am I wrong on here?

    I was therefore thinking of teaming on VM communication network. And I was wondering maybe i should allow management traffic on it or not??(if allowed it is to work as backup for the main management network).

    Am I wrong on these concepts?

     

    Also I am not yet decided on using CSV.

    Any thoughts, anyone?





    • Edited by Hard Stone Saturday, April 28, 2012 9:01 AM
    Saturday, April 28, 2012 8:56 AM
  • I hope someone will reply soon to my points as I need to get this project underway pretty soon.

    Currently I am planning the following, as I have 4 NIC on each server,

    Use NIC 1 for management,

    Team NIC 2 & 3 and use for VM Network.

    Use NIC 4 for heartbeat and live migration/CSV.

     

    IF anyone sees any serious problems with this configuration, or any know issues then please do let me know.

    My own rationale for choosing this configuration is as follows,

     

    1. Redundancy is a major concern for this company. Hence, teaming on VM network so that virtual servers will remain available to users even in case of one NIC failure.

    2. If management NIC goes down, it does not affect the traffic for users and VMs and from users perspective the whole environment would still be UP. The administrators can access the host servers directly via KVM in this case.

    3. Due to Dual HBA and MPIO on servers, dual controllers on SAN storage and also TWO SAN switches, server to storage connectivity is fully redundant  and on Fiber Channel (no iSCSI). Also back-up solution has not yet been selected but any backups will be planned for Off-Peak hours only. Hence I think that likelyhood of "redirected access" on CSV -although possible- is reduced to minimal.  Therefore Choosing to share NIC with heartbeat, which should not be very busy/loaded.

    4. Same reason as #3 for sharing live migration with heartbeat. I *think* heartbeat is a low-usage network and live migrations in this environment are not expected to be that frequent.

    Besides Live migrations can be planned and how long does it take anyway to change the NIC for live migration? The company is far more focused on FAIL-OVER and I need to optimize the environment for that. Are my choices as outlined above the correct ones for this scenario? (and with the available hardware).

     

    All inputs will be highly appreciated.



    Saturday, April 28, 2012 4:44 PM
  • Hi Hard Stone,

    To make like this design , you can check the following links.
    I created all my other vendor harware with these guide.
    Networking is so important because nobody wants downtime for any issues.

     
    Dell™ Networking Solutions Guide for Microsoft® Hyper-V™
    http://support.dell.com/support/edocs/software/HyperV/en/nsg/nsga00.pdf


    Dell™ High Availability Solutions Guide for Microsoft® Hyper-V™
    http://support.dell.com/support/edocs/software/HyperV/en/hsg/hsga00.pdf

    NetApp Storage Best Practices for Microsoft Virtualization and NetApp SnapManager for Hyper-V
    http://media.netapp.com/documents/tr-3702.pdf
    3 NETWORK CONFIGURATION (Page 8)

    Sincerely,
    Murat Demirkiran

    If the post helps you and remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. www.scvmm2012.com (TR Language)

    Saturday, April 28, 2012 11:11 PM
  • Hi,

    1. To use CSV or not, you can check the following blog.

    Hyper-V: CSV not Required for Live Migration

    http://blogs.technet.com/windowsserverexperts/archive/2010/01/29/hyper-v-csv-not-required-for-live-migration.aspx

    2. For Hyper-V Failover Cluster network configuration, you can check the following guides.

    Hyper-V : How many network cards do I need?

    http://blogs.technet.com/b/gavinmcshera/archive/2011/03/27/3416313.aspx

    Designating a Preferred Network for Cluster Shared Volumes Communication

    http://technet.microsoft.com/en-us/library/ff182335(WS.10).aspx

    3. It depends. Someone use teaming without any issues, some other not. And you need to know that teaming is not supported by Microsoft, at this moment.

    • Marked as answer by Vincent Hu Monday, May 7, 2012 8:06 AM
    Tuesday, May 1, 2012 6:15 AM