locked
Server 2016 cluster network best practices RRS feed

  • Question

  • Is there a document anywhere that describes Server 2016 cluster network best practices?  Have they changed significantly since Server 2012 R2?

    I'd also like to understand a little better how the various components of Server 2016 cluster networks are supposed to operate.

    -- Ken


    /kenw K&M Systems Integration Ltd. 403-274-7848

    Tuesday, May 10, 2016 1:25 PM

Answers

  • There are kind of two paths...   you can absolutely keep doing what you were doing in Win2012 R2, and there is the new Software-defined Networking path in Win2016.  What are you interested?

    Or are you more interested in cluster specifics?  We have made changes to heartbeating, how we handle transient network failures between nodes, etc...

    Thanks!
    Elden

    Wednesday, May 11, 2016 1:20 AM
  • We have recently updated the hyper-converged doc for TP5, and it has details on networking.  Does this address your questions?

    https://technet.microsoft.com/en-us/library/mt693395.aspx

    Thanks!
    Elden

    Wednesday, May 11, 2016 9:03 PM

All replies

  • There are kind of two paths...   you can absolutely keep doing what you were doing in Win2012 R2, and there is the new Software-defined Networking path in Win2016.  What are you interested?

    Or are you more interested in cluster specifics?  We have made changes to heartbeating, how we handle transient network failures between nodes, etc...

    Thanks!
    Elden

    Wednesday, May 11, 2016 1:20 AM
  • Hi,

    >Is there a document anywhere that describes Server 2016 cluster network best practices?
    As Windows Server 2016 is still a preview version, there is no detailed document/article/KB about the cluster network best practice on Microsoft official website. 

    >Have they changed significantly since Server 2012 R2?
    New function about WS 2016, you may reference - What's New in Windows Server 2016 Technical Preview 5:
    https://technet.microsoft.com/en-us/library/dn765472.aspx

    Document/article/KB which is applied to previous OS version can also be considered as a reference for deploying WS 2016.

    Best Regards,
    Eve Wang

    Please remember to mark the replies as answers if they help and unmark them if they provide no help. If you have feedback for TechNet Support, contact tnmff@microsoft.com.

    Wednesday, May 11, 2016 2:49 AM
  • I was just looking more for general architectural overview.

    For example, following the documents from TP4 with hyper-converged cluster, it appeared to want two separate virtual networks, but it was unclear to me why that was,   Doesn't seem to be actually necessary.

    So just an overview of what network connectivity is required would be helpful.

    -- Ken


    /kenw K&M Systems Integration Ltd. 403-274-7848

    Wednesday, May 11, 2016 1:23 PM
  • We have recently updated the hyper-converged doc for TP5, and it has details on networking.  Does this address your questions?

    https://technet.microsoft.com/en-us/library/mt693395.aspx

    Thanks!
    Elden

    Wednesday, May 11, 2016 9:03 PM
  • Can you please update the best practices For windows 2019 and 2016.  The 2016 server says software defined settings can be used but the cluster logs shows the the server is dropping 40% of the packets.  The servers on the other side are going into an orphaned state.  I still do not find clear documentation on the best practices for geo clusters.

    Network Best Practices:
    Discuss with customer his network architecture and if he can provide stretched VLANs across sites that can reduce the Multi-Site Failover Cluster complexity against different VLANs.
    Share all implementation networking requirements early with customer network team, especially if you are going to do changes in network design related to the required VLANs.
    Consider encryption over WAN.
    https://blogs.technet.microsoft.com/meamcs/2013/11/09/microsoft-windows-multi-site-failover-cluster-best-practices/

    My other issue is with internal cluster network as Even with the correct metrics the internal cluster network marked with a * is still at the top. 

    If you look at the article it states if the binding order is wrong you will lose network packets.

    https://support.microsoft.com/en-us/help/981953/an-incorrect-ip-address-is-returned-when-you-ping-a-server-by-using-it

    If you go further into this you will see the Connection* which means this is the internal cluster adapter as discussed here.

    http://blogs.technet.com/b/askcore/archive/2010/04/15/windows-server-2008-failover-clusters-networking-part-4.aspx

    According to those articles and the cases I have work with Microsoft if the internal cluster connections were at the top of the binding order we would see the clusters go offline.

    According to the article on AutoMetric This should never happen and the internal cluster networks should never be at the top of the binding order

    My concern is that we appear to be having issues with the cluster communication.


    • Edited by ftpex1 Friday, January 4, 2019 7:15 PM
    Thursday, December 27, 2018 7:42 PM