none
Hyper-V failover solution for systems generating large number of files

    Question

  • Hi All,

    In our environment we have a 2 node cluster, we also have a cluster located at our DR site that we are using to replicate our VM's to in case of the event of a DR failure. Our production site has 10 VMs (plus many others), the 10 servers receive data from our users and write the data back to a PRIMARY server. The PRIMARY server then tags the data, indexes them and saves it to a folder that is created daily. On average, our users/servers generate 10,000 files a day totalling 2GB.

    As you can see we have single point of failure with the PRIMARY server and we are looking for a solution that will allow us to have redundancy should the PRIMARY server fail. We have discussed serveral options however none of them feel right due to other issue they my present or it being extremely complicated.

    We have discussed the option to:

    1. Create two new VM's on the cluster/nodes and within the two VM's create another cluster, we would then add the file share role between them hosting a UNC/file share. We would replicate the VM's to DR via Hyper-V and use SAN replication for the data.  - The issue we have with this is the VM's are hosted on the existing cluster, we could need to connect ISCSI directly to the VM's however the attached storage would not have failover/MPIO capabilities as they are not CSV's.
    2. Add the file server role to the existing cluster and the 10 servers would all write to the UNC path - The issue here is fail-over with regards to DR. Since our DR cluster is separate from our production site how would we replicate the file server/share role to the DR cluster?
      We could use DFRS however i am not sure DFRS is a good idea give the amount of files and data being generated.

    I am reaching out to everyone to see if you have any other solutions we could use to achive a our goal simplisticly. I look forward everyone's suggestions.

    Tuesday, July 24, 2018 6:45 AM

Answers

  • If you want the VMs to use the CSV, you have to make them highly available.  Only cluster resources can access cluster resources.  Besides, there is no real value in not making them highly available.

    The shared storage for the cluster, your iSCSI LUN, can be either a normal cluster disk or a CSV.  Either one will work.


    tim

    Wednesday, July 25, 2018 1:08 PM

All replies

  • "The issue we have with this is the VM's are hosted on the existing cluster, we could need to connect ISCSI directly to the VM's however the attached storage would not have failover/MPIO capabilities as they are not CSV's."

    I don't understand your reasoning here.  VMs can be configured to use MPIO to iSCSI connections.  VM clusters can be configured to use either normal cluster disks or CSV disks.  Either one can be configured for MPIO.

    "Add the file server role to the existing cluster"

    Not recommended.  It is not recommended to add any additional cluster role to a cluster that is configured to host virtual machines.  This gets to be more critical if you expect clustered VMs to access the services of the file services on the host nodes.  Really bad idea.

    Another possibility is look into a solution that uses a database instead of 10,000 files.  Something like SQL Server offers different methods of providing highly available solutions that are much easier to manage than something that is managing 10,000 individual files.


    tim

    Tuesday, July 24, 2018 1:03 PM
  • Tim thanks for getting back to me.

    So just thinking more about option 1. Could we:

    1. Create a new LUN and the two new VM's (NOTE: The new VM's will not be HA enabled)
    2. The VM's system drive(s) would be placed on one of the existing clusters CSV's
    3. Connect the LUN to the VM's via iSCSI
    4. Configure ISCSI/MPIO on both VM's
    5. Create the cluster (eg. HA-DATA-CLUSTER), add the LUN/disk and convert it to a CSV.
    6. Add the file share role, pointing to a share on the LUN

    In essence we are creating two virtual nodes within the cluster, correct?

    If there is node failure:

    The current cluster (HA-PRODUCTION-CLUSTER) will failover all the CSV's and VM's configured for HA.

    The cluster (HA-DATA-CLUSTER) will detect that virtual node is offline and so the file server role will failover to the other node.

    Wednesday, July 25, 2018 2:54 AM
  • If you want the VMs to use the CSV, you have to make them highly available.  Only cluster resources can access cluster resources.  Besides, there is no real value in not making them highly available.

    The shared storage for the cluster, your iSCSI LUN, can be either a normal cluster disk or a CSV.  Either one will work.


    tim

    Wednesday, July 25, 2018 1:08 PM
  • Hi ,

    Just checking in to see if the information provided was helpful.
    Please let us know if you would like further assistance.

    Best Regards,
    Candy


    Please remember to mark the replies as an answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com   

    Monday, July 30, 2018 9:21 AM
  • All, the above has pointed me in the right direction however i have not tested this as yet.

    Theoretically it makes sense and should work so for now i will mark the above suggestions as answered.

    Thanks for all the help.

    Tuesday, July 31, 2018 10:20 PM