none
failover design question

    Question

  • hi, hi,  experts, I have a new project now.
    for the project, we are going to purchase 2 blade servers.
    one for primary and one for secondary.

    on 2 blade servers, Hyper V will be installed for VM solutions.
    there are 3 VMs to be created in each primary blade and secondary blade. totally , 6 VMs.

    on each blade.
    A) windows server 2016 standard, web portal, tomcat (vendor to manage) , 4 cores, 8 gb ram
    B) windows server 2016 standard, IIS, .net web services , 4 cores, 8 gb ram
    C) windows server 2016 standard, SQL server 2016 standard,, 4 cores, 8 gb ram

    web portal on A calls web services on B and database of A is C.

    for this design, may I know , as the VM solution is HyperV. 
    I dont have any HA knowledge using Hyper V.

    what option I have for the DR case?

    can whole VMs set (A+B+C) replicate to 2nd blade server?
    when disaster happends, whole set application failover to 2nd blade server?

    or I have to check how to do per VM in OS level?

    please advise. thank you !

    delaynomore

    delaynomore.

    Friday, March 03, 2017 3:35 AM

Answers

  • If you don't want the expense of a real physical shared storage system I can say that StarWind is a great choice having used it in the past, simple user interface (you can powershell script the setup too as they provide their own PS modules), it's cost effective and great performance. They will also be happy to arrange some product demos to show you how it all works, it turns your locally installed hard disks into a shared storage solution by layering a virtual disk on top and synchronizing the virtual disk on each node (synchronously).

    At the end of the day, it's all dependent on how important your up-time is, automatic fail-over in the event of physical server failure, and how deep your pockets are. figure out your RTO and RPO and go from there.

    • Marked as answer by delaynomore Wednesday, March 15, 2017 2:08 AM
    Friday, March 10, 2017 8:58 PM
  • if it's just the 2 physical servers you have it may well be much more worth while to use hyper-v replica, since the secondary is classed as a DR server (provided it's not running live workloads) you do not have to license that second server, you could even install the hyper-v free edition.for just 3 VMs this is the simplest and easiest and cheapest option, but failover is not automatic.

    or, you could make a 2-node failover cluster, but then you need a solution where shared storage is available between the two nodes AND you have to license both nodes in the cluster for all the VMs which could potentially run, this is the most expensive solution but you do get automatic failover and live migration (if you need to bring the primary down for maintenance) with no downtime.

    for 3 VMs if uptime isn't massively important, go with the first option. you can replicate down to every 30 seconds, 5 minutes or 15 minutes, and you also have a chance to periodically perform a test failover where you can check the VM will boot and log in

    High Availability, Disaster Recovery, and Backup and Recovery are different approaches to business continuity, HA is the most expensive one is what i refer to as a "first tier" if you can afford it, otherwise drop down to DR, worst case BR. ideally you should have all three but it depends on how deep your pockets are and the criticality of your workloads - HA can get extremely complex and very resilient

    • Marked as answer by delaynomore Saturday, March 04, 2017 4:23 AM
    Friday, March 03, 2017 10:37 PM
  • Failover is manual with hyper-v replica, you could come up with genius way to use a scheduled powershell script to detect when the primary site has failed and trigger the failover but i shouldnt put such an idea in your head because it would be an extremely complicated way to approach things and anything could go wrong - probably best not to do such a thing.

    if you install hyper-v free edition you could run some free linux based VM on the DR server which detects failures and sends you a notification (if you put server standard on there you will have to pay for the license if there is any VM running linux or not), if you put windows standard on the DR server and all your VMs are off (as they are in replica's) then you don't have to pay for the server standard license provided it's only used for DR purposes.

    great thing about hyper-v replica is you can specify what IPs you want the VMs to use at the failover site, so if they are in a different subnet then when they do failover they will get the correct IPs for that subnet.

    if you're constrained by bandwidth for replication take a look at the New-NetQoSPolicy https://technet.microsoft.com/en-us/itpro/powershell/windows/qos/new-netqospolicy

    scope if for the vmms.exe process, and the correct port your replica traffic will be using, run the command on the source sending server

    as previously stated if you want automatic failover you're going to have much more expense and complexity with shared storage and multi-subnet cluster if the two half are using different subnets, and the licensing cost increases because you must license every server in the cluster. it's already going to cost you 3 standard licenses for 1 server, a 2 node cluster is going to cost you 6 standard licenses

    • Marked as answer by delaynomore Monday, March 06, 2017 2:05 AM
    Saturday, March 04, 2017 10:23 AM

All replies

  • Hi Delaynomore,

    >>what option I have for the DR case?

    You could use Hyper-V replica for DR.

    >>can whole VMs set (A+B+C) replicate to 2nd blade server?

    Yes, all VMs could be replicated.

    >>when disaster happends, whole set application failover to 2nd blade server?

    Yes, VMs would failover automatically.

    Well, you could also use failover clustering for high availability.

    https://technet.microsoft.com/en-us/library/hh831579(v=ws.11).aspx

    Best Regards,

    Leo


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

    • Marked as answer by delaynomore Friday, March 03, 2017 1:25 PM
    • Unmarked as answer by delaynomore Friday, March 03, 2017 1:26 PM
    Friday, March 03, 2017 7:16 AM
    Moderator
  • if it's just the 2 physical servers you have it may well be much more worth while to use hyper-v replica, since the secondary is classed as a DR server (provided it's not running live workloads) you do not have to license that second server, you could even install the hyper-v free edition.for just 3 VMs this is the simplest and easiest and cheapest option, but failover is not automatic.

    or, you could make a 2-node failover cluster, but then you need a solution where shared storage is available between the two nodes AND you have to license both nodes in the cluster for all the VMs which could potentially run, this is the most expensive solution but you do get automatic failover and live migration (if you need to bring the primary down for maintenance) with no downtime.

    for 3 VMs if uptime isn't massively important, go with the first option. you can replicate down to every 30 seconds, 5 minutes or 15 minutes, and you also have a chance to periodically perform a test failover where you can check the VM will boot and log in

    High Availability, Disaster Recovery, and Backup and Recovery are different approaches to business continuity, HA is the most expensive one is what i refer to as a "first tier" if you can afford it, otherwise drop down to DR, worst case BR. ideally you should have all three but it depends on how deep your pockets are and the criticality of your workloads - HA can get extremely complex and very resilient

    • Marked as answer by delaynomore Saturday, March 04, 2017 4:23 AM
    Friday, March 03, 2017 10:37 PM
  • hi, Milkientia

    if the design is using Hyper-V replica solution in secondary site, let says replicate to vm replica in 2nd site from 1st site in every 30 seconds,

    during the disaster, is there no automatic failover to 2nd site, or I have to manually go to 2nd site to trigger failover operation manually?

    further, if failover operation is triggered manually or automatically and complete, 2nd replica will have same IP as VM in 1st site? let says the VM is actually web server. after the failover operation complete, can users still access the website hosting in the VM?

    I am new to this server , DR, failover knowledge, please kindly help me.

    delaynomore


    delaynomore.

    Saturday, March 04, 2017 9:33 AM
  • Failover is manual with hyper-v replica, you could come up with genius way to use a scheduled powershell script to detect when the primary site has failed and trigger the failover but i shouldnt put such an idea in your head because it would be an extremely complicated way to approach things and anything could go wrong - probably best not to do such a thing.

    if you install hyper-v free edition you could run some free linux based VM on the DR server which detects failures and sends you a notification (if you put server standard on there you will have to pay for the license if there is any VM running linux or not), if you put windows standard on the DR server and all your VMs are off (as they are in replica's) then you don't have to pay for the server standard license provided it's only used for DR purposes.

    great thing about hyper-v replica is you can specify what IPs you want the VMs to use at the failover site, so if they are in a different subnet then when they do failover they will get the correct IPs for that subnet.

    if you're constrained by bandwidth for replication take a look at the New-NetQoSPolicy https://technet.microsoft.com/en-us/itpro/powershell/windows/qos/new-netqospolicy

    scope if for the vmms.exe process, and the correct port your replica traffic will be using, run the command on the source sending server

    as previously stated if you want automatic failover you're going to have much more expense and complexity with shared storage and multi-subnet cluster if the two half are using different subnets, and the licensing cost increases because you must license every server in the cluster. it's already going to cost you 3 standard licenses for 1 server, a 2 node cluster is going to cost you 6 standard licenses

    • Marked as answer by delaynomore Monday, March 06, 2017 2:05 AM
    Saturday, March 04, 2017 10:23 AM
  • Hey,

    Milkieta suggested great options for you.

    For a 2-node failover cluster you will need a shared storage. You can go with Software-Defined Solutions such as HP VSA or StarWind Virtual SAN

     

    • Proposed as answer by Taras Shved Tuesday, March 14, 2017 4:12 PM
    Friday, March 10, 2017 5:01 PM
  • If you don't want the expense of a real physical shared storage system I can say that StarWind is a great choice having used it in the past, simple user interface (you can powershell script the setup too as they provide their own PS modules), it's cost effective and great performance. They will also be happy to arrange some product demos to show you how it all works, it turns your locally installed hard disks into a shared storage solution by layering a virtual disk on top and synchronizing the virtual disk on each node (synchronously).

    At the end of the day, it's all dependent on how important your up-time is, automatic fail-over in the event of physical server failure, and how deep your pockets are. figure out your RTO and RPO and go from there.

    • Marked as answer by delaynomore Wednesday, March 15, 2017 2:08 AM
    Friday, March 10, 2017 8:58 PM