VMware vSphere

 View Only
  • 1.  One or two vDS in a stretched cluster between two datacenters?

    Posted Dec 06, 2018 01:58 PM


    I need to build a stretched cluster between two datacenters. There are 4 hosts on each site, so a total of 8 hosts in the cluster.  I have HPE peer-persistence LUNs implemented, so synchronously mirrored LUNs protect the vms and if one site failes, the vms will be started by vSphere HA automatically on the hosts on the second site. Admission Control host failures is set to 4, so one site can fail completely and 50% of ressorces are reserved by admission control.

    To this day I have vDS implemented only one one datacenter.

    Now in my new scenario described above, I have the challenge if I should configure only one vDS across all hosts (simple setup, easy maintenance) or if I should go with two vDS. The latter means, that the four hosts in "datacenter A" share a vDS and the four in "datacenter B" share another vDS. This is a little bit more complicated to maintain. The most important goal however is reliabilty and system stability. So if I only have one vDS and this one is broken, in the worst case the networking on all hosts/vms is affected. With 2 vDS I have a good chance that half of the hosts/vms still have network connectivity.

    One question I am not shure about is: Will DRS work and move VMs between two ESXi hosts with different vDS? (The portgroups will be labelled with the same name, I know that in the meantime this works in vSphere 6.5, if the two vDS are located in different network folders)

    What would you recommend me? Which approach would you suggest with the assumtion that stability is most important? Will DRS work in this scenario?

    Best regards


  • 2.  RE: One or two vDS in a stretched cluster between two datacenters?

    Posted Dec 06, 2018 02:02 PM

    You need to continue using one vDS spanned between that stretched cluster. DRS will not move VMs to the other hosts even with port groups of the same name (because their underlying IDs differ), and for HA it will not start them on the other side if one half of your stretch fails. And a vDS isn't a physical entity, as you know, so they don't just "break" and stop working, so that shouldn't be a design reason.

  • 3.  RE: One or two vDS in a stretched cluster between two datacenters?

    Posted Dec 07, 2018 11:01 AM


    Thanks for your answer. I didn't await that HA and DRS won't work in this scenario with two vDS. I thought they will work, because when I do a vmotion for test purposes from a "vDS A" to anothver "vDS B" with identically labelled portgroups, in the wizard the portgroup name is suggested correctly, so I was assuming that HA will work as well...

    I know, vDS should not break normally. But probably a fatal configuration failure could break it in the worst case. Or  an upgrade of a vDS to a later version could fail and the connectivity to vmkernel adapters and virtual machines could fail.

    But of course: If you are right and DRS and - even more important - HA don't work with two vDS, I really have to go with only one...

    Best regards and thank you