2 Replies Latest reply on May 15, 2020 4:43 AM by Krillian

    Question regarding stretched cluster design

    Krillian Lurker

      Hello,

       

      we have a customer who wants us to establish redundancy/disaster recovery for our system using vm ware. The customer has 3 sites in a metropolitan area as shown in the picture. We'll only have one VM active that is providing our system, perfmon shows that we have an average of 25 writes per second(mostly log files and a mongo db).

       

      Is it possible to run a stretched cluster given these parameters?

      Thanks!

        • 1. Re: Question regarding stretched cluster design
          TheBobkin Virtuoso
          VMware EmployeesvExpert

          Hello Krillian,

           

          Welcome to Communities.

           

          "perfmon shows that we have an average of 25 writes per second(mostly log files and a mongo db)."

          So that you are aware 'IOPS' or 'writes per second' are not particularly scientific nor precise measurements of storage workload - e.g. you could have one VM that requires '10 IOPS' (512k random writes) and another VM that requires '1000 IOPS' (4k sequential reads), then as you can see the number of 'IOPS' essentially becomes a useless metric as the first VM will have way higher storage performance requirements (even though 10 is less than 1000). You can use VMware fling IOInsight or similar tool to look at the workloads metrics in detail.

           

          When designing clusters you should get an idea of what storage the workload(s) are currently running on and whether it is adequate or being pushed to its limits and then using this information to determine how beefy the storage on the vSAN cluster needs to be (e.g. whether it needs to be All-Flash or whether Hybrid will suffice, what cache/capacity ratio will do, disk and Disk-Group sizing etc.).

           

          If you have 1ms RTT between the sites, this is actually within the recommended max limits for a non-stretched cluster and thus having a stretched cluster with this low latency between sites will be fine.

           

          Bob

          • 2. Re: Question regarding stretched cluster design
            Krillian Lurker

            Hello TheBobkin,

             

            thanks for your quick response. i've setup perfmon to also record Bytes writen per Second as seen below.

              

               

            DateBytes writte/sWrites/s
            05/14/2020 09:49:25,224138046,90717,85449033
            05/14/2020 09:49:30,22473724,908337,399689692
            05/14/2020 09:49:35,224375184,59921,59948181
            05/14/2020 09:49:40,224164645,567713,99884093
            05/14/2020 09:49:45,225183499,099919,19982212
            05/14/2020 09:49:50,225156445,885717,39762973
            05/14/2020 09:49:55,22698291,0017712,59833397
            05/14/2020 09:50:00,226387460,137254,7969646
            05/14/2020 09:50:05,227112222,968713,39911273
            05/14/2020 09:50:10,227137618,473416,39915077
            05/14/2020 09:50:15,227855953,728321,79719189
            05/14/2020 09:50:20,22892562,2068711,59907356
            05/14/2020 09:50:25,229308796,449821,19712036
            05/14/2020 09:50:30,229131066,925510,39959736
            05/14/2020 09:50:35,229149091,564719,79962347
            05/14/2020 09:50:40,230187569,452718,3973177
            05/14/2020 09:50:45,230122876,1294

            11,19964721