2 Replies Latest reply: Jul 2, 2014 7:49 AM by RSS

    Storage Virtualization and Nimble - makes it sense?

    Visibility: Open to anyone

      Hi folks,


      due to customer demand I must ask the question:


      the lack of real high availability features (async repl only) drives me to the question,

      does it make sense to link two (or more) nimble boxes with an state-of-art (OKmakes

      it complicated) storage virtualization product live Datacore or Falconstore?


      Are the any thoughts or drawbacks about this?


      Spring in Germany



        • Re: Storage Virtualization and Nimble - makes it sense?
          Ben Watson Adventurer

          Hello Andreas,


          I've wondered the same thing before, glad to see I'm not the only one! I'm sure mirroring/synchronous replication will probably have been discussed at Nimble already with a view on how it could be implemented.


          Anyway, re the Storage Virtualisation bit; we've been a long-time FalconStor partner (and naturally have had some exposure to Datacore as well). There wouldn't be any issues, and if anything being able to mirror between stretch clustered nodes could be an attractive option. In my experience, I always prefer FC disk behind FalconStor due to latency and the fact that you don't always have to log into FC targets, they just appear when you rescan the adapters. When using iSCSI storage, you'll have to log into the LUN using iscsiadm every time a new volume is presented, which I find a bit of a pain.


          One of the hurdles you'll face with FalconStor is that Nimble hasn't been certified for use with NSS/CDP, FDS or VTL/SiR, despite it having an ALUA statement. This will be a rather large problem for you with FalconStor Support. Could probably get field certified but that is far from ideal...


          There's also no advantage in having (perhaps a storage pool) with all Nimble volumes in and creating a SAN resource across them, as NSS fills up one allocation from a LUN before going onto the next. Datacore would be able to accelerate writes using the caching, but to be honest there isn't any point in that either - the Nimble arrays are so quick.


          If you want failover of physical/virtual servers between sites and don't already have FalconStor/Datacore, what about host-level HA software? You could use SRM (lots of Nimble documentation and references there) or CA/DoubleTake/NeverFail etc. That's another way to achieve the high availability objective, and depending what product you go for, some can monitor the application in case it's just the data drive on shared storage, and failover the machine if that goes offline (clever stuff).





          • Re: Storage Virtualization and Nimble - makes it sense?



            we just finished a testing scenario with DataCore and Nimble.


            2x Dell R720 Servers, 32GB RAM, FC Frontend, iSCSI 10G Backend, FC Mirror links, SANsymhony-V 10

            2x Nimble CS220G-X2


            Testing compared to a Nimble Box without DataCore:


            1. RandomIO increased by about 30%

            2. Throughput increased by about 300%

            3. FC connectivity at frontend

            4. Sync mirror

            5. "Real life benchmarks" for SQL-Server, Exchange were about the same

            6. Slightly increased latency by mirror handshake


            As of now it seems that DataCore and Nimble are a good combination.


            You loose of course some functionality by Nimble e.g.: Snapshots, Application integration, vCenter plugin.

            DataCore provides those in a different fashion.


            If you want lots of TBs on 3RU units with low power consumption and high IOPs, sync mirror and FC -> go for Nimble + DataCore!