Well, that was fun....
After a few weeks of trying various storage setups to help reduce latency for certain VM's on a clients VMware cluster, I can report back that a SRP Target on a ZFS server worked wonders!
Not sure I'd ever put this thing into production though, it was hell to get working (easy on the VMware side), but on Linux and Solaris, it took quite a bit of effort to get it going. Certainly not a KISS approach, but it did work wonders, as you can see by the attached graph.
Plus the fact that the ZFS storage server isn't redundant, it is a single unit, to make it redundant I need a shared SAS JBOD behind a couple of head-ends.
The attached graph shows the VM running normally, which is backed by a group of 3 EQL arrays, with 4 x GE doing MPIO to the arrays. Around 5:30PM the VM is migrated to the SRP Target backed by a ZFS server.
After a few weeks of trying various storage setups to help reduce latency for certain VM's on a clients VMware cluster, I can report back that a SRP Target on a ZFS server worked wonders!
Not sure I'd ever put this thing into production though, it was hell to get working (easy on the VMware side), but on Linux and Solaris, it took quite a bit of effort to get it going. Certainly not a KISS approach, but it did work wonders, as you can see by the attached graph.
Plus the fact that the ZFS storage server isn't redundant, it is a single unit, to make it redundant I need a shared SAS JBOD behind a couple of head-ends.
The attached graph shows the VM running normally, which is backed by a group of 3 EQL arrays, with 4 x GE doing MPIO to the arrays. Around 5:30PM the VM is migrated to the SRP Target backed by a ZFS server.