Hello folks--
I've finally gotten my Esxi 6.5 server running, with mutiple win10 vm's running on an underlying NVME filestore (actually a Samsung 960 Evo m.2 drive). Thanks to VMware for making this great package available to hobbyists like me.
I have a high RAM server (256GB) with a 1TB (physical) NVME drive. Dual Xeon 2690 V1 CPU's. The 4 clients are running the same OS, so I have deduplication/compression running to save HD space. Each client only uses 16GB, which is reserved, so I have a lot of extra RAM. The VM's are using virtual NVME controllers attached to the virtual disks on the local datastore (no nVME passthrough).
I'd like to use the extra RAM to cache the datastore: I expect that would improve latency and throughput signficantly when working with the VM's, even over the fast Samsung NVME drive. I can't for the life of me find any documentation on VMware on how to do that. There are zillions of posts on how to do the reverse of what I'm trying to do (use SSD's to cache oversubscribed RAM). Also, for my applications, I'm not afraid of using write caching (which I also can't figure out how to turn on, but would like to.)
Ideas:
1.I assume the "right" answer is "use a real SAN"; that's probably out for cost reasons (and, from a hardware viewpoint, shouldn't be needed-- hard to beat the throughput of local RAM).
2. Install a windows 2016 server VM on the same host with iSCSI target role and use Windows file caching on that? Seems suboptimal as compared to Esxi doing this iteself. This would also bottleneck things over VMXNet3 adapters, which are mimicking 10Gbe adapters-- much slower than the native RAM should be capable of.
Is there a "native" Esxi way to do this ? What other "additional" software packages on the host might do the job?
Thanks!
