VSAN (free), 2-node, Hyper-V cluster: Best disk volume caching practices

[infMatejM]

So, we are building a new 2-node hyper converged Hyper-V failover cluster and would appreciate some guidance on best caching practices.

We are using a single HDD (for now) per node to create a VSAN to use as Microsoft Clustered Shared Volume (CSV). Now both Starwind and CSV offers cache options:

StarWind: L1 cache, RAM based, can be write-back or write-through (correct me If I'm wrong)
StarWind: L2 cache, image based (image on an SSD for instance), can only be write-through or can it also work as write-back?
CSV cache: RAM based, read-cache only

From our limited testing (sequential writes only, for now) it seems there are major differences. We were testing with a 35 GB file:

SW L1: on, 128 MB, write-back
SW L2: on, 20 GB SSD image, write-back (I don't know if this setting even works)
CSV cache: off
Result: ~90 MB/s

SW L1: on, 128 MB, write-back
SW L2: off
CSV cache: off
Result: ~110 MB/s

SW L1: off
SW L2: off
CSV cache: off
Result: ~150 MB/s

SW L1: off
SW L2: off
CSV cache: on, 512MB
Result: ~200 MB/s

My question is, is it even recommended to use StarWind cache or what are the recommendations for running a Failover Cluster with purely a Hyper-V role?

Edit: Clustered Shared Volume Cache is apparently read-only, I was mistaken, I fixed my earlier text. But that means that the above results are even more confusing. If CSV cache is read only, why does it have such a big impact on sequential writes?

[yaroslav (staff)]

Welcome to StarWind forum.
I'd rather not use caching for this setup. L1 cache impruves performance, but the system has a significant chance of running in full sync during node turn off/restart, or mutual not synchronized during the blackout.
L2 cache is not HA and boosts only reads. So, does not make much sense for most implementations.
Good luck! Please also consider trialing first if you plan commercial use. Switching to free will be impossible though.