StarWind cache - maximum effective size

[agattuso]

We just purchased Native SAN 2-node HA and I'd like a little clarification on StarWind RAM cache sizes before I purchase 2 new servers to run everything.

In the StarWind High Availability Best Practices document it says the following:

It is recommended to provision 256mb–1gb of [write-back] caching per each terabyte of the HA device’s size. Although, the
maximum recommended cache size is 3gb. for most scenarios, bigger cache is not utilized effectively.

Is the cache used for both read/write or just write? If it's used for both, it seems to me that more would be better, especially on the read side. When I was looking at Nexenta (before I decided to go with SW) they had the concept of WSS (Working Set Size), which is the amount of space most commonly used for applications or workloads. If your WSS fits in your RAM and L2ARC caches you could expect better performance because most of the typical files that were accessed would be in cache and wouldn't have to be pulled from the physical disks. I know StarWind doesn't have SSD caching (yet, v8), but does the same concept apply to the StarWind RAM cache? Will more than 3GB/TB make read speeds better for databases (Exchange & SQL) and other files that get accessed on a daily basis?

On the new servers I'm looking to purchase I'll have 12TB of storage (12x2TB SAS, RAID 10), and I can go with 128GB RAM (4 OS, 36 StarWind cache, 88 Hyper-V) or 256GB (up to 128GB could go toward StarWind cache). I can save $1000 per server if I go with 128GB, but if StarWind can effectively utilize the extra 128GB for it's read cache then I'll go with 256GB in the servers.

Since I'm going with 7200 RPM SAS drives for the physical storage, I want to get the best performance I can out of the RAM cache. What is your recommendation for the amount of RAM in the new servers? I also plan to add SSDs for caching when v8 becomes available if the expectation is that it will improve performance.

Thanks,

Anthony

[anton (staff)]

StarWind caches both reads and writes. Write-Back or Write-Through policy. LRU.

Yes, more you'll assign to StarWind to use - more we'll use and faster you'll go. It's recommended to go with 3x clustering with a huge Write-Back caches however.

Upcoming version will also have In-Memory feature so you'll be able to combine RAM and deduplication to run VMs from memory.