Separating capacity from performance

[rrbnc]

If we consider an example of, say, needing 8 TB usable storage, and around 4000 IOPS, is it reasonable to use StarWind with a small number of NL SAS, and then add some L2 cache and use a decent size RAM cache? This is compared to, say, trying to cram 24x SAS drives into a chassis which does not make for a cheap server.

Let's say these are our examples:

Main storage: 4x 4TB NL SAS or 8x 2TB NL SAS in RAID10
L2 cache: 800 GB Intel S3700, or 1.6TB Intel P3700 if larger cache is needed
L1 cache: 32 GB RAM (enough?)

And probably need LSFS enabled?

I guess the question is, can we estimate IOPS based on cache sizes, or if we give our needs to StarWind engineers (usable storage and IOPS required) will they know what to spec based on lab testing and customer base?

[anton (staff)]

That depends on your workload. With read caches it's always a question how effective they are in predicting your I/O pattern. So you may or may not get what you want. Write caches are more immune as the only thing you need to have to make them happy - pulsating traffic so "lazy writer" would have time to flush data to disk (LSFS can do that very fast).