Performance Best Practices - V8 and Server 2012 R2

[anton (staff)]

These are recommended values. Going lower is OK you're going to have less chances to have a reasonable cache hit (which is pure math in case of reads).

[robnicholson]

I appreciate that but guidance based upon realistic real-world configurations would be much appreciated. Nobody except the real petrol heads/more money than sense users are going to set-up that level of caching. I know you keep aiming for uber-performance but whilst we'd all love to drive a Ferrari, most of us are stuck with a Ford.

Won't it have to be more of a sliding scale. The best size for the cache is more a factor of what percentage of the overall volume size is in regular use over a certain period? For example, it's moot that one may have a 32TB disk volume because you've got lots of data. Isn't what's important the rate at which data is been written, i.e. the best overall cache size is that which means that on average normal load, writes are working entirely in cache.

For example, with that 32TB array, if you have one user writing one byte every minute, then a one byte cache is perfectly adequate. It's moot whether that array is 1MB or 100TB or 100PB.

A more accurate idea of this could be achieved maybe by analysing the I/O through the SAN on a per volume basis. Some nice graphs maybe.

But isn't all this moot. Doesn't v8 do dynamic caching? So you tell Starwind to use as much memory as it can balanced real-time across the volumes? So if you have three volumes like this:

1. Exchange & SQL
2. File server
3. Archive server

You have 40GB free RAM. Exchange & SQL are really hammering the SAN, file server is busy but not stupid stilly and the archive server is pretty idle. So Starwind dynamically allocates a 30GB cache to Exchange & SQL, 9.5GB to the file server and just 500MB to the archive server.

But then somebody starts writing some large files to the file server so over a minute, it's usage backs up (cache starts overflowing), so the cache sizes are adjusted so it's more balanced between #1 and #2?

This is all guesswork. Does StarWind work this way?

Cheers, Rob.

[anton (staff)]

Like I've told dynamic cache is out. We're working on implementation but early previews would be in September 2014 only.

Back to your questions. You don't tell MAIN thing: are you running hyper-converged (same hardware for Hyper-V & StarWind) or dedicated (separated layers)?

Assuming separated layers.

1) SQL Server and Exchange. Use StarWind to build SoFS back end and use SMB 3.0 to talk from these apps to a virtual shared storage.

2) File Server. Similar to 1. We provide a block back end so use it to build guest VM cluster for SMB 3.0 file server (no SoFS!!!)

3) Not a good scenario for StarWind.

Assuming dedicated.

1) and 2) are similar, virtual LU/CSV in a loopback. 3) is out. Use Windows dedupe for archives. We're primary VM storage thing.

[robnicholson]

>Back to your questions. You don't tell MAIN thing: are you running hyper-converged (same hardware for Hyper-V & StarWind) or dedicated (separated layers)?

We're not running anything on Starwind v8 yet. I'm currently evaluating it in the lab to determine whether we stay with the product (still on v6) or consider an alternative. And therefore I need to understand all the sizing requirements in advance so I can do a real cost comparison.

We may or may not run Starwind on a dedicated server or on the Hyper-V hosts themselves. As you can see from the post I made on Spiceworks, I really do like the idea of a) less tin in the computer room and b) less cost.

If you say we need 1.6TB of RAM (5%) for a 32TB archive disk system, then Starwind is not a suitable product. But of course, you're not saying that so I'm asking, in your opinion, what would be a reasonable cache size for a volume of that size where uber performance is not critical. We'll have smaller volumes for that.

And given I'm asking this question, you can guarantee that every person sizing up Starwind will be asking the same question.

Cheers, Rob.

[robnicholson]

Like I've told dynamic cache is out.

That is a real shame. Do I there infer there is nothing special about the Hyper-V edition of Starwind? At no point in the code is there a line that says "IF running on Hyper-V THEN do something else"? It's exactly the same product as the version you'd run on a non-Hyper-V native server?

I'm just trying to get to the bottom of all this because my job is on the line if we choose a system based upon my recommendation and it turns out I don't know what I'm talking about, I'll look very foolish.

Cheers, Rob.

[Anatoly (staff)]

The code is the same everywhere. The license makes the difference.
Don`t get me wrong - we did improvements for Hyper-V converged scenarios - loopback accelerator is one of them.

[zfllj]

So no Storage Spaces to present volumes to SW VSAN? Must be hardware based RAID?

[anton (staff)]

I have no idea how you came to that conclusion... Storage Spaces and anything having native 4KB blocks are fully supported with StarWind with V8.

So no Storage Spaces to present volumes to SW VSAN? Must be hardware based RAID?