Re: performance impact of various compression schemes on a zvol

On Wed, Oct 18, 2023 at 09:21:29AM +0200, infoomatic wrote:
>On 18.10.23 00:33, void wrote:
>>What's the perfomance penalty on zvols of compression?
>>The impact on the host?
>
>This totally depends on your data. I did some tests on our mailserver,
>and for us it boiled down to:
>
>*) use lz4 if performance is very important, and compression rate not so
>much. We found that the impact of lz4 is so fast that our workloads were
>quite faster than any test without compression, so I would say you can
>set lz4 in any way (lz4 is very fast on detecting incrompressible data)
>
>*) use between zstd-6 and zstd-9 (not too much difference in our tests)
>if you want a nice balance where performance is not too much affected
>and we reach a nice level of compression, however, of course your
>definition of "maximum performance penalty" may be different.
>
>You could check out lzbench.

Thanks everyone, this is exactly the type of info I was looking for.

In terms of data used within the vm, I'd guess 70% of the space 
currently used is uncompressible multimedia. I'll stay with lz4
(when file-backed the vm was on a lz4 vdev. I'll set the vol to 
use lz4). There is no filesystem compression within the vm as
it's a linux guest which pre-dates zfs use within linux).
--