[Gluster-users] Poor performance on a server-class system vs. desktop

[Gionatan Danti]

Il 2020-11-27 09:40 Amar Tumballi ha scritto:
> Let's get to longer look into performance:
> 

Amar, Xavi, thanks for your input - very appreciated.

However, I found that when facing sync writes (ie: fsync) gluster 
performances are very low - too much for a kernel/syscall overhead.

For more information you can read here: 
https://lists.gluster.org/pipermail/gluster-users/2020-January/037601.html

Short summary: with RAM disk based bricks I got ~250 IOPs when 
replicating a 2-way test cluster via 1 Gb/s ethernet, and at most 
500-600 IOPs when running two RAM disk bricks inside the same host (so 
excluding both disk and network overhead).

When not using fsync I was capped by glusterd taking 100% of a single 
CPU core (at about 5500 IOPs, which is not bad on the ancient machine I 
used for testing).

Any suggestion on how to improve fsync speed?
Thanks.

-- 
Danti Gionatan
Supporto Tecnico
Assyoma S.r.l. - www.assyoma.it
email: g.danti at assyoma.it - info at assyoma.it
GPG public key ID: FF5F32A8