Gluster users - Oct 2018 - High CPU usage with 3 bricks

Hi,

I have a server setup on a powerful processor (i7-8700). I have three
bricks setup on localhost - each their own hard drive. When I'm downloading
content at 100MB/s I see upwards of 180% CPU usage from GlusterFS.

Uploading content at 100MB/s is much more manageable, with around 50%-70%
cpu usage combined from all the glusterfs processes.

My question is, is this normal CPU usage? I thought GlusterFS was designed
to be used in a RAID configuration with having an internal network setup
with upwards of 20Gbps connectivity per node. If 1Gbps write speed is
causing 180% cpu usage on only three hard drives/bricks, how could a raid
setup at 20Gbps possibly be feasible?

This is a default distributed volume. Just the standard 3-4 commands to get
the volume created, started and running. There has been no additional
tuning done.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.gluster.org/pipermail/gluster-users/attachments/20181004/d3f2ed9c/attachment.html>

On Fri, Oct 5, 2018 at 1:54 AM Tyler Salwierz <tyler at irex.me> wrote:
> Hi,
>
> I have a server setup on a powerful processor (i7-8700). I have three
> bricks setup on localhost - each their own hard drive. When I'm
downloading
> content at 100MB/s I see upwards of 180% CPU usage from GlusterFS.
>
> Uploading content at 100MB/s is much more manageable, with around 50%-70%
> cpu usage combined from all the glusterfs processes.
>
Do you see the high CPU usage of clients and bricks?

>
> My question is, is this normal CPU usage? I thought GlusterFS was designed
> to be used in a RAID configuration with having an internal network setup
> with upwards of 20Gbps connectivity per node. If 1Gbps write speed is
> causing 180% cpu usage on only three hard drives/bricks, how could a raid
> setup at 20Gbps possibly be feasible?
>
This looks to be an anomaly, especially with reads/writes. Is it possible
to attach output of "top -bHd >> top.txt"?

Also recently there was a similar thread (though the issue was not resolved)
https://lists.gluster.org/pipermail/gluster-users/2018-July/034414.html

> This is a default distributed volume. Just the standard 3-4 commands to
> get the volume created, started and running. There has been no additional
> tuning done.
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> https://lists.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.gluster.org/pipermail/gluster-users/attachments/20181005/f4fdfda0/attachment.html>