[Gluster-users] Advice needed for network change.

[Strahil]

Hey All,

I have managed to migrate but with some slight changes.
I'm using teaming with balance runner (l3 + l4 for balance hashing) on 6 ports (2 dual port + 2 nics using only 1 port  ) per machine.
Running some tests (8 network connections in parallel)  between the replica nodes shows  an aggregated  bandwidth of 400+ MB/s (megabytes).

Now I need to 'force' gluster to open more connections in order to spread the load to as many ports as possible.
I have tried with client & server event-threads set to 6 , but the performance is bellow my expectations .

Any hints will be appreciated.

Best Regards,
Strahil Nikolov On May 10, 2019 13:36, Strahil <hunter86_bg at yahoo.com> wrote:
>
> Hello Community,
>
> I'm making some changes and I would like to hear  your opinion on the topic.
>
> First, let me share my setup.
> I have 3 systems with in a replica 3 arbiter 1 hyperconverged setup (oVirt) which use 1 gbit networks for any connectivity.
>
> I have added 4 dual-port 1 gbit NICs ( 8 ports per machine in total) and connected them directly between ovirt1 and ovirt2 /data nodes/ with LACP aggregation (layer3+layer4 hashing).
>
> As ovirt1 & ovirt2 are directly connected /trying to reduce costs by avoiding the switch/ I have  setup /etc/hosts for the arbiter  /ovirt3/ to point tothe old IPs .
>
> So they look like:
> ovirt1 & ovirt2 /data nodes/  /etc/hosts:
>
> 127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4
> ::1         localhost localhost.localdomain localhost6 localhost6.localdomain6
> 192.168.1.90 ovirt1.localdomain ovirt1
> 192.168.1.64 ovirt2.localdomain ovirt2
> 192.168.1.41 ovirt3.localdomain ovirt3
> 10.10.10.1   gluster1.localdomain gluster1
> 10.10.10.2   gluster2.localdomain gluster2
>
> ovirt3 /etc/hosts:
>
> 127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4
> ::1         localhost localhost.localdomain localhost6 localhost6.localdomain6
> #As gluster1 & gluster2 are directly connected , we cannot reach them.
> 192.168.1.90 ovirt1.localdomain ovirt1 gluster1
> 192.168.1.64 ovirt2.localdomain ovirt2 gluster2
> 192.168.1.41 ovirt3.localdomain ovirt3
>
> Do you see any obstacles to 'peer probe' and then 'replace brick' the 2 data nodes.
> Downtime is not an issue, but I preffer not to wipe the setup.
>
> Thanks for reading this long post and don't hesitate to recommend any tunings.
> I am still considering what values to put for the  client/server thread count.
>
> Best Regards,
> Strahil Nikolov
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20190513/5ab54a2f/attachment.html>