11:59 a.m.
Hi Sahina,
Thanks for your reply.
Let me share my test results with gluster v3 .
I have a 3-node hyperconverged setup with 1 Gbit/s network and SSD (sata-based) for LVM
caching.
Testing the bricks showed higher than the network performance.
1. Tested ovirt 4.2.7/4.2.8 with fuse mounts. Using 'dd if=/dev/zero
of=/default/gluster/Mount point/from/ovirt bs=1M count=5000'.
Results: 56MB/s directly on the mount point, 20+-2 MB/s from a VM.
Reads on fuse mount point -> 500+ MB/s
Disabling sharing increased performance from the fuse mount point, nothing beneficial from
a VM.
Converting the bricks of a volume to 'tmpfs' does not bring any performance gain
for FUSE mount.
2. Tested ovirt 4.2.7/4.2.8 with gfapi - performance in VM -> approx 30 MB/s
3. Gluster native NFS (now deprecated) on ovirt 4.2.7/4.2.8 -> 120MB/s on mount point,
100+ MB/s in VM
My current setup:
Storhaug + ctdb + nfs-ganesha (ovirt 4.2.7/4.2.8) -> 80+-2 MB/s in the VM. Reads are
around the same speed.
Sadly, I didn't have the time to test performance on gluster v5 (ovirt 4.3.0) , but
I haven't noticed any performance gain for the engine.
My suspicion with FUSE is that when a gluster node is also playing the role of a client
-> it still uses network bandwidth to communicate to itself, but I could be wrong.
According to some people on the gluster lists, the FUSE performance is expected ,but my
tests with disabled sharing shows better performance.
Most of the time 'gtop' does not show any spikes and iftop shows that network
usage is not going over 500 Mbit/s
As I hit some issues with the deployment on 4.2.7 , I decided to stop my tests for now.
Best Regards,
Strahil Nikolov
On Feb 25, 2019 09:17, Sahina Bose <sabose(a)redhat.com> wrote:
The options set on the gluster volume are tuned for data consistency and reliability.
Some of the changes that you can try
1. use gfapi - however this will not provide you HA if the server used to access the
gluster volume is down. (the backup-volfile-servers are not used in case of gfapi). You
can change this using the engine-config tool for your cluster level.
2. Change the remote-dio:enable to turn on client side brick caching. Ensure that you
have a power backup in place, so that you don't end up with data loss in case server
goes down before data is flushed.
If you're seeing issues with a particular version of glusterfs, please provide
gluster profile output for us to help identify bottleneck. (check
https://docs.gluster.org/en/latest/Administrator%20Guide/Monitoring%20Wor... on how to
do this)
On Fri, Feb 22, 2019 at 1:39 PM Sandro Bonazzola <sbonazzo(a)redhat.com> wrote:
>
> Adding Sahina
>
> Il giorno ven 22 feb 2019 alle ore 06:51 Strahil <hunter86_bg(a)yahoo.com> ha
scritto:
>>
>> I have done some testing and it seems that storhaug + ctdb + nfs-ganesha is
showing decent performance in a 3 node hyperconverged setup.
>> Fuse mounts are hitting some kind of limit when mounting gluster -3.12.15
volumes.
>>
>> Best Regards,
>> Strahil Nikolov
>>
>> _______________________________________________
>> Users mailing list -- users(a)ovirt.org
>> To unsubscribe send an email to users-leave(a)ovirt.org
>> Privacy Statement: https://www.ovirt.org/site/privacy-policy/
>> oVirt Code of Conduct:
https://www.ovirt.org/community/about/community-guidelines/
>> List Archives:
https://lists.ovirt.org/archives/list/users@ovirt.org/message/7U5J4KYDJJS...
>
>
>
> --
>
> SANDRO BONAZZOLA
>
> MANAGER, SOFTWARE ENGINEERING, EMEA R&D RHV
>
> Red Hat EMEA
>
> sbonazzo(a)redhat.com