[Gluster-users] Is there any performance impact in setting up every gluster client as a NFS server?

Jeevan Patnaik g1patnaik at gmail.com
Wed Nov 15 04:47:47 UTC 2017


What I meant was that every host acts as nfs server for itself alone. A is
mounted on A, B is mounted on B etc. So, if a host is crashed or storage of
the host is not available, other hosts won't be affected.

And how bad can be the performance overhead in having multiple nfs servers?


On Nov 15, 2017 10:07 AM, "Soumya Koduri" <skoduri at redhat.com> wrote:


On 11/14/2017 11:45 PM, Jeevan Patnaik wrote:

> Hi,
> We have around 60 hosts and each of them acts as glusterFs clients as well
> as server.
> To achieve HA, my underatanding is that we can use Ganesha NFS alone (and
> not Kernel NFS) and for above 3.10 versions, the HA packages are not ready
> yet.
> So, I'm thinking if I can avoid HA by making each client acts as server
> for itself.
> i.e.,the server name specified while mounting is localhost.

IIUC, you are planning to use gluster native client (via FUSE) mounts act
as NFS server by exporting those mount paths via Kernel-NFS/NFS-Ganesha. If
thats the case, yes it shall add a bit of overhead to the performance.

Also without HA packages, wouldn't that NFS server become single point of
failure to its NFS clients. How will you ensure the IPs used by that
particular host migrates to another one.


> So, if a host have issues, then it won't affect other hosts and no need
> for HA.
> But will there be any performance impact in having too many glusterFs
> servers?
> Regards,
> Jeevan.
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://lists.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20171115/86c06ed2/attachment.html>

More information about the Gluster-users mailing list