[Gluster-users] QEMU-GlusterFS native integration demo video

Fernando Frediani (Qube) fernando.frediani at qubenet.net
Mon Sep 3 09:12:43 UTC 2012


Hi Bharata,

I am interested to see the performance for read and writes separately. Aggregated will probably be much influenced by the read performance.

Interesting the last figures, seems quiet a significant performance gain using the QEMU-GlusterFS.

See if you can get the results for read and writes separately.

Best regards,

Fernando

-----Original Message-----
From: Bharata B Rao [mailto:bharata.rao at gmail.com] 
Sent: 03 September 2012 10:01
To: Fernando Frediani (Qube)
Cc: gluster-users at gluster.org
Subject: Re: [Gluster-users] QEMU-GlusterFS native integration demo video

On Mon, Sep 3, 2012 at 2:14 PM, Fernando Frediani (Qube) <fernando.frediani at qubenet.net> wrote:
> Hi Bharata
> Thanks for this, very useful.
>
> Would you be able to specify tests with mainly reads and writes. As far as I know there is a big hit and poor performance on writes in normal fuse mounts.

Ok, here are the aggregated bandwidth numbers from a quick FIO (4 seq writes using libaio) write test:

Base: 189667KB/s
QEMU-GlusterFS native: 150635KB/s
QEMU-GlusterFS FUSE: 43028KB/s

So as you can see, native is much better than FUSE, but still doesn't match the base numbers. When I say base, it means that the guest is run directly from glusterfs brick (w/o gluster FUSE mount or using QEMU-GlusterFS native driver).

> Are you using IOmeter or bonnie ?

FIO at the moment. Plan to include more benchmark numbers in future.
Any help here would be appreciated :)

>
> Seems the results with fuse and the native qemu-glusterfs are pretty similar, am I right ?

No, QEMU-GlusterFS native numbers are way better than FUSE. Let me quote the numbers from the URL I gave earlier:

Base: 63076KB/s
QEMU-GlusterFS native: 53609KB/s
QEMU-GlusterFS FUSE: 29392KB/s

Regards,
Bharata.



More information about the Gluster-users mailing list