[Gluster-users] Glusterfs performance tweaks

[Ben Turner]

----- Original Message -----
> From: "Punit Dambiwal" <hypunit at gmail.com>
> To: "Ben Turner" <bturner at redhat.com>
> Cc: "Vijay Bellur" <vbellur at redhat.com>, gluster-users at gluster.org
> Sent: Thursday, April 9, 2015 9:36:59 PM
> Subject: Re: [Gluster-users] Glusterfs performance tweaks
> 
> Hi Ben,
> 
> But without glusterfs if i run the same command with dsync on the same
> ssd...it gives me good throughput...all setup (CPU,RAM,Network are same)
> the only difference is no glusterfs...
> 
> [root at cpu09 mnt]# dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> 4096+0 records in
> 4096+0 records out
> 268435456 bytes (268 MB) copied, 0.935646 s, 287 MB/s
> [root at cpu09 mnt]#
> 
> [image: Inline image 1]
> 
> But on the top of the glusterfs it gives too slow performance....i run the
> ssd trim every night to clean the garbage collection...i think there is
> something need to do from gluster or OS side to improve the
> performance....otherwise no use to use the ALL SSD with gluster because
> with all SSD you will get the performance slower then SATA....
> 
> 
> 
> On Fri, Apr 10, 2015 at 2:12 AM, Ben Turner <bturner at redhat.com> wrote:
> 
> > ----- Original Message -----
> > > From: "Punit Dambiwal" <hypunit at gmail.com>
> > > To: "Vijay Bellur" <vbellur at redhat.com>
> > > Cc: gluster-users at gluster.org
> > > Sent: Wednesday, April 8, 2015 9:55:38 PM
> > > Subject: Re: [Gluster-users] Glusterfs performance tweaks
> > >
> > > Hi Vijay,
> > >
> > > If i run the same command directly on the brick...

What does this mean then?  Running directly on the brick to me means running directly on the SSD.  The command below is the same thing as above, what changed?

-b

> > >
> > > [root at cpu01 1]# dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> > > 4096+0 records in
> > > 4096+0 records out
> > > 268435456 bytes (268 MB) copied, 16.8022 s, 16.0 MB/s
> > > [root at cpu01 1]# pwd
> > > /bricks/1
> > > [root at cpu01 1]#
> > >
> >
> > This is your problem.  Gluster is only as fast as its slowest piece, and
> > here your storage is the bottleneck.  Being that you get 16 MB to the brick
> > and 12 to gluster that works out to about 25% overhead which is what I
> > would expect with a single thread, single brick, single client scenario.
> > This may have something to do with the way SSDs write?  On my SSD at my
> > desk I only get 11.4 MB / sec when I run that DD command:
> >
> > # dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> > 4096+0 records in
> > 4096+0 records out
> > 268435456 bytes (268 MB) copied, 23.065 s, 11.4 MB/s
> >
> > My thought is that maybe using dsync is forcing the SSD to clean the data
> > or something else before writing to it:
> >
> > http://www.blog.solidstatediskshop.com/2012/how-does-an-ssd-write/
> >
> > Do your drives support fstrim?  It may be worth it to trim before you run
> > and see what results you get.  Other than tuning the SSD / OS to perform
> > better on the back end there isn't much we can do from the gluster
> > perspective on that specific DD w/ the dsync flag.
> >
> > -b
> >
> > >
> > > On Wed, Apr 8, 2015 at 6:44 PM, Vijay Bellur < vbellur at redhat.com >
> > wrote:
> > >
> > >
> > >
> > > On 04/08/2015 02:57 PM, Punit Dambiwal wrote:
> > >
> > >
> > >
> > > Hi,
> > >
> > > I am getting very slow throughput in the glusterfs (dead slow...even
> > > SATA is better) ... i am using all SSD in my environment.....
> > >
> > > I have the following setup :-
> > > A. 4* host machine with Centos 7(Glusterfs 3.6.2 | Distributed
> > > Replicated | replica=2)
> > > B. Each server has 24 SSD as bricks…(Without HW Raid | JBOD)
> > > C. Each server has 2 Additional ssd for OS…
> > > D. Network 2*10G with bonding…(2*E5 CPU and 64GB RAM)
> > >
> > > Note :- Performance/Throughput slower then Normal SATA 7200 RPM…even i
> > > am using all SSD in my ENV..
> > >
> > > Gluster Volume options :-
> > >
> > > +++++++++++++++
> > > Options Reconfigured:
> > > performance.nfs.write-behind- window-size: 1024MB
> > > performance.io-thread-count: 32
> > > performance.cache-size: 1024MB
> > > cluster.quorum-type: auto
> > > cluster.server-quorum-type: server
> > > diagnostics.count-fop-hits: on
> > > diagnostics.latency- measurement: on
> > > nfs.disable: on
> > > user.cifs: enable
> > > auth.allow: *
> > > performance.quick-read: off
> > > performance.read-ahead: off
> > > performance.io-cache: off
> > > performance.stat-prefetch: off
> > > cluster.eager-lock: enable
> > > network.remote-dio: enable
> > > storage.owner-uid: 36
> > > storage.owner-gid: 36
> > > server.allow-insecure: on
> > > network.ping-timeout: 0
> > > diagnostics.brick-log-level: INFO
> > > +++++++++++++++++++
> > >
> > > Test with SATA and Glusterfs SSD….
> > > ———————
> > > Dell EQL (SATA disk 7200 RPM)
> > > —-
> > > [root at mirror ~]#
> > > 4096+0 records in
> > > 4096+0 records out
> > > 268435456 bytes (268 MB) copied, 20.7763 s, 12.9 MB/s
> > > [root at mirror ~]# dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> > > 4096+0 records in
> > > 4096+0 records out
> > > 268435456 bytes (268 MB) copied, 23.5947 s, 11.4 MB/s
> > >
> > > GlsuterFS SSD
> > > —
> > > [root at sv-VPN1 ~]# dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> > > 4096+0 records in
> > > 4096+0 records out
> > > 268435456 bytes (268 MB) copied, 66.2572 s, 4.1 MB/s
> > > [root at sv-VPN1 ~]# dd if=/dev/zero of=test bs=64k count=4k oflag=dsync
> > > 4096+0 records in
> > > 4096+0 records out
> > > 268435456 bytes (268 MB) copied, 62.6922 s, 4.3 MB/s
> > > ————————
> > >
> > > Please let me know what i should do to improve the performance of my
> > > glusterfs…
> > >
> > >
> > > What is the throughput that you get when you run these commands on the
> > disks
> > > directly without gluster in the picture?
> > >
> > > By running dd with dsync you are ensuring that there is no buffering
> > anywhere
> > > in the stack and that is the reason why low throughput is being observed.
> > >
> > > -Vijay
> > >
> > > -Vijay
> > >
> > >
> > >
> > > _______________________________________________
> > > Gluster-users mailing list
> > > Gluster-users at gluster.org
> > > http://www.gluster.org/mailman/listinfo/gluster-users
> >
>