[Gluster-users] Slow performance of gluster volume

[Abi Askushi]

Hi Krutika,

I already have a preallocated disk on VM.
Now I am checking performance with dd on the hypervisors which have the
gluster volume configured.

I tried also several values of shard-block-size and I keep getting the same
low values on write performance.
Enabling client-io-threads also did not have any affect.

The version of gluster I am using is glusterfs 3.8.12 built on May 11 2017
18:46:20.
The setup is a set of 3 Centos 7.3 servers and ovirt 4.1, using gluster as
storage.

Below are the current settings:


Volume Name: vms
Type: Replicate
Volume ID: 4513340d-7919-498b-bfe0-d836b5cea40b
Status: Started
Snapshot Count: 0
Number of Bricks: 1 x (2 + 1) = 3
Transport-type: tcp
Bricks:
Brick1: gluster0:/gluster/vms/brick
Brick2: gluster1:/gluster/vms/brick
Brick3: gluster2:/gluster/vms/brick (arbiter)
Options Reconfigured:
server.event-threads: 4
client.event-threads: 4
performance.client-io-threads: on
features.shard-block-size: 512MB
cluster.granular-entry-heal: enable
performance.strict-o-direct: on
network.ping-timeout: 30
storage.owner-gid: 36
storage.owner-uid: 36
user.cifs: off
features.shard: on
cluster.shd-wait-qlength: 10000
cluster.shd-max-threads: 8
cluster.locking-scheme: granular
cluster.data-self-heal-algorithm: full
cluster.server-quorum-type: server
cluster.quorum-type: auto
cluster.eager-lock: enable
network.remote-dio: off
performance.low-prio-threads: 32
performance.stat-prefetch: on
performance.io-cache: off
performance.read-ahead: off
performance.quick-read: off
transport.address-family: inet
performance.readdir-ahead: on
nfs.disable: on
nfs.export-volumes: on


I observed that when testing with dd if=/dev/zero of=testfile bs=1G count=1
I get 65MB/s on the vms gluster volume (and the network traffic between the
servers reaches ~ 500Mbps), while when testing with dd if=/dev/zero
of=testfile bs=1G count=1 *oflag=direct *I get a consistent 10MB/s and the
network traffic hardly reaching 100Mbps.

Any other things one can do?

On Tue, Sep 5, 2017 at 5:57 AM, Krutika Dhananjay <kdhananj at redhat.com>
wrote:

> I'm assuming you are using this volume to store vm images, because I see
> shard in the options list.
>
> Speaking from shard translator's POV, one thing you can do to improve
> performance is to use preallocated images.
> This will at least eliminate the need for shard to perform multiple steps
> as part of the writes - such as creating the shard and then writing to it
> and then updating the aggregated file size - all of which require one
> network call each, which further get blown up once they reach AFR
> (replicate) into many more network calls.
>
> Second, I'm assuming you're using the default shard block size of 4MB (you
> can confirm this using `gluster volume get <VOL> shard-block-size`). In our
> tests, we've found that larger shard sizes perform better. So maybe change
> the shard-block-size to 64MB (`gluster volume set <VOL> shard-block-size
> 64MB`).
>
> Third, keep stat-prefetch enabled. We've found that qemu sends quite a lot
> of [f]stats which can be served from the (md)cache to improve performance.
> So enable that.
>
> Also, could you also enable client-io-threads and see if that improves
> performance?
>
> Which version of gluster are you using BTW?
>
> -Krutika
>
>
> On Tue, Sep 5, 2017 at 4:32 AM, Abi Askushi <rightkicktech at gmail.com>
> wrote:
>
>> Hi all,
>>
>> I have a gluster volume used to host several VMs (managed through oVirt).
>> The volume is a replica 3 with arbiter and the 3 servers use 1 Gbit
>> network for the storage.
>>
>> When testing with dd (dd if=/dev/zero of=testfile bs=1G count=1
>> oflag=direct) out of the volume (e.g. writing at /root/) the performance of
>> the dd is reported to be ~ 700MB/s, which is quite decent. When testing the
>> dd on the gluster volume I get ~ 43 MB/s which way lower from the previous.
>> When testing with dd the gluster volume, the network traffic was not
>> exceeding 450 Mbps on the network interface. I would expect to reach near
>> 900 Mbps considering that there is 1 Gbit of bandwidth available. This
>> results having VMs with very slow performance (especially on their write
>> operations).
>>
>> The full details of the volume are below. Any advise on what can be
>> tweaked will be highly appreciated.
>>
>> Volume Name: vms
>> Type: Replicate
>> Volume ID: 4513340d-7919-498b-bfe0-d836b5cea40b
>> Status: Started
>> Snapshot Count: 0
>> Number of Bricks: 1 x (2 + 1) = 3
>> Transport-type: tcp
>> Bricks:
>> Brick1: gluster0:/gluster/vms/brick
>> Brick2: gluster1:/gluster/vms/brick
>> Brick3: gluster2:/gluster/vms/brick (arbiter)
>> Options Reconfigured:
>> cluster.granular-entry-heal: enable
>> performance.strict-o-direct: on
>> network.ping-timeout: 30
>> storage.owner-gid: 36
>> storage.owner-uid: 36
>> user.cifs: off
>> features.shard: on
>> cluster.shd-wait-qlength: 10000
>> cluster.shd-max-threads: 8
>> cluster.locking-scheme: granular
>> cluster.data-self-heal-algorithm: full
>> cluster.server-quorum-type: server
>> cluster.quorum-type: auto
>> cluster.eager-lock: enable
>> network.remote-dio: off
>> performance.low-prio-threads: 32
>> performance.stat-prefetch: off
>> performance.io-cache: off
>> performance.read-ahead: off
>> performance.quick-read: off
>> transport.address-family: inet
>> performance.readdir-ahead: on
>> nfs.disable: on
>> nfs.export-volumes: on
>>
>>
>> Thanx,
>> Alex
>>
>> _______________________________________________
>> 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/20170905/ac6e6de4/attachment.html>