[Gluster-users] Tuning for small files?

Prasad, Nirmal nprasad at idirect.net
Thu Jul 10 14:38:39 UTC 2014


Hi,

It depends on the number of files and typical sizes. We ended up re-organizing them as larger entities on Gluster.

I believe there used to be a Gluster BDB translator for small files but it is no longer an option - you can try to have a similar setup.

http://gluster.org/community/documentation/index.php/Translators/storage/bdb

Ceph has LevelDB support released - the size of the "small" objects plays a key role in overall performance.

http://stackoverflow.com/questions/21726728/the-optimum-size-of-objects-in-ceph-object-storage-rados

If you find a specific setting in gluster that makes a huge difference for tackling the small files please do share.

Thanks
Regards
Nirmal

From: gluster-users-bounces at gluster.org [mailto:gluster-users-bounces at gluster.org] On Behalf Of Jocelyn Hotte
Sent: Thursday, July 10, 2014 8:42 AM
To: gluster-users at gluster.org
Subject: Re: [Gluster-users] Tuning for small files?

Hi, can anyone help me with this issue?

From: Jocelyn Hotte
Sent: 8 juillet 2014 11:16
To: gluster-users at gluster.org<mailto:gluster-users at gluster.org>
Subject: Tuning for small files?

Hello Gluster Community,
I was wondering if there are some things I could do to optimize gluster for small files.

Our bricks are 4 SSD in Raid0, and we have a replica  set factor of 3

We have a use-case where we would like to optimize our gluster mount for small files (about 6kb).

Here's my current benchmarks on my mount

# 1mb block size on RAID0 SSD
dd if=/dev/zero of=/ssd/test_ssd_1mb bs=1M count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 0.0577655 s, 1.8 GB/s

# 1kb block size on RAID0 SSD
dd if=/dev/zero of=/ssd/test_ssd_1kb bs=1K count=100000
100000+0 records in
100000+0 records out
102400000 bytes (102 MB) copied, 0.181732 s, 563 MB/s

# 1mb block size on Gluster mount
dd if=/dev/zero of=/gluster/test_gluster_1mb bs=1M count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 0.219619 s, 477 MB/s

# 1kb block size on Gluster mount
dd if=/dev/zero of=/gluster/test_gluster_1kb bs=1K count=100000
100000+0 records in
100000+0 records out
102400000 bytes (102 MB) copied, 6.73986 s, 15.2 MB/s

As you can see, our performances on small block size are very low.

We're running glusterfs 3.4.2, with the following options:
diagnostics.count-fop-hits: on
diagnostics.latency-measurement: off
performance.cache-max-file-size: 64KB
performance.write-behind-window-size: 8MB
performance.cache-size: 2GB
performance.io-thread-count: 64
performance.flush-behind: off
performance.io-cache: on

Thank you for your help

Jocelyn Hotte
Tools Programmer, Ubisoft
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20140710/b420b7dd/attachment.html>


More information about the Gluster-users mailing list