[Gluster-users] How to achieve linear performance as adding nodes to gluster cluster (distributed replica volume)?

H. Ge harryge00 at gmail.com
Tue Jan 17 04:35:55 UTC 2017


Hi,

I have tested 3 kinds of distributed replica volume: 4 * 2, 3* 2 and 2*2. I
suppose 4 * 2 should achieve the best IOPS, however, their performance
seems similar.

I have tested with "*dd if=/dev/zero of=/mnt/glusterfs/block8 bs=128M
count=1*" and "*dd if=/dev/zero of=/mnt/glusterfs/block8 bs=32M count=8*".
All bricks are on virtual machines,  with same hardware: 2 cores cpu, 8G
memory.

The following is my volume configuration:
Volume Name: rep4
Type: Distributed-Replicate
Volume ID: b2ad2871-cfad-4f2c-afdb-38c2c4d6239c
Status: Started
Number of Bricks: 2 x 2 = 4
Transport-type: tcp
Bricks:
Brick1: 192.168.16.145:/home/vagrant/rep4
Brick2: 192.168.16.146:/home/vagrant/rep4
Brick3: 192.168.16.82:/home/vagrant/rep4
Brick4: 192.168.16.114:/home/vagrant/rep4
Volume Name: rep6
Type: Distributed-Replicate
Volume ID: 2cbcefce-da7a-4823-aee7-432c40f3ae55
Status: Started
Number of Bricks: 3 x 2 = 6
Transport-type: tcp
Bricks:
Brick1: 192.168.16.49:/home/vagrant/rep6
Brick2: 192.168.16.114:/home/vagrant/rep6
Brick3: 192.168.16.141:/home/vagrant/rep6
Brick4: 192.168.16.145:/home/vagrant/rep6
Brick5: 192.168.16.146:/home/vagrant/rep6
Brick6: 192.168.16.82:/home/vagrant/rep6
Volume Name: rep8
Type: Distributed-Replicate
Volume ID: ea77934c-bd5d-4578-8b39-c02402d00739
Status: Started
Number of Bricks: 4 x 2 = 8
Transport-type: tcp
Bricks:
Brick1: 192.168.16.145:/home/vagrant/rep8
Brick2: 192.168.16.146:/home/vagrant/rep8
Brick3: 192.168.16.114:/home/vagrant/rep8
Brick4: 192.168.16.82:/home/vagrant/rep8
Brick5: 192.168.16.141:/home/vagrant/rep8
Brick6: 192.168.16.49:/home/vagrant/rep8
Brick7: 192.168.16.144:/home/vagrant/rep8
Brick8: 192.168.16.112:/home/vagrant/rep8

According to "http://moo.nac.uci.edu/~hjm/Performance_in_a_Gluster_Systemv
6F.pdf",
*> To scale out performance, enterprises need simply add additional storage
server nodes, and will generally see linear performance improvements.*

I wonder how can I achieve  linear performance improvements? Have I tested
in the wrong way?

Regards
Harry(Haoyuan) Ge
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170117/250d04ca/attachment.html>


More information about the Gluster-users mailing list