Re: [ovirt-users] gluster config in 4 node cluster
2015-06-01 2:32 GMT+03:00 Soeren Malchow <soeren.malchow(a)mcon.net>:
Hi
For a production environment i would not build a 2 node gluster, i would
build at least 3 nodes to have it much easier with the quorum.
Taking into account that you can use commodity hardware i would also
suggest to split the services, but i would go for at least 3 gluster nodes
– which add up to at least 5 nodes for a HA system.
If you want 4 in any case, then my suggestion would be to go for 4
replicas, each node has anything that it needs to run and you can basically
use NFS to localhost for the storage which would make the nodes always
access the local storage for the VMS, availability wise it makes no
difference.
I have data center with one cluster based on 3 nodes with both services:
virt and gluster, and GlusterFS as a data domain. Every node has a single
HDD as a brick (replica 3) and 2x1G nics with bonding on top (src-dst-ip).
I have 2 vlans: 1 - management, 2 - VMs.
If in a VM I run something like 'dd if=/dev/zero of=/tmp/tmp' I have around
80+ MB/s write speed, at this moment on SPM node i run 'iftop -i bond0.1'
and can see two flows with around 900 Mb/s each of them to the second and
to the third nodes (thanks to the bonding that every flow can be done
through separate nic). All of this is expected result and in this case we
reach a limit of bandwidth our nics. And even if HDD of a node has 150+
MB/s linear write speed I can not reach such speed of write in VMs.
What result would I get with 4 nodes and gluster replica 4? I expect that
write speed in VM will be much slower, because now SPM node will generate
three flows, and two of them will go through one nic and we will have only
500Mb/s per flow and in this way the write speed of VM will be twice lower.
Am I wrong?
Cheers
Soeren
From: Юрий Полторацкий <y.poltoratskiy(a)gmail.com>
Date: Sunday 31 May 2015 18:32
To: "users(a)ovirt.org" <users(a)ovirt.org>
Subject: Re: [ovirt-users] gluster config in 4 node cluster
Hi,
As for me, I would build one cluster with gluster service only based on
two nodes (replica 2), and the other one with virt service only based on
other two nodes. I think this variant is more scalable in future.
PS. I am a new in oVirt, so do not except that I am wrong.
28.05.2015 23:11, paf1(a)email.cz пишет:
Hello,
How to optimal configure 4 node cluster for any one node goes to
maintenance without stopping VM ??
a) replica 4 - but it takes a lot of space
b) disperse 3+1 ( raid 5 ) - but bad performance and not visible by oVirt
3.7.2
c) stripe2+replica2 = but VM paused
any other idea ?
regs.
Pa.
_______________________________________________
Users mailing listUsers@ovirt.orghttp://lists.ovirt.org/mailman/listinfo/users
Attachments:
- attachment.html (text/html — 4.7 KB)