[Gluster-users] Gluster object-storage additional proxy

Gangalwar gaurav at gluster.com
Sun Oct 30 14:07:56 UTC 2011

Hi Daniel.
If you want to to load balance the load on proxy server across multiple machines then you have to use load balancer, Pound is the one we have tested with, it will load balance across all the backend proxies. This is the only way for load balancing on proxy.
If you don't want proxy and other storage (object/container/account) working on the same node/machine, as you mentioned in the latest code there is provision to setup remote object/container/account server. Let me know if you are facing any issue with this.
If your testing with HTTPS workload then CPU usage will be high due to SSL processing, we observer very low CPU usage with HTTP.

From: gluster-users-bounces at gluster.org [gluster-users-bounces at gluster.org] on behalf of Mohammed Junaid [junaid at gluster.com]
Sent: Friday, October 28, 2011 12:12 PM
To: Daniel Pereira
Cc: gluster-users at gluster.org
Subject: Re: [Gluster-users] Gluster object-storage additional proxy

Hi Daniel,

Please find the comments inline.

Yes, I did that. However the CPU usage gets very high (100% for all cores) to have all the Gluster and Swift daemons running on the same machine. We want to separate the processes in the same manner it is already possible in Swift.

>From what I understand, you have started the Gluster Object Storage on one of the GlusterFS servers and then increased the Proxy server count. If there are any other GlusterFS servers then you can start Gluster Object Storage on those machines and do some load balancing over all these machines by using Pound, nginx etc. This will help reduce the cpu usage of the current machine and also utilize the other server resources.

I read the documentation and there's no mention about separating the swift/object-storage servers in such way. I already have a working setup, I just need to add more proxies.
I could, of course, setup two sets of independent servers and somehow do a messy data sharding between the two servers, but it's simpler and cheaper to scale just the proxy servers as needed (this is almost standard procedure in Swift) since, from our measurements, they are the bottleneck.

There is a difference here how the swift and Gluster Object Storage work. Currently, all the servers (proxy, auth, object, etc) are all coupled to each other and hence they cannot be seperated. This can be emulated by configuring the setup as I mentioned above.

Hope this helps.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20111030/e46ed0c4/attachment.html>

More information about the Gluster-users mailing list