also, after defragging, and specifying *NEW* files in my iozone benchmark, the load of reads/writes is now more distributed, but i'm wondering which did the trick? -----Original Message----- From: gluster-users-bounces at gluster.org [mailto:gluster-users-bounces at gluster.org] On Behalf Of Harshavardhana Sent: Thursday, June 24, 2010 2:10 AM To: gluster-users at gluster.org Subject: Re: Brick Preference On 06/23/2010 01:53 PM, Andy Pace wrote: > Just wondering how Gluster determines which brick to send files to. I've got 4 bricks setup right now, and benchmarking 4 clients against them, and it seems that all the clients prefer bricks 3 and 4, rather than 1 and 2... > > It's setup in a replicating/distributing model. > > Is there a way to 'weight' them? It would make sense for doing this on commodity hardware, as Gluster advertises - send less traffic to smaller bricks, and more to the bigger bricks. > > > > Hi Andy, Are you using the brick4 and brick4 from the same set of subvolume from the last time?. Since updating the replicate/distribute pairs you might need to run defrag.sh scripts to repopulate the hash values properly. Also this is the case when you don't defrag the older directories on old exports gets files written to their own set of pairs. Only new directories would help in creating files across evenly. Regards -- Harshavardhana Gluster Inc - http://www.gluster.com +1(408)-770-1887, Ext-113 +1(408)-480-1730 _______________________________________________ Gluster-users mailing list Gluster-users at gluster.org http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
