Like Eric, I too use lvm to partition off bricks for different volumes. You can even specify which physical device a brick is on when you're creating your brick, ie. "lvcreate -n myvol_brick_a -l50 vg_gluster /dev/sda1". This is handy if you have to replace the disk while the old one is still alive as you can just install the replacement and do a "pvmove". Each brick uses memory. I have 15 volumes, 4 disks per server, and one brick per volume per disk. 60 bricks would use a lot of memory. Rather than buy a bunch more memory that would usually sit idle, I set "performance.cache-size" to a number that would use up just enough memory. Do experiment with that setting. It appears that size limit is used for multiple caches so the actual memory used seems to be some multiple of what you set it to. When a server comes back after maintenance and the self-heals start, having multiple bricks healing simultaneously can put quite a load on your servers. Test that and see if it meets your satisfaction. I actually kill bricks for non-essential volumes while the essential volumes are healing, then use "volume start ... force" to start the bricks for the degraded volumes individually to manage that. On 11/12/2013 08:21 AM, Eric Johnson wrote: > I would suggest using different partitions for each brick. We use LVM > and start off with a relativity small amount allocated space, then > grow the partitions as needed. If you were to place 2 bricks on the > same partition then the free space is not going to show correctly. > Example: > > 1TB partition 2 bricks on this partition > > brick: vol-1-a using 200GB > brick: vol-2-a using 300GB. > > Both volumes would show that they have ~500GB free, but in reality > there would be ~500GB that either could use. I don't know if there > would be any other issues with putting 2 or more bricks on the same > partition, but it doesn't seem like a good idea. I had gluster setup > that way when I was first testing it, and it seemed to work other than > the free space issue, but I quickly realized it would be better to > separate out the bricks on to their own partition. Using LVM allows > you to easily grow partitions as needed. > > my 2 cents. > > > On 11/12/13, 9:31 AM, David Gibbons wrote: >> Hi All, >> >> I am interested in some feedback on putting multiple bricks on one >> physical disk. Each brick being assigned to a different volume. Here >> is the scenario: >> >> 4 disks per server, 4 servers, 2x2 distribute/replicate >> >> I would prefer to have just one volume but need to do geo-replication >> on some of the data (but not all of it). My thought was to use two >> volumes, which would allow me to selectively geo-replicate just the >> data that I need to, by replicating only one volume. >> >> A couple of questions come to mind: >> 1) Any implications of doing two bricks for different volumes on one >> physical disk? >> 2) Will the "free space" across each volume still calculate >> correctly? IE, if one volume takes up 2/3 of the total physical disk >> space, will the second volume still reflect the correct amount of >> used space? >> 3) Am I being stupid/missing something obvious? >> >> Cheers, >> Dave >> >> >> _______________________________________________ >> Gluster-users mailing list >> Gluster-users at gluster.org >> http://supercolony.gluster.org/mailman/listinfo/gluster-users > > > -- > Eric Johnson > 713-968-2546 > VP of MIS > Internet America > www.internetamerica.com > > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://supercolony.gluster.org/mailman/listinfo/gluster-users -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20131112/09c0532a/attachment.html>
