i'm not sure wether i got everything right. you are using glusterfs as a substitute for nfs. if you are using it on infiniband, then you probably could get some performance with combining io-threads, write-behind and as many raid-5 partitions the controller could handle. on ethernet i thing glusterfs isnt yet fast enough for io-killers, but probably it could be performing better than nfs using perfomance-translators on the client side. Io Noci Sean Davis schrieb: > We have a largish NAS, a single linux box, with 40 drives. They are > currently configured as 2 RAID-6 arrays. The machine has two RAID > controllers. This serves as a file system for a small cluster with > about 10 nodes, 60 processors, total. We have some very IO-intensive > applications that simply crush our NAS. Load averages go to 40+ and > we see about 40% wait. The RAID is incredibly fast with about 700 > MB/second serial read, but the sustained concurrent access is much > lower. We would like to look at how gluster could be used to speed up > concurrent access to this single machine. Any insights into using > gluster in this type of situation? We thought about slicing up the 40 > drives into 8 RAID-5 partititions and then serving them using > gluster. We wanted to stay away from full AFR given the increased > cost/TB. We are mainly interested in knowing whether we need another > machine or whether we can make do with our current NAS after > reconfiguring to use gluster. > > Thanks, > Sean > > ------------------------------------------------------------------------ > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users >
