Re: [ovirt-users] [Gluster-users] GlusterFS for production use with Ovirt

A few things to clarify 1) are all 4 for Gluster or are they the whole cluster. keep in mind that Ovirt expects that you are running Gluster on different nodes than your VM's. while I do not completely agree with that idea that is how it is now. 2) RAID hardware is not necessarily a bad thing, in fact the cache may help significantly. that said it is not required. things to keep in mind if you are using RAID hardware for performance 2.a) RAID 0 will give you the best speed if you are using 3 nodes for quorum. keep in mind this will require more work in the case of a drive failure to remake the brick. 2.b) if you are only doing a mirror across 2 bricks RAID 0 may be acceptable but consider using 1+0 or RAID 5 the performance on the two may vary depending on the RAID controller but 1+0 is generally faster. 2.c) always optimize for the stripe which in theory should be automatic if the RAID controller reports the required information to do it, but in practice most do not so you should do it manually and Red Hat's support site just added a nice tool to do it for you. 3) for performance more RAM is rarely a bad thing; however it can be with sustained writes.If you are constantly writing faster than the disks can handle you will eventually fill the buffers and block till they flush. so the answer here is if you are mostly dealing with reads more and bursts of writes more ram is better. if you are dealing with constant writes it may cause uneven performance and may in fact be detrimental. 4) On the 10Gb consider using consider using RDMA Iverbs over Ethernet it may help. 5) Look closely at the scheduler on the disks the default these days is CFQ (Completely Fair Queuing). I find DEADLINE works better for Gluster; however depending on if you are using a RAID and you have a properly aligned file system you may find NOOP may provide better performance you need to do some testing with your hardware to determine if this is the case. On Mon, Mar 23, 2015 at 9:43 PM, Punit Dambiwal <hypunit(a)gmail.com&gt; wrote: