It's always interesting seeing people advising against a solution without actually having it tested :)
Thanks John for sharing your results. I guess that you're using XFS as the underlying FS ? Your write-behind options, clearly show a performance improvement for write speed. Have you tried with VM cache enabled? What about read performances? I also hope that qemu native integration will be soon supported in Cloudstack, performances are definitely boosted :) BTW are you happy with the cloudstack KVM integration? Seems, reading this list, that there's some "needed" features that doesn't work very well with KVM - HA, VM snapshots,... Cheers, Olivier On 2013-09-11 16:38, John Skinner wrote: > I currently have GlusterFS deployed into an 8 node KVM cluster running on CloudStack 4.1 for primary storage. Gluster is deployed on 28 1TB drives across 2 separate storage appliances using a distributed-replicated volume with the replica set to 2. The storage network is 10Gb copper. > > These are the options I have configured for the volume in Gluster, most of them are from a Red Hat document on configuring Red Hat Enterprise Storage for VM hosting: > > performance.io-thread-count: 32 > performance.cache-size: 1024MB > performance.write-behind-window-size: 5MB > performance.write-behind: on > network.remote-dio: on > cluster.eager-lock: enable > performance.stat-prefetch: off > performance.io-cache: on > performance.read-ahead: on > performance.quick-read: on > > Here are some of the numbers I was getting when benchmarking the storage from the KVM node directly (not a VM) > > The below table is in KB/s. The test is single stream 1GB file utilizing Direct I/O (no cache). I used iozone to run the benchmark. > > Write 4k 45729 > Read 4k 10189 > Random Write 4k 31983 > Random Read 4k 9859 > Write 16k 182246 > Read 16k 37146 > Random Write 16k 113026 > Random Read 16k 37237 > Write 64k 420908 > Read 64k 125315 > Random Write 64k 383848 > Random Read 64k 125218 > Write 256k 567501 > Read 256k 218413 > Random Write 256k 508650 > Random Read 256k 229117 > > In the above results, I have the volume mounted to each KVM host as a FUSE glusterfs file system. They are added to CloudStack as a shared mount point. In the future it would be great to utilize GlusterFS qemu libvirt integration with libgfapi so I could bypass fuse altogether. However, that would require adding that code to CloudStack to support that. > > I maybe have 15 or so VMs running from the storage now and it is still pretty snappy. Need to do some more testing though and really get it loaded. > > ----- Original Message ----- > > From: "Rafael Weingartner" <rafaelweingart...@gmail.com> > To: users@cloudstack.apache.org > Sent: Wednesday, September 11, 2013 8:48:07 AM > Subject: Re: Cloustack 4.1.0 + GlusterFS > > Right now I can think in three main reasons: > > The first reason, performance (I do not know Gluster and its performance > and if the read and write speed are satisfactory). Please if you can make a > test, post the results. > > Second consistency, I do not know Gluster, but swift that is also a > Distributed File System is not consistency and they make it pretty clear on > their page (http://docs.openstack.org/developer/swift/) > > "Swift is a highly available, distributed, eventually consistent > object/blob store...". > > I would not accept to storage my VMs images on a FS that is eventually > consistent. > > Third, network, I haven't used this kind of FS, but I can image that it > uses a lot of bandwidth to keep synchronizing, managing and securing the > data. So, managing the networking would be a pain. > > 2013/9/11 Olivier Mauras <oliv...@core-hosting.net> > >> Hi, Those thinking that it's not a good idea, do you mind explaining your point of view? GlusterFS seems like the only real alternative to a highly priced SAN for the primary storage... Thanks, Olivier On 2013-09-11 15:08, Rafael Weingartner wrote: >> >>> I >> totally agree with Tomasz, I do not think that using a distributed FS as >> >>> primary storage is a good idea, but as a secondary it sounds >> interesting. And maybe as a Swift backend. -- @shankerbalan M: +91 98860 60539 | O: +91 (80) 67935867 shanker.ba...@shapeblue.com [1] [2] | www.shapeblue.com [2][3] | Twitter:@shapeblue ShapeBlue Services India LLP, 22nd floor, Unit 2201A, World Trade Centre, Bangalore - 560 055 This email and any attachments to it may be confidential and are intended solely for the use of the indivi >> >>> sarily represent those of Shape Blue Ltd or related companies. If you are not the intended recipient of this email, you must neit >> action based upon its contents, nor copy or show it to anyone. Please contact the sender if you believe you have received this email in error. Shape Blue Ltd is a company incorporated in England & Wales. ShapeBlue Services India LLP is operated under license from Shape Blue Ltd. ShapeBlue is a registered trademark. -- Rafael Weingartner Links: ------ [1] mailto:t.a.zi...@gmail.com [3] [2] mailto:shanker.ba...@shapeblue.com [4] [3] http://www.shapeblue.com [5] >> >>> > > -- Rafael Weingartner Links: ------ [1] mailto:shanker.ba...@shapeblue.com [2] http://www.shapeblue.com [3] mailto:t.a.zi...@gmail.com [4] mailto:shanker.ba...@shapeblue.com [5] http://www.shapeblue.com