Hello, On Wed, 15 Jun 2016 00:22:51 +0000 pixelfairy wrote: > We have a small cluster, 3mons, each which also have 6 4tb osds, and a > 20gig link to the cluster (2x10gig lacp to a stacked pair of switches). > well have at least replica pool (size=3) and one erasure coded pool. I'm neither particular knowledgeable nor a fan of EC pools, but keep in mind that the coding is dictated by the number of OSD nodes, so 3 doesn't give a lot of options, IIRC. In fact, it will be the same as a RAID5 and only sustain the loss of one OSD/disk, something nobody in their right mind does these days. > current plan is to have journals coexist with osds as that seems to the > be safest and most economical. > You will be thoroughly disappointed by the performance if you do this, unless your use case is something like a backup server with very few random I/Os. Any performance optimizations will suggest looking at journal SSDs first. > what levels of striping would you recommend for this size cluster? any > other optimization conciderations? looking for a starting point to work > from. > Striping is one of the last things to ponder. Not only does it depend a LOT on your use case, it's also not possible to change later on, so getting it right for the initial size and future growth is an interesting challenge. > also, any recommendations for testing / benchmarking these > configurations? > > so far, looking at > https://www.sebastien-han.fr/blog/2012/08/26/ceph-benchmarks/ > bsd rebuilding itself, and maybe phoronix. > Those benchmarks are very much out-dated, both in terms of Ceph versions and capabilities as well as the tools used (fio is the most common benchmark tool for some time now). Once bluestore comes along (in a year or so), there will be another performance and HW design shift. Christian -- Christian Balzer Network/Systems Engineer chibi@xxxxxxx Global OnLine Japan/Rakuten Communications http://www.gol.com/ _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |