Hi Andrey, On Fri, 16 Mar 2012, Andrey Stepachev wrote: > Ceph has a several problems right now: > a) due of strict posix sematics ceph can't handle high throughput. > it very eays to hit max iops (even if we on tmpfs with journal) If I had to guess I wouldn't suspect the POSIX/metadata stuff initially, as that generally doesn't get in the way of data iops (although it's possible). I take it TestDSFIO is a standard hadoop benchmark? I continue to be annoyed and frustrated that we don't have a java/hadoop person in-house to work on this stuff (hint hint, http://ceph.newdream.net/jobs). I hope we can carve some resources out to work on this stuff soon. > b) broken timestamps. in my case it affected HBase HLog functionality, > so it is possible to not recover after fail. I take it this the same problem Noah was seeing? (http://tracker.newdream.net/issues/1666) Thanks! sage > > For 6 nodes cluster (I use xfs, because btrfs degrades on high volumes) > > Ceph 0.43 (from my github) > ------------------------------------------------------------------ > Date & time: Thu Mar 01 18:38:16 MSK 2012 > Number of files: 36 > Total MBytes processed: 360000 > Throughput mb/sec: 2.8411840839866644 > Average IO rate mb/sec: 2.8668723106384277 > IO rate std deviation: 0.29681573660544924 > Test exec time sec: 3991.457 > > > Original hadoop > ------------------------------------------------------------------ > ----- TestDFSIO ----- : write > Date & time: Thu Mar 01 14:08:53 MSK 2012 > Number of files: 36 > Total MBytes processed: 360000 > Throughput mb/sec: 4.59152733368745 > Average IO rate mb/sec: 4.596996784210205 > IO rate std deviation: 0.16118933245583172 > Test exec time sec: 2313.589 > > 2012/3/15 Matt Weil <mweil@xxxxxxxxxxxxxxxx>: > >> http://ceph.newdream.net/wiki/Using_Hadoop_with_Ceph > > > > > > What are the plans going forward. > > > > Are there any benchmarks out there between HDFS and ceph? > > -- > > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > > the body of a message to majordomo@xxxxxxxxxxxxxxx > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > > > -- > Andrey. > -- > To unsubscribe from this list: send the line "unsubscribe ceph-devel" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > >
