Hi, thanks for your response. Yes, I ran my benchmark tests with a striping configuration of 1 MB stripe unit, 10 stripe count and 10 MB object size. Therefore when reading with a blocksize of 10 MB, 10 stripe units in 10 different objects could be read in parallel [1]. With an (artificially large) blocksize of 1 GB the CephFS kernel client could read from 1024 MB / 10 MB = 103 objects (ideally stored on different OSDs) in parallel.
What is the practical implication of storing multiple stripe units into a single object vs. the default configuration with one stripe unit per object? In both cases the file is splitted into multiple chunks and could be read and written in parallel if I'm not mistaken. With multiple stripe units per object fewer objects are required - is this the reason (object creation+management is probably costly)? [1] http://docs.ceph.com/docs/master/dev/file-striping/ Andreas 2016-12-08 4:02 GMT+01:00 Goncalo Borges <goncalo.bor...@sydney.edu.au>: > Hi... > > Are you actually playing with file layout? > > http://docs.ceph.com/docs/jewel/cephfs/file-layouts/ > > By increasing the stripe count and tuning the stripe unit to your > application block size, you may have an increase of performance. > > Cheers > > Goncalo > > > > > On 12/08/2016 12:45 PM, Andreas Gerstmayr wrote: >> >> Hi, >> >> does the CephFS kernel module (as of kernel version 4.8.8) support >> parallel reads of file stripes? >> When an application requests a 500MB block from a file (which is >> splitted into multiple objects and stripes on different OSDs) at once, >> does the CephFS kernel client request these blocks in parallel? >> >> My benchmarks suggest it does not (there is no significant difference >> in throughput whether I'm reading a file in chunks of 64 KB, 500 MB or >> 1 GB blocks). >> In the architecture docs [1] under the protocol section data striping >> and the resulting performance gains are explained - but I'm not sure >> if this means that this optimization is already implemented in the >> current CephFS kernel module or not. >> >> >> [1] http://docs.ceph.com/docs/master/architecture/#data-striping >> >> Thanks, >> Andreas >> _______________________________________________ >> ceph-users mailing list >> ceph-users@lists.ceph.com >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com >> > > -- > Goncalo Borges > Research Computing > ARC Centre of Excellence for Particle Physics at the Terascale > School of Physics A28 | University of Sydney, NSW 2006 > T: +61 2 93511937 > _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
