lkml.org 
[lkml]   [2005]   [Apr]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Subjecta few dbench datapoints across various filesystems
    From
    Date
    I ran some quick tests with dbench to see the effects of various
    performance improvements, and found the results interesting. Although
    dbench is too write oriented, and not particularly favorable to a few
    filesystems (who are otherwise good performers), dbench can still can be
    useful.

    System was a Pentium 4 2.40 GHz, uni, IDE drive, 1GB RAM
    Kernel was mainline: 2.6.12-rc2. dbench version 3.03 from samba.org


    Unfortunately the two target partitions were not the exact same size
    (the small partition size may skew some ext3 results somewhat higher,
    but it saved time to avoid reformatting my development machine).
    partition sizes were 16.5 GB vs JFS 2 GB EXT3

    Network filesystems tests (nfs v3, cifs, smbfs) were mounted over
    localhost to minimize the effects of network adapter differences.
    The cifs code was using the current version of cifs.ko (version 1.33)
    newer than what is in mainline.

    JFS Throughput 52.6572 MB/sec 20 procs
    ext3 Throughput 179.726 MB/sec 20 procs
    ext3 (2nd run) Throughput 180.409 MB/sec 20 procs
    cifs mounted to samba/JFS 9.67401 Throughput MB/sec 20 procs
    cifs mounted to samba/ext3 Throughput 12.2919 MB/sec 20 procs
    nfs mounted to JFS Throughput 16.3588 MB/sec 20 procs
    nfs mounted to ext3 Throughput 13.5945 MB/sec 20 procs
    nfs mounted to ext3 (2nd run) Throughput 12.4307 MB/sec 20 procs

    Then using newer JFS code with the faster JFS code from current mm tree
    for the following five:
    JFS Throughput 58.3836 MB/sec 20 procs
    nfs mount to jfs Throughput 16.562 MB/sec 20 procs
    smbfs mount to Samba/jfs Throughput 16.4742 MB/sec 20 procs
    cifs larger mempool (40 buffers) (to Samba/jfs)
    Throughput 12.8196 MB/sec 20 procs
    cifs larger mempool (40 buffers), mount with directio mount option
    (to Samba/JFS)
    Throughput 14.1643 MB/sec 20 procs

    The cifs numbers are much improved, and should be pretty easy to get to
    another 20 or 30% faster, which I will look at doing once we finishing
    testing cifs version 1.33

    Interesting that JFS was slower than ext3 for the local test, but faster
    than ext3 when mounting via NFS to the same system, same filesystem.
    The JFS performance guys are apparently close to improving the local
    dbench numbers since dbench shows a performance issue with writing
    synchr. to the journal where other fs are able to do it faster (JFS does
    well apparently on most of the other benchmarks, and my informal tests
    showed that JFS was faster on various simple perf tests as well).

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-04-21 22:09    [W:4.768 / U:0.040 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site