[ https://issues.apache.org/jira/browse/CASSANDRA-2975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13153112#comment-13153112 ]
David Allsopp commented on CASSANDRA-2975: ------------------------------------------ The benchmark does several rounds of warmup for each iteration (i.e. for each buffer size from 1 to 32 bytes). It reduces the number of iterations as the input buffer size grows, so that each run processes a similar number of bytes - though this is probably irrelevant since the performance seems fairly constant with respect to buffer size. {noformat} Running test for buffer lengths from 1 to 32 *| Ratio: 0.96 for keylength 1 iterations=100000000 *| Ratio: 0.95 for keylength 2 iterations=50000000 *| Ratio: 0.95 for keylength 3 iterations=33333333 *| Ratio: 0.96 for keylength 4 iterations=25000000 *| Ratio: 0.94 for keylength 5 iterations=20000000 *| Ratio: 0.94 for keylength 6 iterations=16666666 *| Ratio: 0.96 for keylength 7 iterations=14285714 *| Ratio: 0.93 for keylength 8 iterations=12500000 * | Ratio: 0.89 for keylength 9 iterations=11111111 *| Ratio: 0.93 for keylength 10 iterations=10000000 *| Ratio: 0.95 for keylength 11 iterations=9090909 *| Ratio: 0.95 for keylength 12 iterations=8333333 *| Ratio: 0.93 for keylength 13 iterations=7692307 *| Ratio: 0.90 for keylength 14 iterations=7142857 *| Ratio: 0.95 for keylength 15 iterations=6666666 * | Ratio: 0.86 for keylength 16 iterations=6250000 * | Ratio: 0.87 for keylength 17 iterations=5882352 *| Ratio: 0.91 for keylength 18 iterations=5555555 * | Ratio: 0.83 for keylength 19 iterations=5263157 * | Ratio: 0.83 for keylength 20 iterations=5000000 * | Ratio: 0.80 for keylength 21 iterations=4761904 * | Ratio: 0.88 for keylength 22 iterations=4545454 *| Ratio: 0.91 for keylength 23 iterations=4347826 *| Ratio: 0.91 for keylength 24 iterations=4166666 * | Ratio: 0.88 for keylength 25 iterations=4000000 *| Ratio: 0.92 for keylength 26 iterations=3846153 * | Ratio: 0.85 for keylength 27 iterations=3703703 * | Ratio: 0.88 for keylength 28 iterations=3571428 * | Ratio: 0.88 for keylength 29 iterations=3448275 * | Ratio: 0.89 for keylength 30 iterations=3333333 *| Ratio: 0.92 for keylength 31 iterations=3225806 -------- Old (ms): 18938 New (ms): 17470 Overall ratio: 0.9224838948146583 {noformat} i.e. 8% improvement on average. > Upgrade MurmurHash to version 3 > ------------------------------- > > Key: CASSANDRA-2975 > URL: https://issues.apache.org/jira/browse/CASSANDRA-2975 > Project: Cassandra > Issue Type: Improvement > Components: Core > Reporter: Brian Lindauer > Assignee: Brian Lindauer > Priority: Trivial > Labels: lhf > Fix For: 1.1 > > Attachments: > 0001-Convert-BloomFilter-to-use-MurmurHash-v3-instead-of-.patch, > 0002-Backwards-compatibility-with-files-using-Murmur2-blo.patch, > Murmur3Benchmark.java > > > MurmurHash version 3 was finalized on June 3. It provides an enormous speedup > and increased robustness over version 2, which is implemented in Cassandra. > Information here: > http://code.google.com/p/smhasher/ > The reference implementation is here: > http://code.google.com/p/smhasher/source/browse/trunk/MurmurHash3.cpp?spec=svn136&r=136 > I have already done the work to port the (public domain) reference > implementation to Java in the MurmurHash class and updated the BloomFilter > class to use the new implementation: > https://github.com/lindauer/cassandra/commit/cea6068a4a3e5d7d9509335394f9ef3350d37e93 > Apart from the faster hash time, the new version only requires one call to > hash() rather than 2, since it returns 128 bits of hash instead of 64. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira