Note that in order to compare apples to apples the instances should be configured the same. 0.98 increases the block cache to 40% by default, enables scanner caching by default, and disabled Nagle's (I think). The same config changes should be done to 0.94 (unless we truly want to compare the out of the box configurations).
-- Lars ________________________________ From: Jean-Marc Spaggiari <[email protected]> To: dev <[email protected]>; lars hofhansl <[email protected]> Sent: Thursday, July 3, 2014 6:10 AM Subject: Re: [VOTE] The 1st hbase 0.94.21 release candidate is available for download Here are the perfs tests results. Someone on the user list asked few days ago about perfs between 0.94 and 0.98. So I have put also 0.98 on my results here. Basically, as I said yesterday, no negativ perf impact for 0.94.21. Each test run 10 times (except the last 4) and I remove the 2 extremes. JM 0.94.20 0.94.21 0.98.2 FilteredScanTest 0,23 0,22 0,25 RandomReadTest 808 815 1 107 RandomSeekScanTest 172 169 201 RandomScanWithRange10Test 286 279 277 RandomScanWithRange100Test 147 145 161 RandomScanWithRange1000Test 38,22 37,27 50 SequentialReadTest 1 217 1 199 1 392 SequentialWriteTest 13 732 13 391 19 455 RandomWriteTest 14 091 13 642 19 708 GaussianRandomReadBenchmark 9 404 9 435 10 777 SequentialReadBenchmark 2 938 638 3 074 815 3 335 239 SequentialWriteBenchmark 912 469 913 573 833 654 UniformRandomReadBenchmark 10 299 10 358 11 762 UniformRandomSmallScan 233 583 234 083 277 850 LoadTestTool real 19m30.113s user 37m12.300s sys 11m19.224s real 19m26.099s user 36m1.740s sys 11m41.704s real 14m40.709s user 28m56.892s sys 12m6.364s IntegrationTestLoadAnVerify real 4m11.269s user 1m32.904s sys 0m7.176s real 4m7.288s user 1m31.704s sys 0m6.780s real 2m38.295s user 1m24.500s sys 0m6.036s HLogPerformanceEvaluation 10431,988 10629,025 n/a IntegrationTestBigLinkedList real 6m0.125s user 3m0.004s sys 0m11.312s real 6m17.616s user 2m58.684s sys 0m10.492S real 9m9.284s user 3m31.576s sys 0m19.828s 2014-07-02 21:22 GMT-04:00 Jean-Marc Spaggiari <[email protected]>: So far performances are similar to 0.94.21. So green from my side too. I will post detailed results later tonight or tomorrow morning. > >JM > > > > >2014-07-02 18:29 GMT-04:00 lars hofhansl <[email protected]>: > > >Thanks Ted & Andy. >>With my +1 we have the required 3 binding votes, will release 0.94.21 >>tomorrow. >> >>-- Lars >> >> >> >>________________________________ >> From: Ted Yu <[email protected]> >>To: "[email protected]" <[email protected]> >>Sent: Wednesday, July 2, 2014 1:25 PM >> >>Subject: Re: [VOTE] The 1st hbase 0.94.21 release candidate is available for >>download >> >> >> >>+1 >> >>Checked md5 sum >>Checked signature >>Apache RAT check passes. >>Unit test suite passed (using jdk 1.7.0_25) >> >>Cheers >> >> >> >> >> >>On Wed, Jul 2, 2014 at 1:19 PM, Andrew Purtell <[email protected]> wrote: >> >>> +1 >>> >>> MD5 sum is good >>> Signature is good >>> Unpacked tarball, structure looks good. >>> Ran Apache RAT, check passed**. >>> Local compile and build was successful (with 7u60) >>> Unit test suite passes 10 times out of 10 >>> Ran LoadTestTool, no errors, logs look good. >>> >>> ** - You have to remove docs/ and generated web.xml files under >>> hbase-webapps/ first. Going forward I think we should release a source only >>> tarball and a second -bin "binary convenience" artifact. >>> >>> >>> >>> On Fri, Jun 27, 2014 at 2:20 PM, lars hofhansl <[email protected]> wrote: >>> >>> > The 1st 0.94.21 RC is available for download at >>> > http://people.apache.org/~larsh/hbase-0.94.21-rc0/ >>> > Signed with my code signing key: C7CFE328 >>> > >>> > 0.94 development is continuing to wind down, HBase 0.94.21 is a small bug >>> > fix release with 9 fixes: >>> > [HBASE-10692] - The Multi TableMap job don't support the security >>> > HBase cluster >>> > [HBASE-11052] - Sending random data crashes thrift service >>> > [HBASE-11096] - stop method of Master and RegionServer coprocessor is >>> > not invoked >>> > [HBASE-11234] - FastDiffDeltaEncoder#getFirstKeyInBlock returns wrong >>> > result >>> > [HBASE-11341] - ZKProcedureCoordinatorRpcs should respond only to >>> > members >>> > [HBASE-11414] - Backport to 0.94: HBASE-7711 rowlock release problem >>> > with thread interruptions in batchMutate >>> > [HBASE-8495] - Change ownership of the directory to bulk load >>> > [HBASE-10871] - Indefinite OPEN/CLOSE wait on busy RegionServers >>> > [HBASE-10935] - support snapshot policy where flush memstore can be >>> > skipped to prevent production cluster freeze >>> > >>> > The list of changes is also available here: >>> > >>> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12326794 >>> > >>> > Here's the test run for this RC: >>> > https://builds.apache.org/job/HBase-0.94.21/30/ >>> > >>> > Please try out the RC, check out the doc, take it for a spin, etc, and >>> > vote +1/-1 by EOD July 3rd on whether we should release this as 0.94.21. >>> > >>> > Thanks. >>> > >>> > -- Lars >>> > >>> >>> >>> >>> -- >>> Best regards, >>> >>> - Andy >>> >>> Problems worthy of attack prove their worth by hitting back. - Piet Hein >>> (via Tom White) >>> >
