hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Purtell <andrew.purt...@gmail.com>
Subject Re: [VOTE] The second HBase 1.4.9 release candidate (RC1) is available
Date Sat, 08 Dec 2018 06:00:59 GMT
YCSB workloads are not that interesting. I think it’s still useful as a “standard” tool
though that indicates we aren’t doing worse. Also, there may be more difference under overload
conditions. Next time. 

Using Phoenix and some of our internal test tools we can generate workloads that turn up interesting
differences at the day job, but mostly between 0.98 and any 1.x, and the changes were small
and additive over time. Some of our folks contributed perf fixes that return some of the difference,
10%, 20% here and there. There are opportunities for more. 

For 1.5 branch-1 has some interesting changes, like pressure aware compaction rate limiting
on by default, which may change things (via influence on GC). TinyLRU if it goes in might
also be interesting. 


> On Dec 7, 2018, at 8:02 PM, Stack <stack@duboce.net> wrote:
> 
> Thanks for the test runs.
> 
> The diffs are miniscule. After so many releases, would have expected a
> tendency up or down but not constant (smile).
> 
> S
> 
>> On Fri, Dec 7, 2018 at 5:34 PM Andrew Purtell <apurtell@apache.org> wrote:
>> 
>> Today I did a comparison between 1.2.6.1 and 1.4.9RC1 with YCSB. The
>> results are close. Overall runtimes are almost the same. In the average and
>> high percentile measures there is a general upward trend but nothing that
>> looks like a significant regression. Still for 1.5.0 I think we should see
>> if it's possible to lower measures made by YCSB closer to those observed
>> with 1.2.
>> 
>> Linux version 4.14.55-62.37.amzn1.x86_64
>> c3.8xlarge x 5
>> OpenJDK Runtime Environment (build 1.8.0_181-shenandoah-b13)
>> -Xms20g -Xmx20g -XX:+UseG1GC -XX:+AlwaysPreTouch -XX:+UseNUMA
>> -XX:-UseBiasedLocking -XX:+ParallelRefProcEnabled
>> Hadoop 2.8.5
>> Init: Load 100 M rows and snapshot
>> Run: Delete table, clone and redeploy from snapshot, run 10 M operations
>> (except 1M for workload E)
>> Args: -threads 100 -target 50000 (except 10000 for workload E)
>> 
>> YCSB Workload A
>> 
>> target 50k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 200581 200605
>> [OVERALL], Throughput(ops/sec) 49855 49849
>> [READ], AverageLatency(us) 491 517
>> [READ], MinLatency(us) 233 273
>> [READ], MaxLatency(us) 140287 165503
>> [READ], 95thPercentileLatency(us) 605 647
>> [READ], 99thPercentileLatency(us), 880 1031
>> [UPDATE], AverageLatency(us) 1332 1327
>> [UPDATE], MinLatency(us) 711 692
>> [UPDATE], MaxLatency(us) 137215 160383
>> [UPDATE], 95thPercentileLatency(us) 1790 1800
>> [UPDATE], 99thPercentileLatency(us) 2333 2495
>> 
>> YCSB Workload B
>> 
>> target 50k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 200569 200566
>> [OVERALL], Throughput(ops/sec) 49858 49859
>> [READ], AverageLatency(us),  438 445
>> [READ], MinLatency(us) 198 216
>> [READ], MaxLatency(us) 143615 153343
>> [READ], 95thPercentileLatency(us) 539 552
>> [READ], 99thPercentileLatency(us) 806 957
>> [UPDATE], AverageLatency(us) 1098 1064
>> [UPDATE], MinLatency(us) 746 762
>> [UPDATE], MaxLatency(us) 141183 149503
>> [UPDATE], 95thPercentileLatency(us) 1334 1336
>> [UPDATE], 99thPercentileLatency(us) 1695 1777
>> 
>> YCSB Workload C
>> 
>> target 50k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 200543 200559
>> [OVERALL], Throughput(ops/sec) 49865 49861
>> [READ], AverageLatency(us) 348 340
>> [READ], MinLatency(us) 174 182
>> [READ], MaxLatency(us) 140287 161279
>> [READ], 95thPercentileLatency(us) 428 421
>> [READ], 99thPercentileLatency(us) 793 841
>> 
>> YCSB Workload D
>> 
>> target 50k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 200557 200577
>> [OVERALL], Throughput(ops/sec) 49861 49856
>> [READ], AverageLatency(us) 528 503
>> [READ], MinLatency(us) 187 213
>> [READ], MaxLatency(us) 134655 147199
>> [READ], 95thPercentileLatency(us) 1395 991
>> [READ], 99thPercentileLatency(us) 1880 1721
>> [INSERT], AverageLatency(us) 1272 1245
>> [INSERT], MinLatency(us) 830 827
>> [INSERT], MaxLatency(us) 124479 140671
>> [INSERT], 95thPercentileLatency(us) 1518 1505
>> [INSERT], 99thPercentileLatency(us) 2249 2553
>> 
>> YCSB Workload E
>> 
>> target 10k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 100564 100584
>> [OVERALL], Throughput(ops/sec) 9944 9942
>> [SCAN], AverageLatency(us) 4297 3700
>> [SCAN], MinLatency(us) 765 740
>> [SCAN], MaxLatency(us) 1229823 1056767
>> [SCAN], 95thPercentileLatency(us) 10503 9855
>> [SCAN], 99thPercentileLatency(us) 22655 19007
>> [INSERT], AverageLatency(us) 3178 2707
>> [INSERT], MinLatency(us) 935 885
>> [INSERT], MaxLatency(us) 1020415 148479
>> [INSERT], 95thPercentileLatency(us) 5795 4927
>> [INSERT], 99thPercentileLatency(us) 13791 9727
>> 
>> YCSB Workload F
>> 
>> target 50k/op/s 1.2.6.1 1.4.9
>> 
>> 
>> 
>> [OVERALL], RunTime(ms) 200619 200583
>> [OVERALL], Throughput(ops/sec) 49846 49855
>> [READ], AverageLatency(us) 577 610
>> [READ], MinLatency(us) 246 270
>> [READ], MaxLatency(us) 131455 127743
>> [READ], 95thPercentileLatency(us) 815 909
>> [READ], 99thPercentileLatency(us) 1525 1549
>> [READ-MODIFY-WRITE], AverageLatency(us) 2006 2050
>> [READ-MODIFY-WRITE], MinLatency(us) 1105 1116
>> [READ-MODIFY-WRITE], MaxLatency(us) 164095 178303
>> [READ-MODIFY-WRITE], 95thPercentileLatency(us) 2723 2855
>> [READ-MODIFY-WRITE], 99thPercentileLatency(us) 4423 4415
>> [UPDATE], AverageLatency(us) 1427 1438
>> [UPDATE], MinLatency(us) 703 727
>> [UPDATE], MaxLatency(us) 80767 128703
>> [UPDATE], 95thPercentileLatency(us) 1993 2031
>> [UPDATE], 99thPercentileLatency(us) 2727 2751
>> 
>> 
>>> On Thu, Dec 6, 2018 at 4:45 PM Andrew Purtell <apurtell@apache.org> wrote:
>>> 
>>> The second HBase 1.4.9 release candidate (RC1) is available for download
>>> at https://dist.apache.org/repos/dist/dev/hbase/hbase-1.4.9RC1/ and
>> Maven
>>> artifacts are available in the temporary repository
>>> https://repository.apache.org/content/repositories/orgapachehbase-1240/
>>> 
>>> There was no vote on 1.4.9RC0.
>>> 
>>> The git tag corresponding to the candidate is '1.4.9RC1' (d625b212e4).
>>> 
>>> A detailed source and binary compatibility report for this release is
>>> available for your review at
>>> 
>> https://dist.apache.org/repos/dist/dev/hbase/hbase-1.4.9RC1/compat-check-report.html
>>> . There are no reported compatibility issues.
>>> 
>>> A list of the 22 issues resolved in this release can be found at
>>> https://s.apache.org/fNPx .
>>> 
>>> Please try out the candidate and vote +1/0/-1.
>>> 
>>> The vote will be open for at least 72 hours. Unless objection I will try
>>> to close it Friday December 14, 2018 if we have sufficient votes.
>>> 
>>> Prior to making this announcement I made the following preflight checks:
>>> 
>>>    RAT check passes (7u80)
>>>    Unit test suite passes 5/5 (7u80, 8u172)
>>>    LTT load 100M rows with 100% verification and 20% updates (8u181)
>>>    ITBLL 500M rows with slowDeterministic monkey (8u181)
>>> 
>> 
>> --
>> Best regards,
>> Andrew
>> 
>> Words like orphans lost among the crosstalk, meaning torn from truth's
>> decrepit hands
>>   - A23, Crosstalk
>> 

Mime
View raw message