Sorry, just saw your vote on RC now.
On Thu, Jul 17, 2014 at 5:16 PM, Enis Söztutar <e...@apache.org> wrote:
> This looks indeed concerning. It seems that Workload E is 95% scan, and
> the other workloads have no scan, so it seems that we have some regression
> in scans.
>
> Should this sink the RC, what do you think?
>
> Enis
>
>
> On Thu, Jul 17, 2014 at 5:08 PM, Andrew Purtell <apurt...@apache.org>
> wrote:
>
>> Comparing the relative performance of 0.98.4 RC0 and 0.98.0 on Hadoop
>> 2.2.0
>> using YCSB.
>>
>> This will be the last report of these from me for a while, as I will be
>> losing my current access to EC2 resources tomorrow.
>>
>> 5 concurrent YCSB clients on 5 servers target 100,000 ops/second in
>> aggregate. Reported average values are averages of readings from all
>> clients over 3 runs. Min values are the minimum reported by any client on
>> any run. Max and percentile values are the maximum reported by any client
>> on any run. What is interesting is relative differences, because each EC2
>> testbed has a varying baseline. 0.98.0 and 0.98.4 tests were run on the
>> same instance set.
>>
>> These tests were run with no security coprocessors installed, using HFile
>> V2. The workload E results are a concern. *It appears we have a 23%
>> decline
>> in measured scan throughput and an 23% increase in average op time from 27
>> ms to 35 ms. *This does not correspond to any active security feature
>> (though that could worsen results potentially, untested) so is something
>> changed in core code. Other workloads are not affected so this is
>> something
>> specific to scanning. Perhaps delete tracking.
>>
>>
>> *Hardware and Versions*
>>
>> Hadoop 2.2.0
>>
>> HBase 0.98.0-hadoop2 + HBASE-11277
>>
>> HBase 0.98.4-hadoop2 RC0
>>
>> YCSB 1.0.4
>>
>>
>> 11x EC2 c3.8xlarge: 1 master, 5 slaves, 5 test clients
>>
>> 32 cores
>>
>> 60 GB RAM
>>
>> 2 x 320 GB directly attached SSD
>>
>> NameNode: 4 GB heap
>>
>> DataNode: 1 GB heap
>>
>> Master: 1 GB heap
>>
>> RegionServer: 8 GB heap, 24 GB bucket cache offheap engine
>>
>>
>> *Methodology*
>>
>>
>> Setup:
>>
>> 0. Start cluster
>> 1. shell: create "seed", { NAME=>"u", COMPRESSION=>"snappy" }
>> 2. YCSB: Preload 100 million rows into table "seed"
>> 3. shell: flush "seed" ; compact "seed"
>> 4. Wait for compaction to complete
>> 5. shell: create_snapshot "seed", "seed_snap"
>> 6. shell: disable "seed"
>>
>>
>> For each test:
>>
>> 7. shell: clone_snapshot "seed_snap", "test"
>> 8. YCSB: On each client (5 clients), run test -p
>> operationcount=2000000 -threads 20 -target 20000
>> 9. shell: disable "test"
>> 10. shell: drop "test"
>>
>>
>>
>> *Workload A*
>> *0.98.0* *0.98.4*
>>
>>
>>
>> [OVERALL] RunTime(ms) 100743 100693 [OVERALL] Throughput(ops/sec) 99263
>> 99312 [UPDATE] Operations 4997918 4999620 [UPDATE] AverageLatency(us)
>> 633
>> 647 [UPDATE] MinLatency(us) 269 268 [UPDATE] MaxLatency(us) 1450432
>> 713191
>> [UPDATE] 95thPercentileLatency(ms) 0 0 [UPDATE]
>> 99thPercentileLatency(ms) 5
>> 4 [READ] Operations 5002242 5000540 [READ] AverageLatency(us) 151 144
>> [READ] MinLatency(us) 0 0 [READ] MaxLatency(us) 1104157 952392 [READ]
>> 95thPercentileLatency(ms) 0 0 [READ] 99thPercentileLatency(ms) 0 0
>>
>>
>>
>> *Workload B*
>>
>>
>>
>>
>>
>>
>> [OVERALL] RunTime(ms) 100465 100458 [OVERALL] Throughput(ops/sec) 99537
>> 99544 [UPDATE] Operations 9499627 9499891 [UPDATE] AverageLatency(us)
>> 556
>> 589 [UPDATE] MinLatency(us) 268 264 [UPDATE] MaxLatency(us) 709604
>> 695863
>> [UPDATE] 95thPercentileLatency(ms) 0 0 [UPDATE]
>> 99thPercentileLatency(ms) 1
>> 2 [READ] Operations 500533 500269 [READ] AverageLatency(us) 147 144
>> [READ] MinLatency(us) 0 0 [READ] MaxLatency(us) 571294 495148 [READ]
>> 95thPercentileLatency(ms) 0 0 [READ] 99thPercentileLatency(ms) 0 0
>>
>>
>>
>> *Workload C*
>>
>>
>>
>>
>>
>>
>> [OVERALL] RunTime(ms) 100091 100022 [OVERALL] Throughput(ops/sec) 99909
>> 99978 [READ] Operations 9916831 10000000 [READ] AverageLatency(us) 524
>> 526
>> [READ] MinLatency(us) 273 269 [READ] MaxLatency(us) 737108 741634 [READ]
>> 95thPercentileLatency(ms) 0 0 [READ] 99thPercentileLatency(ms) 1 2
>>
>>
>>
>> *Workload D*
>>
>>
>>
>>
>>
>>
>> [OVERALL] RunTime(ms) 114244 103308 [OVERALL] Throughput(ops/sec) 89114
>> 96809 [INSERT] Operations 9499965 9500306 [INSERT] AverageLatency(us)
>> 1145
>> 668 [INSERT] MinLatency(us) 270 271 [INSERT] MaxLatency(us) 4598999
>> 3291540 [INSERT] 95thPercentileLatency(ms) 6 1 [INSERT]
>> 99thPercentileLatency(ms) 13 3 [READ] Operations 500035 499694 [READ]
>> AverageLatency(us) 14 15 [READ] MinLatency(us) 4 4 [READ] MaxLatency(us)
>> 494730 495198 [READ] 95thPercentileLatency(ms) 0 0 [READ]
>> 99thPercentileLatency(ms) 0 0
>>
>>
>>
>> *Workload E*
>>
>>
>>
>>
>>
>>
>> [OVERALL] RunTime(ms) 1600910 2078826 [OVERALL] Throughput(ops/sec) 6308
>> 4835 [INSERT] Operations 499131 500322 [INSERT] AverageLatency(us) 14 17
>> [INSERT] MinLatency(us) 5 5 [INSERT] MaxLatency(us) 506079 564468
>> [INSERT]
>> 95thPercentileLatency(ms) 0 0 [INSERT] 99thPercentileLatency(ms) 0 0
>> [SCAN] Operations 9500869 9499678 [SCAN] AverageLatency(us)
>>
>>
>> 26636 34620 [SCAN] MinLatency(us) 746 755 [SCAN] MaxLatency(us) 8067864
>> 4615914 [SCAN] 95thPercentileLatency(ms) 117 136 [SCAN]
>> 99thPercentileLatency(ms) 169 187
>>
>>
>>
>> *Workload F*
>>
>>
>>
>>
>>
>>
>> [OVERALL] RunTime(ms) 100876 100820 [OVERALL] Throughput(ops/sec) 99133
>> 99187 [UPDATE] Operations 10000000 10000000 [UPDATE] AverageLatency(us)
>> 737 746 [UPDATE] MinLatency(us) 273 272 [UPDATE] MaxLatency(us) 759812
>> 747124 [UPDATE] 95thPercentileLatency(ms) 1 1 [UPDATE]
>> 99thPercentileLatency(ms) 5 6 [READ-MODIFY-WRITE] Operations 5000370
>> 5000082 [READ-MODIFY-WRITE] AverageLatency(us) 742 750
>> [READ-MODIFY-WRITE]
>> MinLatency(us) 280 279 [READ-MODIFY-WRITE] MaxLatency(us) 756180 747197
>> [READ-MODIFY-WRITE] 95thPercentileLatency(ms) 1 1 [READ-MODIFY-WRITE]
>> 99thPercentileLatency(ms) 5 6 [READ] Operations 5000530 5000242 [READ]
>> AverageLatency(us) 22 17 [READ] MinLatency(us) 0 0 [READ] MaxLatency(us)
>> 1551953 1097394 [READ] 95thPercentileLatency(ms) 0 0 [READ]
>> 99thPercentileLatency(ms) 0 0
>>
>>
>>
>> --
>> Best regards,
>>
>> - Andy
>>
>> Problems worthy of attack prove their worth by hitting back. - Piet Hein
>> (via Tom White)
>>
>
>
