[
https://issues.apache.org/jira/browse/HBASE-10156?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HBASE-10156:
--------------------------
Attachment: 10156v2.txt
Raw patch. Much in need of cleanup. I owe a writeup. This patch would seem
to have better perf -- test completes about 30% quicker -- but need to do more
test runs to ensure this finding holds up. I just tested 100 threads at mo.
This patch seems we have more throughput and slightly better latency. Will put
up more comprehensize numbers and a cleaned up patch in a while.
{code}
WITH PATCH
..
2014-01-02 09:30:10,213 INFO [main] wal.HLogPerformanceEvaluation: Summary:
threads=100, iterations=1000000, syncInterval=0 took 2137.059s 46793.277ops/s
...
Performance counter stats for './bin/hbase --config /home/stack/conf_hbase
org.apache.hadoop.hbase.regionserver.wal.HLogPerformanceEvaluation -threads
100':
9469922.337766 task-clock # 4.425 CPUs utilized
150,849,553 context-switches # 0.016 M/sec
50,123,607 CPU-migrations # 0.005 M/sec
73,772 page-faults # 0.008 K/sec
14,710,061,505,302 cycles # 1.553 GHz
[83.32%]
10,451,901,705,698 stalled-cycles-frontend # 71.05% frontend cycles idle
[83.32%]
5,540,518,559,443 stalled-cycles-backend # 37.66% backend cycles idle
[66.71%]
10,468,567,537,009 instructions # 0.71 insns per cycle
# 1.00 stalled cycles per insn
[83.38%]
1,729,960,111,202 branches # 182.679 M/sec
[83.32%]
21,231,188,285 branch-misses # 1.23% of all branches
[83.33%]
2140.159656084 seconds time elapsed
WITHOUT PATCH
...
2014-01-02 10:38:51,479 INFO [main] wal.HLogPerformanceEvaluation: Summary:
threads=100, iterations=1000000, syncInterval=0 took 3087.523s 32388.424ops/s
...
Performance counter stats for './bin/hbase --config /home/stack/conf_hbase
org.apache.hadoop.hbase.regionserver.wal.HLogPerformanceEvaluation -threads
100':
7708297.226562 task-clock # 2.494 CPUs utilized
300,427,649 context-switches # 0.039 M/sec
18,194,920 CPU-migrations # 0.002 M/sec
63,271 page-faults # 0.008 K/sec
11,267,121,080,680 cycles # 1.462 GHz
[83.30%]
8,257,311,140,280 stalled-cycles-frontend # 73.29% frontend cycles idle
[83.33%]
4,895,094,825,672 stalled-cycles-backend # 43.45% backend cycles idle
[66.71%]
6,186,011,736,133 instructions # 0.55 insns per cycle
# 1.33 stalled cycles per insn
[83.36%]
1,034,182,786,238 branches # 134.165 M/sec
[83.34%]
32,306,693,838 branch-misses # 3.12% of all branches
[83.33%]
3090.537038965 seconds time elapsed
{code}
> Fix up the HBASE-8755 slowdown when low contention
> --------------------------------------------------
>
> Key: HBASE-10156
> URL: https://issues.apache.org/jira/browse/HBASE-10156
> Project: HBase
> Issue Type: Sub-task
> Components: wal
> Reporter: stack
> Assignee: stack
> Attachments: 10156.txt, 10156v2.txt, Disrupting.java
>
>
> HBASE-8755 slows our writes when only a few clients. Fix.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
