minor optimize IndexMerger's MMappedIndexRowIterable#2084
minor optimize IndexMerger's MMappedIndexRowIterable#2084fjy merged 1 commit intoapache:masterfrom binlijin:master
Conversation
|
The performance numbers is: 2015-12-11T15:01:42,938 INFO [main] io.druid.segment.IndexMerger - outDir[/var/folders/z7/g5zy3kfj7t54y1f074hsqw7h0000gn/T/base1258220153873171419flush/merged/v8-tmp] walked 500,000/500,000 rows in 15,017 millis. After: 2015-12-11T14:50:33,574 INFO [main] io.druid.segment.IndexMerger - outDir[/var/folders/z7/g5zy3kfj7t54y1f074hsqw7h0000gn/T/base2937734964603563157flush/merged/v8-tmp] walked 500,000/500,000 rows in 14,069 millis. |
|
+1 |
|
@binlijin what did you run the benchmarks on, ec2 or your laptop? (or some other dedicated machine) |
|
@drcrallen China doesn't have EC2, it has AliCloud edit: my bad, it does have EC2 :P |
|
allright I'll be more explicit: Given the short timerange of the tests presented, and the small improvement (small improvements are good if they are real!) are variations such as frequency governor fluctuations taken into account? |
|
@drcrallen I run it on my laptop. |
|
@binlijin Could you include the test you've used in druid? We might need standard test for the performance. |
|
@navis, i test it with our one million real data. |
|
@binlijin can you try running on a dedicated machine, the difference you are seeing could just be due to clock rate / temperature fluctuations on your laptop is what @drcrallen is saying |
|
I test it on our test machine, the performance number is: 2015-12-15 06:35:17,978 INFO [main] segment.IndexMerger (Logger.java:info(70)) - outDir[/tmp/base8564891586222722664flush/merged/v8-tmp] walked 500,000/500,000 rows in 22,278 millis. 2015-12-15 06:37:38,199 INFO [main] segment.IndexMerger (Logger.java:info(70)) - outDir[/tmp/base7286842343919247533flush/merged/v8-tmp] walked 500,000/500,000 rows in 21,865 millis. After: 2015-12-15 06:10:55,844 INFO [main] segment.IndexMerger (Logger.java:info(70)) - outDir[/tmp/base3754891995514154122flush/merged/v8-tmp] walked 500,000/500,000 rows in 19,548 millis. 2015-12-15 06:13:21,332 INFO [main] segment.IndexMerger (Logger.java:info(70)) - outDir[/tmp/base2471600163749106438flush/merged/v8-tmp] walked 500,000/500,000 rows in 21,338 millis. It is slow because my laptop is SSD and this machine is not. processor : 23 |
|
👍 |
minor optimize IndexMerger's MMappedIndexRowIterable
Copy from IndexMaker.