Add GetNeighborsBenchmark #2270
Conversation
critical27
requested review from
darionyaphet,
dangleptr,
liuyu85cn,
bright-starry-sky and
panda-sheep
| GetNeighborsBenchmark.cpprelative time/iter iters/s | ||
| ============================================================================ | ||
| OneVertexOneProperty 629.54us 1.59K | ||
| OneVertexOnePropertyWithFilter 72.33% 870.39us 1.15K |
There was a problem hiding this comment.
It seems the more data filtered out, the better performance acquired.
So the bottleneck is encoder(RowWriter) ?
There was a problem hiding this comment.
Hard to say whether bottleneck is encoder or not. With bigger filter_ratio, it will need to eval the exp which is more likely to be useless. But with a small filter_ratio, we can filter out more data, so there will be less data to encode. We can only say with a big filter_ratio, it has worse performance.
The result in 2.0 is same.
There was a problem hiding this comment.
What if we comment the collector? What's the performance looks like?
There was a problem hiding this comment.
Looks like this, do not collect any value.
There was a problem hiding this comment.
Good job.
By the way, I think that even with the same data, there is no comparison between filters and no filters, right?
There was a problem hiding this comment.
Yeah, you're right.
There was a problem hiding this comment.
Looks like this, do not collect any value.
Em.... So we can ensure the cost comes from encoder.
There was a problem hiding this comment.
Let's run the benchmark for a long time, and check the perf top about it. @critical27
| namespace nebula { | ||
| namespace storage { | ||
|
|
||
| // only add two tag and two edge |
There was a problem hiding this comment.
This comment is not right,
Here have 10 tags and 10,000 edges
| GetNeighborsBenchmark.cpprelative time/iter iters/s | ||
| ============================================================================ | ||
| OneVertexOneProperty 629.54us 1.59K | ||
| OneVertexOnePropertyWithFilter 72.33% 870.39us 1.15K |
There was a problem hiding this comment.
Good job.
By the way, I think that even with the same data, there is no comparison between filters and no filters, right?
* add getneighbors benchmark * address @CPWstatic's comments * fix filter_ratio, update bm * address @panda-sheep's comments Co-authored-by: dangleptr <37216992+dangleptr@users.noreply.github.com>
* add getneighbors benchmark * address @CPWstatic's comments * fix filter_ratio, update bm * address @panda-sheep's comments Co-authored-by: dangleptr <37216992+dangleptr@users.noreply.github.com>
Add GetNeighborsBenchmark, no concurrency, used to compare with 2.0 under almost same data.
The copy of lastRank and lastDstId has trivial impact in 1.0, so leave it alone. @dangleptr.