msokolov commented on PR #15430:
URL: https://github.com/apache/lucene/pull/15430#issuecomment-3601795726
I've done many runs, and for the most part I'm seeing identical recall, so I
think we can safely ignore these occasionaly variants. It probably is down to
order differences from concurrent indexing, as @benwtrent says.
I can also report that the merge time difference is greatly reduced if I
enable concurrent reordering by passing the mergestate intra-merge executor in.
EG:
```
recall latency(ms) netCPU avgCpuCount visited index(s) index_docs/s
force_merge(s) index_size(MB) bp-reorder
0.937 3.022 3.002 0.993 5139 155.35 6437.20
129.43 1510.60 false
0.938 2.679 2.669 0.996 5140 160.86 6216.51
150.13 1507.24 true
```
and then by reducing the reorder MAX_ITERS from 20 to 10 we can regain some
more:
```
latency(ms) netCPU avgCpuCount visited index(s) index_docs/s
force_merge(s) index_size(MB) bp-reorder
3.095 3.085 0.997 5145 156.18 6402.87
127.41 1510.59 false
2.705 2.695 0.996 5133 160.26 6239.82
144.53 1507.26 true
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
