zhaih opened a new pull request, #15184:
URL: https://github.com/apache/lucene/pull/15184
### Description
As title, add more comments about concurrency, and also removes
`addGraphNode(int node, UpdateableRandomVectorScorer scorer)` and ensure all
`addGraphNode(int node)` call will not recreate a new scorer.
### Test
Did run benchmark on 100d vectors, didn't find any regression
#### candidate runs
single thread
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.804 0.230 0.225 0.978 500000 100 50 64
250 no 14.07 35549.24 57.87 1
216.95 190.735 190.735 HNSW
```
multi thread
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.835 0.244 0.239 0.980 500000 100 50 64
250 no 14.05 35584.66 18.36 1
220.28 190.735 190.735 HNSW
```
#### baseline runs
single thread
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.805 0.230 0.226 0.983 500000 100 50 64
250 no 13.96 35808.92 57.34 1
216.93 190.735 190.735 HNSW
```
multi thread
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized index(s) index_docs/s force_merge(s) num_segments
index_size(MB) vec_disk(MB) vec_RAM(MB) indexType
0.834 0.252 0.248 0.984 500000 100 50 64
250 no 14.32 34916.20 18.31 1
220.28 190.735 190.735 HNSW
```
--
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]
