jeantil commented on PR #2700:
URL: https://github.com/apache/james-project/pull/2700#issuecomment-2794952792
I added a microbenchmark harness for `CRLFInputStream` that I committed
along with the rest but disabled to prevent the bench from further slowing down
the CI
I also aligned the FIFO queue buffer size with the inpustream default buffer
size used in transferTo to minimize unnecessary allocations in the common case.
Here are the bench results I get on my computer
```
Benchmark
Mode Cnt Score Error Units
CRLFInputStreamBenchmark.measure_force_crlf_inputstream_wellformed
avgt 5 410.651 ± 27.984 us/op
CRLFInputStreamBenchmark.measure_force_crlf_inputstream_wellformed:gc.alloc.rate
avgt 5 3342.409 ± 230.725 MB/sec
CRLFInputStreamBenchmark.measure_force_crlf_inputstream_wellformed:gc.alloc.rate.norm
avgt 5 1439000.726 ± 0.049 B/op
CRLFInputStreamBenchmark.measure_force_crlf_inputstream_wellformed:gc.count
avgt 5 273.000 counts
CRLFInputStreamBenchmark.measure_force_crlf_inputstream_wellformed:gc.time
avgt 5 185.000 ms
CRLFInputStreamBenchmark.measure_force_crlf_outputstream_wellformed
avgt 5 66.754 ± 2.271 us/op
CRLFInputStreamBenchmark.measure_force_crlf_outputstream_wellformed:gc.alloc.rate
avgt 5 2670.461 ± 90.060 MB/sec
CRLFInputStreamBenchmark.measure_force_crlf_outputstream_wellformed:gc.alloc.rate.norm
avgt 5 186928.118 ± 0.003 B/op
CRLFInputStreamBenchmark.measure_force_crlf_outputstream_wellformed:gc.count
avgt 5 218.000 counts
CRLFInputStreamBenchmark.measure_force_crlf_outputstream_wellformed:gc.time
avgt 5 160.000 ms
CRLFInputStreamBenchmark.measure_wellformed
avgt 5 6.159 ± 0.042 us/op
CRLFInputStreamBenchmark.measure_wellformed:gc.alloc.rate
avgt 5 17021.441 ± 115.950 MB/sec
CRLFInputStreamBenchmark.measure_wellformed:gc.alloc.rate.norm
avgt 5 109944.011 ± 0.001 B/op
CRLFInputStreamBenchmark.measure_wellformed:gc.count
avgt 5 1396.000 counts
CRLFInputStreamBenchmark.measure_wellformed:gc.time
avgt 5 967.000 ms
```
If I understand this correctly, the CRLF inputstream implementation is an
order of magnitude slower than the current CRLFoutputstream which itself is an
order of magnitude slower than no CRLF handling.
I am really bad at manually managing buffers I always end up mixing offsets
so I'm not confident i can achieve a faster implementation but I think it
should be possible to do it.
--
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]
