PublisherAsBlockingIterable LinkedBlockingQueue -> LinkedTransferQueue #2386
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Motivation:
LinkedBlockingQueue goes through LockSupport park and unpark methods
which can incur relatively expensive context switching if the EventLoop
thread has to unpark an application thread. This has been shown to be
a bottleneck as throughput increases.
Modifications:
- Use LinkedTransferQueue which does a `Thread.yield()` before parking
on the consumer thread. This may use more CPU on the consumer thread
but the assumption is there will be many more application threads than
EventLoop threads and we want to minimize producer costs.
Results:
LinkedTransferQueue
```
Running 30s test @ http://localhost:8080/medium, using 'ServiceTalkGrpcBlockingClientStrAgg' client
1024 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency - - - -
Req/Sec 0.01k - 0.01k -
290977 requests in 30s
Requests/sec: 9699.23
Transfer/sec: -
OK: 290977
KO: 0
```
LinkedBlockingQueue
```
Running 30s test @ http://localhost:8080/medium, using 'ServiceTalkGrpcBlockingClientStrAgg' client
1024 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency - - - -
Req/Sec 0.01k - 0.01k -
256778 requests in 30s
Requests/sec: 8559.27
Transfer/sec: -
OK: 256778
KO: 0
```
Scottmitch
force-pushed
the
blocking_iterable_xfer
branch
from
7145e7c to
1f064c1
Compare
idelpivnitskiy
added a commit
to idelpivnitskiy/servicetalk
that referenced
this pull request
Dec 20, 2022
…sferQueue (apple#2386)" This reverts commit 1c463da.
idelpivnitskiy
added a commit
that referenced
this pull request
Dec 21, 2022
…sferQueue (#2386)" (#2472) This reverts commit 1c463da (#2386). We didn't observe visible improvements for `ServiceTalkGrpcBlockingClientStrAgg`, but saw negative impact on HTTP/1.1 blocking-streaming cases: 1. On the client-side for `GET` : -10k RPS, +5-10ms p99 2. On the server-side for `POST` : -10k RPS, +5ms p99
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation:
LinkedBlockingQueue goes through LockSupport park and unpark methods which can incur relatively expensive context switching if the EventLoop thread has to unpark an application thread. This has been shown to be a bottleneck as throughput increases.
Modifications:
Thread.yield()before parking on the consumer thread. This may use more CPU on the consumer thread but the assumption is there will be many more application threads than EventLoop threads and we want to minimize producer costs.Results:
LinkedTransferQueue
LinkedBlockingQueue