Concurrency helpers for the JVM
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
src Update ParallelGZIPOutputStream.java Aug 6, 2017
README.md across the board updates Aug 5, 2017
build.gradle add test case for CustomBlockingMpmcQueue Aug 6, 2017

README.md

Concurrent utilities

PRs welcome, these are collections of utilities I have written or wished was available when attempting to build a faster and more concurrent system.

Please read the source before applying them to your system as they will kick off new threads with waiting loops etc. You might need to tune them for your solution.

AsyncOutputStreamQueue

Allows offload of the work to the next stage of an OutputStream on another thread via a queue. This means your producer can keep generating data to put on the output stream while the next stage is performing the actual output work.

Ideally this is what the PipedInputStream / PipedOutputStream should do. Unfortunately the PipedOutputStream has 1sec sleeps and requires wait/notify or other solutions to get a reasonable performance.

BufferedOutputStream

A BufferedOutputStream drop-in replacement that uses atomic CAS instead of synchronized blocks to get data through. If you have many threads attempting to write to a fast OutputStream, then this stream should operate a lot faster.

As an example, both other OutputStreams here queue the work on other threads, and tend to return very quickly, this means the overhead of a synchronized call can be quite large with lots of writers

BufferedOutputStream2

Another attempt, using an MPSC queue and a thread. In my testing this was a little slower than the CAS solution, so you should probably prefer that. Left here for testing and/or further improvements.

ParallelGZIPOutputStream

A parallel implementation of gzip for the JVM. A drop-in replacement for the existing GZIPOutputStream. Achieve a significant improvement in your gzip output performance.

Inspired by Shevek's work, only without a buffer and with a little less locking https://github.com/shevek/parallelgzip

CustomBlockingMpmcQueue

When Java's out of the box ArrayBlockingQueue and similar become a bit slow, you can switch to NitsanW's JCTools queues. Unfortunately they don't out of the box implement BlockingQueue, which you need for an Executor. Use this adapter in your executor and you can take advantage of the JCTools queue performance in your executor.