Make use of relaxed*() methods in PoolThreadCache.MemoryRegionCache when possible #13734
Conversation
| @@ -368,7 +369,7 @@ protected abstract void initBuf(PoolChunk<T> chunk, ByteBuffer nioBuffer, long h | |||
| @SuppressWarnings("unchecked") | |||
| public final boolean add(PoolChunk<T> chunk, ByteBuffer nioBuffer, long handle, int normCapacity) { | |||
| Entry<T> entry = newEntry(chunk, nioBuffer, handle, normCapacity); | |||
| boolean queued = queue.offer(entry); | |||
| boolean queued = queue.relaxedOffer(entry); | |||
There was a problem hiding this comment.
relaxed offer on mpsc doesn't make much difference https://github.com/JCTools/JCTools/blob/master/jctools-core/src/main/java/org/jctools/queues/MpscArrayQueue.java#L456
while relaxedPoll would save it to spin awaiting the elements (and breaking linearizability if the queue);
I don't suggest to go there unless there's some evidence of pathologic cases which can by fixing by this change...
There was a problem hiding this comment.
relaxed offer on mpsc doesn't make much difference
True for relaxedOffer.
About the relaxedPoll part, I may try some benchmark if possible.
There was a problem hiding this comment.
@laosijikaichele any update ?
There was a problem hiding this comment.
@normanmaurer Sorry for the delay, been busy with other things recently, I may still need some time to back on this PR...
There was a problem hiding this comment.
please ping me once you are ready
…che.MemoryRegionCache to reduce the overhead
Motivation:
This is inspired by #11960.
We should be able to make use of
MessagePassingQueue.relaxedPoll()methods inPoolThreadCache.MemoryRegionCacheto reduce the overhead.Modification:
Use
MessagePassingQueue.relaxedPoll(...).Result:
Less overhead.