Which Java blocking queue is most efficient for single-producer single-consumer scenarios

Question

I\'m working on a standard Java system with critical timing requirements for my producers (1/100s of ms matters).

I have a producer placing stuff in a blocking queu

Answer 1

You can write a latency sensitive system in Java but you have to be careful of memory allocations, information passing between threads and locking. You should write some simple tests to see how much time these things cost on your server. (They vary bases on the OS and the memory architecture)

If you do this you can get stable timings for operations up to 99.99% of the time. This is not proper real-time, but it may be close enough. On a trading system, using Java instead of C costs might cost £100/d, however the cost of developing in C/C++ rather than Java is likely to be much higher than this. e.g. in terms of the flexibility it gives us and the number of bugs it saves.

You can get fairly close the same amount of jitter you would see in a C program doing the same thing. Some call this "C-like" Java.

Unfortunately it takes about 8 micro-second to pass an object between threads in Java via a ArrayBlockingQueue on the servers I work on and I suggest you test this on your servers. A simple test is to pass the System.nanoTime() between threads and see how long it takes.

ArrayBlockingQueue has a "feature" where it creates an object on each element added (though there is not much of a good reason to do so) So if you find a standard implementation which doesn't do this, let me know.