Benchmark BlockingConcurrentQueue against dlib::pipe

[pfeatherstone]

Would it be possible to update the benchmarks to include dlib::pipe from http://dlib.net/other.html#pipe ?

[cameron314]

I'm open to pull requests :-)

[pfeatherstone]

How come the plots here do not show moodycamel::BlockingConcurrentQueue graphs?

[pfeatherstone]

@cameron314 I've created a PR. Hopefully the benchmarks can be updated by running on AWS. Hopefully the graphs can be updated too, and include moodycamel::BlockingConcurrentQueue as well as dlib::pipe. Usually a user wants to use moodycamel::BlockingConcurrentQueue anyway to have some upper bound on memory consumption.

[cameron314]

Thanks for the pull request! I'll take a look.

The plots were made when the queue was initially released. At the time, BlockingConcurrentQueue didn't exist yet. And the AWS results were contributed by a friendly Redditor, so unfortunately I'm unable to reproduce them.

Note also that the BlockingConcurrentQueue is no better than the ConcurrentQueue in terms of memory consumption, and that a circular-buffer based queue (similar to dlib::pipe, if I understand correctly) may better suit your needs in such a case.

[topin89]

similar to dlib::pipe

Yeah, simple fixed-size circular buffer, but it uses std::swap internally, no constructors or destructors. It requires some modification in the benchmark code to work. See #229