Monday, March 08, 2010

Reading Notes #3: Ring Benchmark

I have been working my way through Joe Armstrong's Programming in Erlang. One of the exercises in chapter 8 is to write a "ring benchmark" in Erlang and any other language, compare the results and publish them.

Write a ring benchmark. Create N processes in a ring. Send a message round the ring M times so that a total of N * M messages get sent. Time how long this takes for different values of N and M. Write a similar program in some other programming language you are familiar with. Compare the results. Write a blog, and publish the results on the Internet!

Here are my benchmarks written in Erlang and Python respectively. I have presumed that the author meant sending messages synchronously; i.e. a message has to go around the whole ring before the next can be sent.

The Erlang version was easy to write. After all the language is designed to use a processes and messages paradigm of programming. Here is the source code:

I wrote the "other programming language" version in Python. The benchmark code makes use of Python's multiprocessing module (available in version 2.6 and above). Since I couldn't find any Erlang-style way to directly send a message to a process I chose to use a set of Queue objects to accomplish message passing. Each process has access to two queues - an "in" queue to receive messages and an "out" queue to relay them to the next process in the ring. Each process' "out" queue serves as the "in" for the next. After constructing N - 1 processes, a "plug" process is created to complete the process ring.

I am not entirely convinced that Python's processes are the equivalent of Erlang ones. Certainly threads are not the answer given their use of shared memory. Perhaps the difficulty in creating lightweight, shared-nothing processes in other languages is what the author wanted to illustrate using the example.

The results are shown below. The program was executed on a machine with a dual core Pentium, 2GB memory and running Ubuntu Jaunty.

Please note that:
  • I could not increase the number of Python processes to the order of 1000 (10^3). When I tried to do this the program crashed with a number of errors.
  • I got a curious QueueFeederThread exception when I executed the process for N = 100, M = 10000.

N M N * M (Wall) Time in seconds
Erlang Python
10 10 10^2 0.0 0.0732
10 100 10^3 0.003 0.1795
100 100 10^4 0.036 2.3976
100 1000 10^5 0.242 12.2252
1000 1000 10^6 1.877 -
1000 10000 10^7 18.593 -
10000 10000 10^8 197.477 -

1 comment:

bhumika said...

franchement, je comprends rien la :P