Small Block Allocator Challenge!

I recieved quite a few requests about the mini-coding contest (here, but scroll about half way down to the "Who's your Daddy?" section) that I mentioned.

To refresh your memory, the idea was to write a memory allocator that could substantially outperform the Microsoft CRT (and Windows OS) for small block allocations - inspired by some problems we had. You can read the rules or download the source to the benchmark.

Although I'm not releasing the code to my winning entry (just yet), I did include executables for both the winner (me :)) and for malloc()/free() compiled against the benchmark - they're in the source package referenced above.

The benchmark is NOT normalized for CPU/machine speed, so you should only compare your score to the winning executable (SREE_memtest.exe) run on your machine to see how you're doing. The benchmark will run twice when you launch it - the first run generates your actual "Score"; the second run is only to test validity.

Additionally, there is an "OS score" that it generates - this uses the same "total time" as the regular (official) score, but gets its efficiency measures from the OS rather than relying on you to call blockalloc()/blockfree(). We used that both to validate memory usage - and to compare to SmartHeap, which of course, does NOT strictly follow all the rules, but whatever :P

Be warned that you'll need a tiny bit of patience (the test performs well over 1000000 allocations) and a decent amount of RAM (it allocates over a half a gig) if you're going to run the benchmark(s).

Read More or Download it

It is absolutely possible to beat the winner- we spent a week (after hours) on the problem in isolation (i.e. to each, his own); it was clear after we compared notes at the end that there were lots of things that we each did that would hugely benefit the others' attempts.