Hello Dan, Monday, April 6, 2009, 12:35:14 PM, you wrote:
the size of the sub-array. The test then fills a 10 million element array.
However, something about the benchmark makes it perform poorly for both small chunks and large chunks. -sstderr reports that lots of copying occurs for small chunk sizes, and I haven't bothered to figure out why this is the case. You can, however, see that marking dirty chunks in this fashion would be profitable. The un-chunked array takes around a minute here, while with chunks of 10,000 (which seems to be about the optimal value with the above copying tradeoff), it takes about 6 seconds, and that's still with 60+% GC time.
i don't think that 60% GC time is bad for *this* benchmark. array filling is very trivial operation, after all. important part is 10x GC times reduce, apply these numbers to original benchmark -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com