[MLton-devel] mark-compact-ratio

[Stephen Weeks]

I did some experiments on my laptop with the mark-compact-ratio (mcr),
the minimum ratio of heap size to live data size that the runtime will
use.  I wanted to check out my claim that it made sense to make the
mcr bigger (1.5) than it used to be (1.25) to cut down on paging by
more use of the generational gc.

Boy was I wrong.

In fact, a much smaller mcr gives much better performance in the
presence of paging.  Below, are the the results of a bunch of self
compiles on my laptop with different mcrs, and the corresponding
elapsed, mutator and gc time.

mcr	elapsed	mutator	gc
----	------- -------	---
1.01	21:07	541	568
1.02	20:57	547	541
1.03	20:51	550	532
1.04	19:59	555	508
1.05	20:49	563	515
1.06	21:07	553	495
1.07	23:54	567	492
1.08	21:16	578	507
1.09	23:55	580	485
1.10	22:14	586	508
...
1.15	28:57	548	446
...
1.20	26:45	542	441
...
1.25	39:47	541	438

The best was at 1.04, with only 10% paging cost.  Pretty nice to be
able to do a self-compile in <20 minutes with only 256M.

Anyways, I've set the default mcr to 1.04 for now.  People can play
around with it with their own applications.


-------------------------------------------------------
This sf.net email is sponsored by: OSDN - Tired of that same old
cell phone?  Get a new here for FREE!
https://www.inphonic.com/r.asp?r=sourceforge1&refcode1=vs3390
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel