Because the average ms calculation is rounded, it's not apparent that the incremental GC actually runs 25% faster unless you do the math.
On Tuesday, December 31, 2002, at 01:40 AM, Marcus Denker wrote:
On Mon, Dec 30, 2002 at 12:27:19AM -0800, John M McIntosh wrote:
This change set provides a 18 - 28% improvement in the GC logic using an test image that has 1.1 million objects in it on the PowerPC.
Here's a macroBenchmark (on a G3-400 Powerbook):
normal VM: #(23176 165464 71470 26862 0 23717 14090) with GC changes: #(22306 154462 67454 24451 0 21615 14134) app. %faster %3.9 %7.1 %5.9 %9.8 %9.7 0%
Benchmark #2 triggers 10 full collections, it shows nicely how much faster the GC got:
Benchmark #2: 165464ms
uptime 165.5s full 10 totalling 5,920ms (4.0% uptime), avg 592.0ms incr 15932 totalling 33,728ms (20.0% uptime), avg 2.0ms tenures 1,550 (avg 10 GCs/tenure)
Benchmark #2: 154462ms
uptime 154.5s full 10 totalling 4,781ms (3.0% uptime), avg 478.0ms incr 16036 totalling 25,407ms (16.0% uptime), avg 2.0ms tenures 1,554 (avg 10 GCs/tenure)
-- Marcus Denker marcus@ira.uka.de -- Squeak! http://squeak.de
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com 1-800-477-2659 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===