From: Godmar Back (gback@cs.utah.edu)
Date: Tue Jan 05 1999 - 14:19:49 EST
So, I picked 5MB as the default value for the initial heap size.
Given that kaffe wastes about 2.5MB on we don't know what yet this
seemed a reasonable size. (actually, it was the size that made run
iute the fastest with the fewest # of gcs --- here I am again, optimizing
random benchmarks)
More seriously, this should maybe only 4MB once we've figured out where
all this memory goes. It shouldn't matter much.
I also added an option "-as" to specify the heap increment, which
defaults to 1MB.
Note also that I fixed cmd line parsing such that both -ms 64m and -ms64m
work now.
Enjoy,
- Godmar
PS:
btw, Sun 1.2 under Windows runs that same test on a slightly faster
machine (350 vs. 333Mhz) in 1.4 seconds!
Aside from the better quality of their jit, messages like this also
tell me that there is no way around precise and compacting gc:
<GC: managing allocation failure: need 1032 bytes, type=1, action=1>
<GC: 62 milliseconds since last GC>
<GC: compactHeap took 5 ms, swap time = 1 ms, blocks_moved=22275>
<GC: 0 explicitly pinned objects, 3 conservatively pinned objects>
<GC: last free block at 0x01A089E4 of length 1560>
<GC: last free block is at end>
<GC: freed 22259 objects, 655008 bytes in 17 ms, 78% free (656056/838856)>
<GC: init&scan: 1 ms, scan handles: 4 ms, sweep: 6 ms, compact: 6 ms>
<GC: 0 register-marked objects, 3 stack-marked objects>
<GC: 0 register-marked handles, 43 stack-marked handles>
<GC: refs: soft 0 (age >= 32), weak 0, final 0, phantom 0>
Yet another reason to not invest time in Boehm.
This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 19:57:29 EDT