Leopold Toetsch <[EMAIL PROTECTED]> wrote: > Dan Sugalski <[EMAIL PROTECTED]> wrote:
>> *) Lay in a generational garbage collector. > Good for that degenerated case, probably, but see below. A generational GC still has to mark the root set and the old generation's objects. So it wouldn't be of any help with the degenerated case. $ parrot -p 10m.pasm 1e6 CODE J OP FULL NAME CALLS TOTAL TIME AVG T. ms ---- - ----------------- -------- ---------- ---------- -5 DOD_mark_next 21 1.244545 59.2640 751 j new_p_ic 1000001 0.357744 0.0004 935 j set_p_ki_p 1000000 0.294898 0.0003 -4 DOD_collect_PMC 21 0.191554 9.1216 DOD_mark_next is from marking the root set, i.e. the array. DOD_collect_PMCs is that part, that could profit from a generational approach. But ... $ parrot -p 10m.pasm 2e6 CODE J OP FULL NAME CALLS TOTAL TIME AVG T. ms ---- - ----------------- -------- ---------- ---------- -5 DOD_mark_next 40 4.751935 118.7984 -4 DOD_collect_PMC 40 0.750301 18.7575 751 j new_p_ic 2000001 0.687316 0.0003 ... as you can see, the collection phases are still linear in execution time and collection is fast. The accumulated DOD run counts + ever increasing DOD times together give a nice quadratic slow down: $ parrot -p 10m.pasm 3e6 CODE J OP FULL NAME CALLS TOTAL TIME AVG T. ms ---- - ----------------- -------- ---------- ---------- -5 DOD_mark_next 59 10.502405 178.0069 -4 DOD_collect_PMC 59 1.667929 28.2700 leo
