On Fri, Feb 7, 2020 at 5:53 PM Han-Wen Nienhuys <hanw...@gmail.com> wrote:
> Thanks, I ran the carver score successfully now. > > I took some pauses between runs so thermal throttling wasn't an issue. > > My standard guile2.2 branch (with 40M initial heap), takes about 2m wall > time, 1m of GC time. > > The stats printing shows that we have a very full heap: > > In-use heap: 85% (370824 KiB pointers + 60065 KiB other) > In-use heap: 85% (370725 KiB pointers + 59737 KiB other) > In-use heap: 85% (370143 KiB pointers + 59687 KiB other) > .. > > so we have to scan a lot of live data to reclaim just 15% of the heap. > > Tinkering with INITIAL_HEAP lets us tune things a little bit: > > initial heap 2G: 50s (2.69s GC) > initial heap 900M: 0:54 (6s GC) > initial heap 500M: 1:09 (33s GC) > > The GC timings seem wonky, but since BDW is multithreaded, it's possible > that the computation times don't completely add up. > > Single threaded, the numbers make more sense: MAX=INIT=2G gc time taken: 1.843968805 User time (seconds): 49.36 MAX=INIT=1G gc time taken: 3.291264925 User time (seconds): 51.74 MAX=INIT=800M gc time taken: 5.760042906 User time (seconds): 54.62 500M gc time taken: 17.921457247 User time (seconds): 68.24 It's interesting to note that the multithreaded collector doesn't really help. With a 500M heap, wall clock is 1:08 for the single threaded case, and 1:09 for the multithreaded case. The 800M case seems like a good configuration: it spends about 10% of the time doing GC. $ grep In-use log In-use heap: 0% (0 KiB pointers + 0 KiB other) In-use heap: 2% (20959 KiB pointers + 2430 KiB other) In-use heap: 2% (14698 KiB pointers + 2415 KiB other) In-use heap: 39% (292008 KiB pointers + 29486 KiB other) In-use heap: 47% (334968 KiB pointers + 53702 KiB other) In-use heap: 45% (317237 KiB pointers + 53424 KiB other) In-use heap: 57% (407528 KiB pointers + 67897 KiB other) In-use heap: 53% (378243 KiB pointers + 60449 KiB other) In-use heap: 53% (378249 KiB pointers + 60519 KiB other) In-use heap: 53% (377488 KiB pointers + 60322 KiB other) so we should scale the heap so that approximately 50% is collected on GC. -- Han-Wen Nienhuys - hanw...@gmail.com - http://www.xs4all.nl/~hanwen
