More scalable pool allocator #50137

d-netto · 2023-06-12T00:39:08Z

Still very experimental and depends on #49644. Mostly for later comparisons with #48969 (e.g. if concurrent sweeping would perform better on a potentially less contended pool allocator).

Seems to perform better than master on some multithreaded allocation benchmarks (e.g. tree_mutable.jl), but a lot more benchmarking is needed:

master:

../julia-master/julia run_benchmarks.jl multithreaded binary_tree tree_mutable -n5 -t16 --gcthreads=1
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       7058 │    4033 │      2414 │       1619 │          350 │              1518 │     1631 │         57 │
│  median │       7535 │    4302 │      2484 │       1772 │          384 │              1820 │     1642 │         58 │
│ maximum │       7805 │    4848 │      2710 │       2369 │          415 │              1868 │     1667 │         62 │
│   stdev │        298 │     324 │       121 │        300 │           24 │               161 │       15 │          2 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

PR:

../julia-allocator/julia run_benchmarks.jl multithreaded binary_tree tree_mutable -n5 -t16 --gcthreads=1
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       6344 │    3894 │      1967 │       1922 │          356 │              1487 │     1950 │         59 │
│  median │       6486 │    4021 │      2058 │       2029 │          359 │              1645 │     1971 │         62 │
│ maximum │       7225 │    4321 │      2230 │       2232 │          413 │              1854 │     1982 │         63 │
│   stdev │        368 │     189 │       107 │        120 │           24 │               146 │       13 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

src/gc.h

oscardssmith · 2023-06-14T01:07:49Z

With all the changes since the beginning, are the benchmarks still the same?

d-netto · 2023-06-14T21:03:53Z

On 2983c05:

master:

../julia-master/julia run_benchmarks.jl multithreaded binary_tree -n10 -t32 --gcthreads=1
bench = "tree_immutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4392 │    2295 │      1280 │        868 │          183 │              1884 │      828 │         52 │
│  median │       4580 │    2435 │      1435 │       1058 │          203 │              2183 │      849 │         54 │
│ maximum │       4778 │    2624 │      1502 │       1138 │          212 │              2256 │      893 │         55 │
│   stdev │        128 │      99 │        66 │         82 │           10 │               111 │       20 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       7178 │    4143 │      2173 │       1939 │          403 │              1734 │     1802 │         58 │
│  median │       7512 │    4377 │      2229 │       2142 │          431 │              1852 │     1872 │         59 │
│ maximum │       7968 │    4903 │      2376 │       2661 │          468 │              2078 │     2007 │         62 │
│   stdev │        316 │     292 │        67 │        265 │           21 │               102 │       64 │          2 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

PR:

../julia-allocator/julia run_benchmarks.jl multithreaded binary_tree -n10 -t32 --gcthreads=1
bench = "tree_immutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       3890 │    2215 │      1059 │       1091 │          198 │              1457 │      947 │         57 │
│  median │       3980 │    2382 │      1176 │       1206 │          227 │              1591 │      968 │         60 │
│ maximum │       4258 │    2615 │      1219 │       1435 │          245 │              2832 │      981 │         61 │
│   stdev │        114 │     113 │        48 │        118 │           14 │               411 │        9 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       6526 │    4093 │      1983 │       2037 │          412 │              1428 │     2109 │         63 │
│  median │       6919 │    4439 │      2053 │       2387 │          465 │              1607 │     2127 │         64 │
│ maximum │       7981 │    5336 │      2456 │       3020 │          489 │              3584 │     2142 │         67 │
│   stdev │        518 │     402 │       151 │        275 │           25 │               647 │       11 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

oscardssmith · 2023-06-14T21:55:39Z

that looks pretty good!

src/gc.h

d-netto · 2023-06-15T22:46:20Z

Seeing some regressions on this PR on the single-threaded benchmarks. Marking as draft until further investigated.

d-netto · 2023-06-17T15:40:04Z

Benchmarks look fine on the latest commit (the goal of this PR is be performance-neutral on the serial benchmarks and to provide better scalability when there are multiple threads allocating).

Serial benchmarks (master)

../julia-master/julia run_benchmarks.jl serial all -n10
category = "TimeZones"
bench = "TimeZones.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1369 │     648 │       643 │          5 │          326 │                51 │     5099 │         47 │
│  median │       1384 │     663 │       657 │          5 │          333 │                61 │     5099 │         48 │
│ maximum │       2108 │    1268 │      1261 │         18 │          692 │                87 │     5099 │         60 │
│   stdev │        246 │     212 │       210 │          4 │          141 │                10 │        0 │          5 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "append"
bench = "append.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1984 │      42 │        23 │         19 │           59 │               314 │     1488 │          2 │
│  median │       1998 │      43 │        23 │         20 │           59 │               373 │     1488 │          2 │
│ maximum │       2098 │      60 │        29 │         32 │           89 │               407 │     1488 │          3 │
│   stdev │         35 │       7 │         2 │          5 │           12 │                30 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "bigint"
bench = "pollard.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1009 │     211 │       191 │         20 │           59 │               123 │      127 │         21 │
│  median │       1020 │     214 │       194 │         20 │           59 │               137 │      127 │         21 │
│ maximum │       1046 │     234 │       212 │         23 │           87 │               157 │      127 │         22 │
│   stdev │         11 │       7 │         6 │          1 │            9 │                 9 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "linked"
bench = "list.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4831 │    3193 │      2913 │        277 │         1056 │               224 │     2630 │         65 │
│  median │       4844 │    3209 │      2926 │        286 │         1060 │               240 │     2631 │         66 │
│ maximum │       5489 │    3592 │      3288 │        319 │         1307 │               278 │     2631 │         67 │
│   stdev │        203 │     122 │       115 │         13 │           78 │                18 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4525 │     588 │       578 │         10 │          391 │                43 │      129 │         13 │
│  median │       4564 │     602 │       589 │         12 │          394 │                73 │      131 │         13 │
│ maximum │       4860 │     649 │       629 │         21 │          421 │                77 │      131 │         13 │
│   stdev │        144 │      24 │        20 │          5 │           11 │                12 │        1 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "obj_arrays"
bench = "many_refs.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       2442 │    2052 │      1920 │        131 │          715 │                19 │      869 │         83 │
│  median │       2460 │    2067 │      1931 │        132 │          719 │                26 │      869 │         84 │
│ maximum │       2968 │    2475 │      2317 │        158 │          843 │                36 │      870 │         85 │
│   stdev │        168 │     137 │       129 │          9 │           39 │                 5 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "single_ref.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │        864 │     534 │       512 │         22 │          173 │                25 │      869 │         62 │
│  median │        875 │     541 │       518 │         22 │          176 │                39 │      869 │         62 │
│ maximum │       1096 │     697 │       670 │         26 │          290 │                44 │      869 │         64 │
│   stdev │         71 │      50 │        49 │          1 │           36 │                 7 │        0 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "strings"
bench = "strings.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │      23712 │    3470 │      3093 │        376 │           62 │               890 │      581 │         14 │
│  median │      24002 │    3491 │      3112 │        380 │           74 │              1279 │      581 │         15 │
│ maximum │      24757 │    3705 │      3291 │        421 │           87 │              1481 │      581 │         15 │
│   stdev │        325 │      82 │        67 │         16 │           10 │               177 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

Serial benchmarks (PR)

../julia-allocator/julia run_benchmarks.jl serial all -n10
category = "TimeZones"
bench = "TimeZones.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1399 │     650 │       644 │          5 │          324 │                60 │     5099 │         43 │
│  median │       1427 │     672 │       664 │          6 │          333 │                66 │     5099 │         47 │
│ maximum │       2083 │     989 │       985 │         19 │          649 │                90 │     5099 │         55 │
│   stdev │        227 │     124 │       123 │          4 │          113 │                 9 │        0 │          4 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "append"
bench = "append.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1952 │      44 │        24 │         20 │           60 │               330 │     1488 │          2 │
│  median │       2034 │      45 │        24 │         21 │           60 │               374 │     1488 │          2 │
│ maximum │       2282 │      62 │        33 │         31 │           91 │               458 │     1488 │          3 │
│   stdev │         92 │       7 │         3 │          4 │           14 │                44 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "bigint"
bench = "pollard.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       1010 │     208 │       187 │         21 │           60 │               122 │      127 │         20 │
│  median │       1024 │     211 │       190 │         21 │           60 │               138 │      127 │         21 │
│ maximum │       1101 │     241 │       214 │         26 │           96 │               171 │      127 │         22 │
│   stdev │         27 │      11 │         9 │          2 │           14 │                13 │        0 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "linked"
bench = "list.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4868 │    3232 │      2954 │        276 │         1075 │               165 │     2630 │         65 │
│  median │       4897 │    3261 │      2976 │        283 │         1077 │               244 │     2631 │         67 │
│ maximum │       5342 │    3462 │      3160 │        365 │         1117 │               294 │     2631 │         67 │
│   stdev │        143 │      71 │        64 │         27 │           12 │                37 │        1 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4583 │     625 │       615 │          9 │          425 │                53 │      130 │         14 │
│  median │       4625 │     629 │       619 │         10 │          427 │                76 │      131 │         14 │
│ maximum │       4757 │     647 │       634 │         13 │          429 │                86 │      131 │         14 │
│   stdev │         47 │       6 │         5 │          1 │            1 │                11 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "obj_arrays"
bench = "many_refs.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       2455 │    2061 │      1931 │        130 │          720 │                25 │      869 │         84 │
│  median │       2471 │    2074 │      1942 │        131 │          723 │                30 │      869 │         84 │
│ maximum │       2802 │    2355 │      2215 │        146 │          853 │                33 │      869 │         85 │
│   stdev │        125 │     110 │       105 │          5 │           46 │                 2 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "single_ref.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │        869 │     527 │       505 │         21 │          171 │                26 │      869 │         51 │
│  median │        879 │     537 │       515 │         22 │          175 │                38 │      869 │         61 │
│ maximum │       1140 │     680 │       654 │         26 │          273 │                42 │      869 │         62 │
│   stdev │        108 │      46 │        45 │          2 │           31 │                 6 │        0 │          3 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
category = "strings"
bench = "strings.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │      23607 │    3500 │      3109 │        390 │           63 │               963 │      581 │         15 │
│  median │      24122 │    3553 │      3159 │        393 │           76 │              1319 │      581 │         15 │
│ maximum │      24898 │    3705 │      3273 │        432 │           91 │              1453 │      581 │         15 │
│   stdev │        323 │      58 │        46 │         13 │           11 │               169 │        0 │          0 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

Multithreaded benchmarks (master)

../julia-master/julia run_benchmarks.jl multithreaded binary_tree -n10 -t32 --gcthreads=1
bench = "tree_immutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       4370 │    2455 │      1310 │       1093 │          185 │              1706 │      798 │         55 │
│  median │       4624 │    2565 │      1386 │       1166 │          208 │              1948 │      914 │         57 │
│ maximum │       4854 │    2809 │      1474 │       1409 │          235 │              1998 │     1005 │         58 │
│   stdev │        162 │     127 │        52 │        126 │           15 │                97 │       60 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       7251 │    4282 │      2150 │       2092 │          400 │              1575 │     1806 │         58 │
│  median │       7776 │    4644 │      2327 │       2313 │          416 │              1844 │     1905 │         60 │
│ maximum │       8286 │    5207 │      2442 │       2824 │          450 │              2543 │     1964 │         63 │
│   stdev │        361 │     296 │       104 │        222 │           15 │               275 │       45 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

Multithreaded benchmarks (PR)

../julia-allocator/julia run_benchmarks.jl multithreaded binary_tree -n10 -t32 --gcthreads=1
bench = "tree_immutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       3963 │    2355 │      1089 │       1216 │          200 │              1450 │      912 │         58 │
│  median │       4203 │    2467 │      1193 │       1303 │          222 │              1523 │      954 │         59 │
│ maximum │       4566 │    2779 │      1225 │       1575 │          269 │              7082 │      969 │         63 │
│   stdev │        163 │     129 │        42 │        125 │           20 │              1750 │       18 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘
bench = "tree_mutable.jl"
┌─────────┬────────────┬─────────┬───────────┬────────────┬──────────────┬───────────────────┬──────────┬────────────┐
│         │ total time │ gc time │ mark time │ sweep time │ max GC pause │ time to safepoint │ max heap │ percent gc │
│         │         ms │      ms │        ms │         ms │           ms │                us │       MB │          % │
├─────────┼────────────┼─────────┼───────────┼────────────┼──────────────┼───────────────────┼──────────┼────────────┤
│ minimum │       6814 │    4322 │      1930 │       2385 │          432 │              1300 │     2077 │         63 │
│  median │       7081 │    4644 │      2021 │       2634 │          466 │              1449 │     2120 │         66 │
│ maximum │       7752 │    5257 │      2082 │       3197 │          487 │              1641 │     2142 │         68 │
│   stdev │        278 │     264 │        54 │        237 │           17 │                85 │       18 │          1 │
└─────────┴────────────┴─────────┴───────────┴────────────┴──────────────┴───────────────────┴──────────┴────────────┘

Machine information

Architecture:            x86_64
  CPU op-mode(s):        32-bit, 64-bit
  Address sizes:         43 bits physical, 48 bits virtual
  Byte Order:            Little Endian
CPU(s):                  128
  On-line CPU(s) list:   0-127
Vendor ID:               AuthenticAMD
  Model name:            AMD EPYC 7502 32-Core Processor
    CPU family:          23
    Model:               49
    Thread(s) per core:  2
    Core(s) per socket:  32
    Socket(s):           2
    Stepping:            0
    Frequency boost:     enabled
    CPU max MHz:         2500.0000
    CPU min MHz:         1500.0000
    BogoMIPS:            4999.53
    Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonsto
                         p_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse
                         4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase
                          bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdp
                         ru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov
                          succor smca sme sev sev_es
Virtualization features: 
  Virtualization:        AMD-V
Caches (sum of all):     
  L1d:                   2 MiB (64 instances)
  L1i:                   2 MiB (64 instances)
  L2:                    32 MiB (64 instances)
  L3:                    256 MiB (16 instances)
NUMA:                    
  NUMA node(s):          2
  NUMA node0 CPU(s):     0-31,64-95
  NUMA node1 CPU(s):     32-63,96-127
Vulnerabilities:         
  Itlb multihit:         Not affected
  L1tf:                  Not affected
  Mds:                   Not affected
  Meltdown:              Not affected
  Mmio stale data:       Not affected
  Retbleed:              Mitigation; untrained return thunk; SMT enabled with STIBP protection
  Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl and seccomp
  Spectre v1:            Mitigation; usercopy/swapgs barriers and __user pointer sanitization
  Spectre v2:            Mitigation; Retpolines, IBPB conditional, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
  Srbds:                 Not affected
  Tsx async abort:       Not affected

vchuravy · 2023-06-24T12:17:55Z

@nanosoldier runtests(configuration = (buildflags=["LLVM_ASSERTIONS=1", "FORCE_ASSERTIONS=1"],), vs_configuration = (buildflags = ["LLVM_ASSERTIONS=1", "FORCE_ASSERTIONS=1"],))

nanosoldier · 2023-06-24T22:06:38Z

Your package evaluation job has completed - possible new issues were detected.
A full report can be found here.

gbaraldi · 2023-06-26T13:53:55Z

This made the gcext tests fail, they are allow fail but it might deserve a loock (the tests might just be doing some wrong assumptions)

vchuravy · 2023-06-26T13:58:16Z

Yeah it's on my to-do list

d-netto · 2023-06-26T15:30:23Z

Also taking a look at this.

src/gc.c

src/gc.h

src/gc.c

…ect pool freelist * addresses follow-up comments from #50137, particularly #50137 (comment)

d-netto marked this pull request as draft June 12, 2023 00:39

d-netto requested review from gbaraldi and vchuravy June 12, 2023 00:39

d-netto added the GC Garbage collector label Jun 12, 2023

vtjnash reviewed Jun 12, 2023

View reviewed changes

src/gc.h Outdated Show resolved Hide resolved

d-netto force-pushed the dcn/allocator branch 6 times, most recently from b01d688 to 369ced8 Compare June 14, 2023 00:48

d-netto force-pushed the dcn/allocator branch 2 times, most recently from 65a2b4d to 2983c05 Compare June 14, 2023 20:41

vtjnash reviewed Jun 15, 2023

View reviewed changes

src/gc.h Outdated Show resolved Hide resolved

d-netto force-pushed the dcn/allocator branch from 2983c05 to 8a31167 Compare June 15, 2023 11:56

d-netto marked this pull request as ready for review June 15, 2023 18:46

d-netto marked this pull request as draft June 15, 2023 22:46

d-netto force-pushed the dcn/allocator branch from 8a31167 to acb4110 Compare June 17, 2023 02:56

d-netto marked this pull request as ready for review June 17, 2023 15:40

d-netto changed the title ~~Less contended pool allocator~~ More scalable pool allocator Jun 17, 2023

d-netto added this to the 1.10 milestone Jun 17, 2023

d-netto added the performance Must go faster label Jun 17, 2023

d-netto mentioned this pull request Jun 17, 2023

Concurrent page sweeping #48969

Merged

d-netto force-pushed the dcn/allocator branch 2 times, most recently from 4fe05fa to acb4110 Compare June 17, 2023 22:51

vchuravy approved these changes Jun 25, 2023

View reviewed changes

vchuravy merged commit 5939e2d into JuliaLang:master Jun 25, 2023

vtjnash reviewed Jun 27, 2023

View reviewed changes

src/gc.c Show resolved Hide resolved

vtjnash reviewed Jun 27, 2023

View reviewed changes

src/gc.c Show resolved Hide resolved

vtjnash reviewed Jun 27, 2023

View reviewed changes

src/gc.c Show resolved Hide resolved

vtjnash reviewed Jun 28, 2023

View reviewed changes

src/gc.h Show resolved Hide resolved

vtjnash reviewed Jun 28, 2023

View reviewed changes

src/gc.c Show resolved Hide resolved

dkarrasch mentioned this pull request Jun 30, 2023

ensure objects beyond bump allocated region are inserted into the object pool freelist #50357

Merged

inkydragon mentioned this pull request Jun 30, 2023

Build with cflag GC_FINAL_STATS failed #50368

Closed

d-netto added a commit that referenced this pull request Jun 30, 2023

ensure objects beyond bump allocated region are inserted into the obj…

530113f

…ect pool freelist * addresses follow-up comments from #50137, particularly #50137 (comment)

d-netto mentioned this pull request Jul 13, 2023

gcext test is often failing #50434

Closed

d-netto added a commit to RelationalAI/julia that referenced this pull request Aug 6, 2023

Backport Improve scalability of page allocator (JuliaLang#50137)

10c2c85

d-netto added a commit to RelationalAI/julia that referenced this pull request Aug 6, 2023

Backport Improve scalability of page allocator (JuliaLang#50137)

ad9a47d

d-netto mentioned this pull request Aug 6, 2023

Backport allocation improvements RelationalAI/julia#29

Closed

d-netto mentioned this pull request Sep 6, 2023

Backport allocation improvements RelationalAI/julia#37

Merged

kpamnany pushed a commit to RelationalAI/julia that referenced this pull request Oct 19, 2023

More scalable pool allocator (JuliaLang#50137)

d7d5a07

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Oct 20, 2023

More scalable pool allocator (JuliaLang#50137)

381bdef

kpamnany pushed a commit to RelationalAI/julia that referenced this pull request Oct 21, 2023

More scalable pool allocator (JuliaLang#50137)

14fd2c7

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Oct 23, 2023

More scalable pool allocator (JuliaLang#50137)

dd62648

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 1, 2023

More scalable pool allocator (JuliaLang#50137)

673b3ee

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 2, 2023

More scalable pool allocator (JuliaLang#50137)

0bbb9a9

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 7, 2023

More scalable pool allocator (JuliaLang#50137)

fbc9441

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 10, 2023

More scalable pool allocator (JuliaLang#50137)

87a861d

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 14, 2023

More scalable pool allocator (JuliaLang#50137)

f73192b

DelveCI pushed a commit to RelationalAI/julia that referenced this pull request Nov 15, 2023

More scalable pool allocator (JuliaLang#50137)

e1816e9

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

More scalable pool allocator #50137

More scalable pool allocator #50137

d-netto commented Jun 12, 2023

oscardssmith commented Jun 14, 2023

d-netto commented Jun 14, 2023

oscardssmith commented Jun 14, 2023

d-netto commented Jun 15, 2023

d-netto commented Jun 17, 2023

vchuravy commented Jun 24, 2023

nanosoldier commented Jun 24, 2023

gbaraldi commented Jun 26, 2023

vchuravy commented Jun 26, 2023

d-netto commented Jun 26, 2023

More scalable pool allocator #50137

More scalable pool allocator #50137

Conversation

d-netto commented Jun 12, 2023

oscardssmith commented Jun 14, 2023

d-netto commented Jun 14, 2023

oscardssmith commented Jun 14, 2023

d-netto commented Jun 15, 2023

d-netto commented Jun 17, 2023

vchuravy commented Jun 24, 2023

nanosoldier commented Jun 24, 2023

gbaraldi commented Jun 26, 2023

vchuravy commented Jun 26, 2023

d-netto commented Jun 26, 2023