mirror
/
restic
mirror of https://github.com/restic/restic.git
1
0
Fork 0
Code Issues Releases Wiki Activity
5,980 Commits 7 Branches 48 Tags 92 MiB
Commit Graph

5 Commits

Author SHA1 Message Date
greatroar 5141228e0c repository: Re-tune indexmap allocation strategy 
fd05037e1a changed the allocation batch
size from 256 to 128 under the assumption that an indexEntry is 60 bytes
on amd64, but it's 64: structs are padded out to a multiple of 8 for
alignment reasons. That means we'd waste no space in malloc even without
the batch allocation, at least on 64-bit machines. While that strategy
cuts the overallocation down dramatically for many small indexes, it also
seems to slow allocation down (Go 1.18, Linux, amd64, -benchtime=2s):

    name                   old time/op    new time/op    delta
    DecodeIndex-8             4.67s ± 5%     4.60s ± 1%      ~     (p=0.953 n=10+5)
    DecodeIndexParallel-8     4.67s ± 3%     4.60s ± 1%      ~     (p=0.953 n=10+5)
    IndexHasUnknown-8        37.8ns ± 8%    36.5ns ±14%      ~     (p=0.841 n=5+5)
    IndexHasKnown-8          38.5ns ±12%    37.7ns ±10%      ~     (p=0.968 n=5+5)
    IndexAlloc-8              615ms ±18%     607ms ± 1%      ~     (p=1.000 n=10+5)
    IndexAllocParallel-8      245ms ±11%     285ms ± 6%   +16.40%  (p=0.001 n=10+5)
    MasterIndexAlloc-8        286ms ± 9%     275ms ± 2%      ~     (p=1.000 n=10+5)
    LoadIndex/v1-8           27.0ms ± 4%    26.8ms ± 1%      ~     (p=0.690 n=5+5)
    LoadIndex/v2-8           22.4ms ± 1%    22.8ms ± 2%    +1.48%  (p=0.016 n=5+5)

    name                   old alloc/op   new alloc/op   delta
    IndexAlloc-8              446MB ± 0%     446MB ± 0%    -0.00%  (p=0.000 n=8+4)
    IndexAllocParallel-8      446MB ± 0%     446MB ± 0%    -0.00%  (p=0.008 n=8+5)
    MasterIndexAlloc-8        213MB ± 0%     159MB ± 0%   -25.47%  (p=0.000 n=10+5)

    name                   old allocs/op  new allocs/op  delta
    IndexAlloc-8               913k ± 0%     2632k ± 0%  +188.19%  (p=0.008 n=5+5)
    IndexAllocParallel-8       913k ± 0%     2632k ± 0%  +188.21%  (p=0.008 n=5+5)
    MasterIndexAlloc-8         318k ± 0%     1172k ± 0%  +267.86%  (p=0.008 n=5+5)

Instead, this patch sets a batch size of 4, which means no space is
wasted by malloc on 64-bit and very little on 32-bit. It still gets very
close to the savings from not allocating in batches, without requiring
special code for bits.UintSize==64. Benchmark results, again for
Linux/amd64:

    name                   old time/op    new time/op    delta
    DecodeIndex-8             4.67s ± 5%     4.83s ± 9%     ~     (p=0.315 n=10+10)
    DecodeIndexParallel-8     4.67s ± 3%     4.68s ± 4%     ~     (p=0.315 n=10+10)
    IndexHasUnknown-8        37.8ns ± 8%    44.5ns ±19%     ~     (p=0.095 n=5+5)
    IndexHasKnown-8          38.5ns ±12%    36.9ns ± 8%     ~     (p=0.690 n=5+5)
    IndexAlloc-8              615ms ±18%     628ms ±18%     ~     (p=0.218 n=10+10)
    IndexAllocParallel-8      245ms ±11%     262ms ± 9%   +7.02%  (p=0.043 n=10+10)
    MasterIndexAlloc-8        286ms ± 9%     287ms ±13%     ~     (p=1.000 n=10+10)
    LoadIndex/v1-8           27.0ms ± 4%    26.8ms ± 0%     ~     (p=1.000 n=5+5)
    LoadIndex/v2-8           22.4ms ± 1%    22.5ms ± 0%     ~     (p=0.056 n=5+5)

    name                   old alloc/op   new alloc/op   delta
    IndexAlloc-8              446MB ± 0%     446MB ± 0%     ~     (p=1.000 n=8+10)
    IndexAllocParallel-8      446MB ± 0%     446MB ± 0%   -0.00%  (p=0.000 n=8+8)
    MasterIndexAlloc-8        213MB ± 0%     160MB ± 0%  -25.02%  (p=0.000 n=10+9)

    name                   old allocs/op  new allocs/op  delta
    IndexAlloc-8               913k ± 0%     1333k ± 0%  +45.94%  (p=0.000 n=8+10)
    IndexAllocParallel-8       913k ± 0%     1333k ± 0%  +45.94%  (p=0.000 n=8+8)
    MasterIndexAlloc-8         318k ± 0%      525k ± 0%  +64.99%  (p=0.000 n=10+10)

The allocation method indexmap.newEntry has also been rewritten in a
form that is a few instructions shorter.
 2022-05-11 21:22:14 +02:00
Michael Eischer fd05037e1a repository: recalibrate index batch allocation size 2022-04-30 11:34:10 +02:00
Michael Eischer 6fb408d90e repository: implement pack compression 2022-04-30 11:34:10 +02:00
greatroar 8d2996eaaa Replace siphash by hash/maphash 
In Go 1.17.1, maphash has become quite a bit faster than siphash, so we
can drop one third-party dependency. maphash is just an interface to the
standard Go map's hash function, which we already trust for other use
cases.

Benchmark results on linux/amd64, -benchtime=3s:

name                                             old time/op    new time/op    delta
IndexHasUnknown-8                                  50.6ns ±10%    41.0ns ±19%  -18.92%  (p=0.000 n=9+10)
IndexHasKnown-8                                    52.6ns ±12%    41.5ns ±12%  -21.13%  (p=0.000 n=9+10)
IndexMapHash-8                                     3.64µs ± 1%    2.00µs ± 0%  -45.09%  (p=0.000 n=10+9)
IndexAlloc-8                                        700ms ± 1%     601ms ± 6%  -14.18%  (p=0.000 n=8+10)
IndexAllocParallel-8                                205ms ± 5%     192ms ± 8%   -6.18%  (p=0.043 n=10+10)
MasterIndexAlloc-8                                  319ms ± 1%     279ms ± 5%  -12.58%  (p=0.000 n=10+10)
MasterIndexLookupSingleIndex-8                      156ns ± 8%     147ns ± 6%   -5.46%  (p=0.023 n=10+10)
MasterIndexLookupMultipleIndex-8                    150ns ± 7%     142ns ± 8%   -5.69%  (p=0.007 n=10+10)
MasterIndexLookupSingleIndexUnknown-8              74.4ns ± 6%    72.0ns ± 9%     ~     (p=0.175 n=10+9)
MasterIndexLookupMultipleIndexUnknown-8            67.4ns ± 9%    65.5ns ± 7%     ~     (p=0.340 n=9+9)
MasterIndexLookupParallel/known,indices=25-8        461ns ± 2%     445ns ± 2%   -3.49%  (p=0.000 n=10+10)
MasterIndexLookupParallel/unknown,indices=25-8      408ns ±11%     378ns ± 5%   -7.22%  (p=0.035 n=10+9)
MasterIndexLookupParallel/known,indices=50-8        479ns ± 1%     437ns ± 4%   -8.82%  (p=0.000 n=10+10)
MasterIndexLookupParallel/unknown,indices=50-8      406ns ± 8%     343ns ±15%  -15.44%  (p=0.001 n=10+10)
MasterIndexLookupParallel/known,indices=100-8       480ns ± 1%     455ns ± 5%   -5.15%  (p=0.000 n=8+10)
MasterIndexLookupParallel/unknown,indices=100-8     391ns ±18%     382ns ± 8%     ~     (p=0.315 n=10+10)
MasterIndexLookupBlobSize-8                        71.0ns ± 8%    57.2ns ±11%  -19.36%  (p=0.000 n=9+10)
PackerManager-8                                     254ms ± 1%     254ms ± 1%     ~     (p=0.285 n=15+15)

name                                             old speed      new speed      delta
IndexMapHash-8                                   1.12GB/s ± 1%  2.05GB/s ± 0%  +82.13%  (p=0.000 n=10+9)
PackerManager-8                                   208MB/s ± 1%   207MB/s ± 1%     ~     (p=0.281 n=15+15)

name                                             old alloc/op   new alloc/op   delta
IndexMapHash-8                                      0.00B          0.00B          ~     (all equal)
IndexAlloc-8                                        400MB ± 0%     400MB ± 0%     ~     (p=1.000 n=9+10)
IndexAllocParallel-8                                401MB ± 0%     401MB ± 0%   +0.00%  (p=0.000 n=10+10)
MasterIndexAlloc-8                                  258MB ± 0%     262MB ± 0%   +1.42%  (p=0.000 n=9+10)
PackerManager-8                                    73.1kB ± 0%    73.1kB ± 0%     ~     (p=0.382 n=13+13)

name                                             old allocs/op  new allocs/op  delta
IndexMapHash-8                                       0.00           0.00          ~     (all equal)
IndexAlloc-8                                         907k ± 0%      907k ± 0%   -0.00%  (p=0.000 n=10+10)
IndexAllocParallel-8                                 907k ± 0%      907k ± 0%   +0.00%  (p=0.009 n=10+10)
MasterIndexAlloc-8                                   327k ± 0%      317k ± 0%   -3.06%  (p=0.000 n=10+10)
PackerManager-8                                       744 ± 0%       744 ± 0%     ~     (all equal)
 2021-09-19 16:05:18 +02:00
greatroar 7bda28f31f Chaining hash table for repository.Index 
These are faster to construct but slower to access. The allocation rate
is halved, the peak memory usage almost halved compared to standard map.

Benchmark results on linux/amd64, -benchtime=3s -count=20:

name                                            old time/op    new time/op    delta
PackerManager-8                                    178ms ± 0%     178ms ± 0%     ~     (p=0.231 n=20+20)
DecodeIndex-8                                      4.54s ± 0%     4.30s ± 0%   -5.20%  (p=0.000 n=18+17)
DecodeIndexParallel-8                              4.54s ± 0%     4.30s ± 0%   -5.22%  (p=0.000 n=19+18)
IndexHasUnknown-8                                 44.4ns ± 5%    50.5ns ±11%  +13.82%  (p=0.000 n=19+17)
IndexHasKnown-8                                   48.3ns ± 0%    51.5ns ±12%   +6.68%  (p=0.001 n=16+20)
IndexAlloc-8                                       758ms ± 1%     616ms ± 1%  -18.69%  (p=0.000 n=19+19)
IndexAllocParallel-8                               234ms ± 3%     204ms ± 2%  -12.60%  (p=0.000 n=20+18)
MasterIndexLookupSingleIndex-8                     122ns ± 0%     145ns ± 9%  +18.44%  (p=0.000 n=14+20)
MasterIndexLookupMultipleIndex-8                   369ns ± 2%     429ns ± 8%  +16.27%  (p=0.000 n=20+20)
MasterIndexLookupSingleIndexUnknown-8             68.4ns ± 5%    74.9ns ±13%   +9.47%  (p=0.000 n=20+20)
MasterIndexLookupMultipleIndexUnknown-8            315ns ± 3%     369ns ±11%  +17.14%  (p=0.000 n=20+20)
MasterIndexLookupParallel/known,indices=5-8        743ns ± 1%     816ns ± 2%   +9.87%  (p=0.000 n=17+17)
MasterIndexLookupParallel/unknown,indices=5-8      238ns ± 1%     260ns ± 2%   +9.14%  (p=0.000 n=19+20)
MasterIndexLookupParallel/known,indices=10-8      1.01µs ± 3%    1.11µs ± 2%   +9.79%  (p=0.000 n=19+20)
MasterIndexLookupParallel/unknown,indices=10-8     222ns ± 0%     269ns ± 2%  +20.83%  (p=0.000 n=16+20)
MasterIndexLookupParallel/known,indices=20-8      1.06µs ± 2%    1.19µs ± 2%  +12.95%  (p=0.000 n=19+18)
MasterIndexLookupParallel/unknown,indices=20-8     413ns ± 1%     530ns ± 1%  +28.19%  (p=0.000 n=18+20)
SaveAndEncrypt-8                                  30.2ms ± 1%    30.4ms ± 0%   +0.71%  (p=0.000 n=19+19)
LoadTree-8                                         540µs ± 1%     576µs ± 1%   +6.73%  (p=0.000 n=20+20)
LoadBlob-8                                        5.64ms ± 0%    5.64ms ± 0%     ~     (p=0.883 n=18+17)
LoadAndDecrypt-8                                  5.93ms ± 0%    5.95ms ± 1%     ~     (p=0.247 n=20+19)
LoadIndex-8                                       25.1ms ± 0%    24.5ms ± 1%   -2.54%  (p=0.000 n=18+17)

name                                            old speed      new speed      delta
PackerManager-8                                  296MB/s ± 0%   296MB/s ± 0%     ~     (p=0.229 n=20+20)
SaveAndEncrypt-8                                 139MB/s ± 1%   138MB/s ± 0%   -0.71%  (p=0.000 n=19+19)
LoadBlob-8                                       177MB/s ± 0%   177MB/s ± 0%     ~     (p=0.890 n=18+17)
LoadAndDecrypt-8                                 169MB/s ± 0%   168MB/s ± 1%     ~     (p=0.227 n=20+19)

name                                            old alloc/op   new alloc/op   delta
PackerManager-8                                   91.8kB ± 0%    91.8kB ± 0%     ~     (p=0.772 n=12+19)
IndexAlloc-8                                       786MB ± 0%     400MB ± 0%  -49.04%  (p=0.000 n=20+18)
IndexAllocParallel-8                               786MB ± 0%     401MB ± 0%  -49.04%  (p=0.000 n=19+15)
SaveAndEncrypt-8                                  21.0MB ± 0%    21.0MB ± 0%   +0.00%  (p=0.000 n=19+19)

name                                            old allocs/op  new allocs/op  delta
PackerManager-8                                    1.41k ± 0%     1.41k ± 0%     ~     (all equal)
IndexAlloc-8                                        977k ± 0%      907k ± 0%   -7.18%  (p=0.000 n=20+20)
IndexAllocParallel-8                                977k ± 0%      907k ± 0%   -7.17%  (p=0.000 n=19+15)
SaveAndEncrypt-8                                    73.0 ± 0%      73.0 ± 0%     ~     (all equal)
 2020-07-19 13:58:22 +02:00