- table
- setup 0: macos histo, macos moving avg, comanche histo, comanche moving avg
- setup 1: macos histo, macos moving avg, comanche histo, comanche moving avg
- setup 2: macos histo, macos moving avg
- setup 3: macos histo, macos moving avg, comanche histo, comanche moving avg
- setup 4: macos histo, macos moving avg
The commit per sec
difference between macos
and comanche
is mainly due to the performances decay of index
over time. See any moving_average
png file.
- 0 - Baseline
- 1 - New index
- 2 - New index, index log size x 2
- 3 - Flattened paths, new index, new inode conf (new hashes)
- 4 - Flattened paths, new index, new inode conf, index log size x 2 (new hashes)
- On macbook
- 100k commits, repeated 3 times
- Ocaml 4.11.1
- On comanche
- 300k commits
- OCaml 4.11.0
- Irmin version
0
: 2.5.1 (branch Ngoguey42/no-irmin)1, 2, 3, 4
: branch Ngoguey42/split-inode-config-io-allocations
- Index version
0
: 1.3.01, 2, 3, 4
: mirage/index#279
- Inode config:
0, 1, 2
:entries=32
,stable_hash=256
(max_leaf_size = branching_factor = 32
,stable_hash=256
)3, 4
:max_leaf_size = stable_hash = 21
,branching_factor = 8
(no more stable hashes for trees)