be more memory friendly#5865
Merged
edolstra merged 6 commits intoNixOS:masterfrom Mar 11, 2022
pennae:memory-friendliness
Merged
be more memory friendly#5865edolstra merged 6 commits intoNixOS:masterfrom pennae:memory-friendliness
edolstra merged 6 commits intoNixOS:masterfrom
pennae:memory-friendliness
Conversation
andir
reviewed
Jan 5, 2022
edolstra
reviewed
Jan 5, 2022
thufschmitt
approved these changes
Jan 13, 2022
|
This pull request has been mentioned on NixOS Discourse. There might be relevant details there: |
Contributor
Author
|
@edolstra except rebasing onto current master (which we're currently doing, plus rerunning benchmarks), is there anything this needs done? the boehm guard patches should definitely go into master, the others are optional but do help performance a bit. |
Member
|
I've added it to the 2.8 milestone since it's best not to merge a few days before release :-) |
edolstra
requested changes
Mar 8, 2022
turns out it's only necessary for MANUAL_VDB, which nix doesn't use. omitting
them gives a slight performance improvement on eval.
before:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.988 s ± 0.061 s [User: 5.935 s, System: 0.845 s]
Range (min … max): 6.865 s … 7.075 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 332.6 ms ± 3.9 ms [User: 299.6 ms, System: 32.9 ms]
Range (min … max): 328.1 ms … 339.1 ms 20 runs
Benchmark 3: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.681 s ± 0.049 s [User: 2.382 s, System: 0.228 s]
Range (min … max): 2.607 s … 2.776 s 20 runs
after:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.946 s ± 0.041 s [User: 5.875 s, System: 0.835 s]
Range (min … max): 6.834 s … 7.005 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 330.3 ms ± 2.5 ms [User: 299.2 ms, System: 30.9 ms]
Range (min … max): 327.5 ms … 337.7 ms 20 runs
Benchmark 3: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.671 s ± 0.035 s [User: 2.370 s, System: 0.232 s]
Range (min … max): 2.597 s … 2.749 s 20 runs
vast majority of envs is this size.
before:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.946 s ± 0.041 s [User: 5.875 s, System: 0.835 s]
Range (min … max): 6.834 s … 7.005 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 330.3 ms ± 2.5 ms [User: 299.2 ms, System: 30.9 ms]
Range (min … max): 327.5 ms … 337.7 ms 20 runs
Benchmark 3: nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'
Time (mean ± σ): 2.671 s ± 0.035 s [User: 2.370 s, System: 0.232 s]
Range (min … max): 2.597 s … 2.749 s 20 runs
after:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.935 s ± 0.052 s [User: 5.852 s, System: 0.853 s]
Range (min … max): 6.808 s … 7.026 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 329.8 ms ± 2.7 ms [User: 299.0 ms, System: 30.8 ms]
Range (min … max): 326.6 ms … 336.5 ms 20 runs
Benchmark 3: nix flakes eval --raw --impure --file expr.nix
Time (mean ± σ): 2.655 s ± 0.038 s [User: 2.364 s, System: 0.220 s]
Range (min … max): 2.574 s … 2.737 s 20 runs
these functions are called a whole lot, and they're all comparatively small.
always inlining them gives ~0.7% performance boost on eval.
before:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.935 s ± 0.052 s [User: 5.852 s, System: 0.853 s]
Range (min … max): 6.808 s … 7.026 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 329.8 ms ± 2.7 ms [User: 299.0 ms, System: 30.8 ms]
Range (min … max): 326.6 ms … 336.5 ms 20 runs
Benchmark 3: nix flakes eval --raw --impure --file expr.nix
Time (mean ± σ): 2.655 s ± 0.038 s [User: 2.364 s, System: 0.220 s]
Range (min … max): 2.574 s … 2.737 s 20 runs
after:
Benchmark 1: nix flakes search --no-eval-cache --offline ../nixpkgs hello
Time (mean ± σ): 6.912 s ± 0.036 s [User: 5.823 s, System: 0.856 s]
Range (min … max): 6.849 s … 6.980 s 20 runs
Benchmark 2: nix flakes eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix
Time (mean ± σ): 325.1 ms ± 2.5 ms [User: 293.2 ms, System: 31.8 ms]
Range (min … max): 322.2 ms … 332.8 ms 20 runs
Benchmark 3: nix flakes eval --raw --impure --file expr.nix
Time (mean ± σ): 2.636 s ± 0.024 s [User: 2.352 s, System: 0.226 s]
Range (min … max): 2.574 s … 2.681 s 20 runs
reduces peak hep memory use on eval of our test system from 264.4MB to 242.3MB, possibly also a slight performance boost. theoretically memory use could be cut down by another eight bytes per Pos on average by turning it into a tuple containing an index into a global base position table with row and column offsets, but that doesn't seem worth the effort at this point.
no need for function<> with c++17 deduction. this saves allocations and virtual calls, but has the same semantics otherwise. not going through function has the side effect of giving compilers more insight into the cleanup code, so we need a few local warning disables.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
we do a lot of single-object allocations that needn't be done at all, are could at least be batched like the Value allocations are as of recently. singleton Envs are a good candidate since they make up the vast majority of envs allocated. we can also inline some allocation and check functions to avoid stack churn, and with memory batching get about 1.5% improvement on eval.