Performance Optimisations #19

AndriusA · 2019-05-20T21:15:43Z

Optimisations to get in under the performance budget:

implements optimisation that merges together equivalent rules with different domain options
memory copying/allocation optimisations across request parsing and tokenisation

Also removes full-string domain option matching from compile-time option-enabled features

squeezes in under the performance target by ~5%


        adds filter optimisation merging together rules that only differ by d…

…omain options


        moves to a different hashing algorithm compatible with wasm


        removes full domain matching option for clearer code


        Updates regex-based URL parser

Implementation left for reference, however not enabled as it is the slower option


        clippy recommended code cleanups


        Streamlines request parser a little more

Gains a bit more on performance (~10% on request parsing)


        Memory-focused optimisations that get us under the performance target

Overall ~2-3 us saved, now 0.009us through Node.js wrapper


        reduces occurrences of `unwrap` to avoid any panics

Also removes most `[#inline]` annotations to an apparent gain/no loss in performance


        doing away with thread-safe structures for Regex caching for speed

The RwLock gets hit for most rules and requests, making it relatively costly. Going to simpler ones saves another few percent relative to target. Downside is non-parallelised rule parsing from text lists

AndriusA added 7 commits May 17, 2019

adds filter optimisation merging together rules that only differ by d…

5bf11da

…omain options

moves to a different hashing algorithm compatible with wasm

e91b129

removes full domain matching option for clearer code

b24259d

Updates regex-based URL parser

48d43fd

Implementation left for reference, however not enabled as it is the slower option

clippy recommended code cleanups

9f1d8cb

Streamlines request parser a little more

ce424c1

Gains a bit more on performance (~10% on request parsing)

Memory-focused optimisations that get us under the performance target

814ac87

Overall ~2-3 us saved, now 0.009us through Node.js wrapper

AndriusA requested a review from bbondy May 20, 2019

AndriusA added 2 commits May 21, 2019

reduces occurrences of `unwrap` to avoid any panics

9d9ee67

Also removes most `[#inline]` annotations to an apparent gain/no loss in performance

doing away with thread-safe structures for Regex caching for speed

19ac71a

The RwLock gets hit for most rules and requests, making it relatively costly. Going to simpler ones saves another few percent relative to target. Downside is non-parallelised rule parsing from text lists

bbondy approved these changes May 22, 2019

View changes

bbondy merged commit 3e84575 into master May 22, 2019

AndriusA deleted the optimisations branch Jun 6, 2019

brave / adblock-rust

Performance Optimisations #19

Performance Optimisations #19

AndriusA commented May 20, 2019

brave / adblock-rust

Join GitHub today

Performance Optimisations #19

Performance Optimisations #19

Conversation

AndriusA commented May 20, 2019