name: unagi-chan version: 0.4.1.4 synopsis: Fast concurrent queues with a Chan-like API, and more description: This library provides implementations of concurrent FIFO queues (for both general boxed and primitive unboxed values) that are fast, perform well under contention, and offer a Chan-like interface. The library may be of limited usefulness outside of x86 architectures where the fetch-and-add instruction is not available. . We export several variations of our design; some support additional functionality while others try for lower latency by removing features or making them more restrictive (e.g. in the @Unboxed@ variants). . - @Unagi@: a general-purpose near drop-in replacement for @Chan@. . - @Unagi.Unboxed@: like @Unagi@ but specialized for primitive types; this may perform better if a queue grows very large. . - @Unagi.Bounded@: a bounded variant with blocking and non-blocking writes, and other functionality where a notion of the queue's capacity is required. . - @Unagi.NoBlocking@: lowest latency implementations for when blocking reads aren't required. . - @Unagi.NoBlocking.Unboxed@: like @Unagi.NoBlocking@ but for primitive types. . Some of these may be deprecated in the future if they are found to provide little performance benefit, or no unique features; you should benchmark and experiment with them for your use cases, and please submit pull requests for additions to the benchmark suite that reflect what you find. . Here is an example benchmark measuring the time taken to concurrently write and read 100,000 messages, with work divided amongst increasing number of readers and writers, comparing against the top-performing queues in the standard libraries. The inset graph shows a zoomed-in view on the implementations here. . <> . license: BSD3 license-file: LICENSE author: Brandon Simmons maintainer: brandon.m.simmons@gmail.com category: Concurrency build-type: Simple cabal-version: >=1.10 -- currently uploaded to imgur; move to this eventually --extra-doc-files: images/*.png --cabal-version: >=1.18 extra-source-files: CHANGELOG.markdown Tested-With: GHC ==7.8.4 || ==7.10.3 || ==8.0.2 || ==8.2.2 || ==8.4.4 || ==8.6.4 || ==8.8.1 || ==8.10.7 source-repository head type: git location: https://github.com/jberryman/unagi-chan.git branch: master library hs-source-dirs: src exposed-modules: Control.Concurrent.Chan.Unagi , Control.Concurrent.Chan.Unagi.Unboxed , Control.Concurrent.Chan.Unagi.Bounded , Control.Concurrent.Chan.Unagi.NoBlocking , Control.Concurrent.Chan.Unagi.NoBlocking.Unboxed other-modules: Control.Concurrent.Chan.Unagi.Internal , Control.Concurrent.Chan.Unagi.Unboxed.Internal , Control.Concurrent.Chan.Unagi.Bounded.Internal , Control.Concurrent.Chan.Unagi.NoBlocking.Internal , Control.Concurrent.Chan.Unagi.NoBlocking.Types , Control.Concurrent.Chan.Unagi.NoBlocking.Unboxed.Internal , Control.Concurrent.Chan.Unagi.Constants , Utilities , Data.Atomics.Counter.Fat ghc-options: -Wall -funbox-strict-fields build-depends: base >= 4.7 && < 5 , atomic-primops >= 0.8 , primitive>=0.5.3 , ghc-prim default-language: Haskell2010 if !arch(i386) && !arch(x86_64) && !arch(aarch64) cpp-options: -DNOT_optimised -- TODO: more complete list of 64-bit archs: if arch(x86_64) || arch(aarch64) cpp-options: -DIS_64_BIT -- tryReadMVar is only available and non-broken on ghc >= 7.8.3 if impl(ghc >= 7.8.3) cpp-options: -DTRYREADMVAR -- TODO -- For v0,4: -- - More benchmarks, and test code we can analyze with ghc-events-analyze. -- - Explore faster single-threaded write (see #11) -- - Explore Stream interface for variants other than NoBlocking (see #11) -- - Experiments w/ new GHC 7.10 stuff, and at least make sure buildable -- ------- -- - For GHC 7.10+ -- - look at small arrays (w/out card-marking) -- - re-benchmark array creation and adjust next segment wait -- - Do a benchmark of multiple queues running in parallel, to see if we are -- affected by global allocator issues with pinned memory: -- http://thread.gmane.org/gmane.comp.lang.haskell.parallel/218 -- -- Possibly-similar prior work to look at: -- -- - maybe implement "Fast Concurrent Queues for x86 Processors" by Morrison & Afek (non-blocking, probably more clever) -- - Also looks like a similar (but lockfree, as above) counter-based queue has been developed by FB: -- https://github.com/facebook/folly/blob/master/folly/MPMCQueue.h -- Please just build tests and run: -- $ time ./dist/build/test/test -- Doing `cabal test` takes forever for some reason. test-suite test type: exitcode-stdio-1.0 ghc-options: -Wall -funbox-strict-fields ghc-options: -O2 -rtsopts -threaded -- NOTE: configure --enable-profiling overrides this: ghc-options: -with-rtsopts=-N ghc-options: -fno-ignore-asserts -- for some hacks for Addr: ghc-options: -fno-warn-orphans ghc-options: -fno-warn-missing-methods -- I guess we need to put 'src' here to get access to Internal modules hs-source-dirs: tests, src main-is: Main.hs other-modules: Atomics , Deadlocks , DupChan , Implementations , IndexedMVar , Smoke , Unagi , UnagiUnboxed , UnagiBounded , UnagiNoBlocking , UnagiNoBlockingUnboxed build-depends: base , primitive>=0.5.3 , atomic-primops >= 0.8 , containers , ghc-prim default-language: Haskell2010 -- These have to be copied from 'library' section too! if !arch(i386) && !arch(x86_64) && !arch(aarch64) cpp-options: -DNOT_optimised if arch(x86_64) || arch(aarch64) cpp-options: -DIS_64_BIT if impl(ghc >= 7.8.3) cpp-options: -DTRYREADMVAR -- compare benchmarks with Chan, TQueue, and (eventually) lockfree-queue? flag compare-benchmarks default: False manual: True benchmark single type: exitcode-stdio-1.0 ghc-options: -Wall -O2 -threaded -funbox-strict-fields -fforce-recomp -rtsopts hs-source-dirs: benchmarks default-language: Haskell2010 default-extensions: CPP build-depends: base , unagi-chan , criterion if flag(compare-benchmarks) cpp-options: -DCOMPARE_BENCHMARKS build-depends: stm -- , lockfree-queue main-is: single.hs ghc-options: -with-rtsopts=-N1 -- To run comparison benchmark used in graph above, run: -- $ cabal configure --enable-benchmarks -fcompare-benchmarks -- $ cabal bench multi --benchmark-option=-omulti3.html --benchmark-option='Demo' benchmark multi type: exitcode-stdio-1.0 ghc-options: -Wall -O2 -threaded -funbox-strict-fields -fforce-recomp -rtsopts hs-source-dirs: benchmarks default-language: Haskell2010 default-extensions: CPP build-depends: base , unagi-chan , criterion if flag(compare-benchmarks) cpp-options: -DCOMPARE_BENCHMARKS build-depends: stm -- , lockfree-queue main-is: multi.hs ghc-options: -with-rtsopts=-N build-depends: async -- flag dev -- default: False -- manual: True -- for profiling, checking out core, etc -- executable dev-example -- -- for n in `find dist/build/dev-example/dev-example-tmp -name '*dump-simpl'`; do cp $n "core-example/$(basename $n).$(git rev-parse --abbrev-ref HEAD)"; done -- if !flag(dev) -- buildable: False -- else -- build-depends: -- base -- , stm -- , unagi-chan -- -- ghc-options: -ddump-to-file -ddump-simpl -dsuppress-module-prefixes -dsuppress-uniques -ddump-core-stats -ddump-inlinings -- ghc-options: -O2 -rtsopts -- -- -- Either do threaded for eventlogging and simple timing... -- ghc-options: -threaded -eventlog -- -- and run e.g. with +RTS -N -l -- -- -- ...or do non-threaded runtime -- --ghc-prof-options: -fprof-auto -- --Relevant profiling RTS settings: -xt -- -- TODO also check out +RTS -A10m, and look at output of -sstderr -- -- hs-source-dirs: core-example -- main-is: Main.hs -- default-language: Haskell2010