The unagi-bloomfilter package
This library implements a fast concurrent bloom filter, based on bloom-1 from Fast Bloom Filters and Their Generalization by Y Qiao, et al.
A bloom filter is a probabilistic, constant-space, set-like data structure supporting insertion and membership queries. This implementation is backed by SipHash so can safely consume untrusted inputs.
The implementation here compares favorably with traditional set implementations in a single-threaded context, e.g. here are 10 inserts or lookups compared across some sets of different sizes:
With the llvm backend benchmarks take around 75-85% of the runtime of the native code gen.
Unfortunately writes in particular don't seem to scale currently; i.e. distributing writes across multiple threads may be slower than in a single-threaded context, because of memory effects. We plan to export functionality that would support using the filter here in a concurrent context with better memory behavior (e.g. a server that shards to a thread-pool which handles only a portion of the bloom array).
|Dependencies||atomic-primops (>=0.8), base (>=4.7 && <5), bytestring, hashabler (>=1.3.0), primitive [details]|
|Source repository||head: git clone https://github.com/jberryman/unagi-bloomfilter.git|
|Uploaded||Mon Sep 5 18:03:47 UTC 2016 by BrandonSimmons|
|Downloads||62 total (11 in the last 30 days)|
|Status||Docs available [build log]
Last success reported on 2016-09-05 [all 1 reports]
|dev||To build tests, executables and benchmarks do `configure -fdev --enable-tests` and run the built executables by hand (i.e. not with `cabal test` etc.; we put all our different executables in test-suite sections in order to hide their dependencies from hackage)||Disabled||Manual|
|instrumented||Enables assertions in library code. When --enable-library-profiling and --enable-executable-profiling is turned on, you can get stacktraces as well||Disabled||Manual|
Use -f <flag> to enable a flag, or -f -<flag> to disable that flag. More info
For package maintainers and hackage trustees