https://github.com/raftlib/benchmarks

Repository of raftlib benchmarks
https://github.com/raftlib/benchmarks

Last synced: about 1 month ago
JSON representation

Repository of raftlib benchmarks

• Host: GitHub
• URL: https://github.com/raftlib/benchmarks
• Owner: RaftLib
• Created: 2020-05-10T00:25:24.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2020-10-27T14:04:24.000Z (over 5 years ago)
• Last Synced: 2025-03-11T17:49:07.367Z (about 1 year ago)
• Language: C++
• Homepage: https://www.raftlib.io
• Size: 124 MB
• Stars: 4
• Watchers: 4
• Forks: 2
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Notes

* Most of the benchmarks are derived from other sources,

where this is so we've listed them as such to give credit

to the original source. 

# benchmarks

* parallel-bzip

    * Original open source code is contained within this folder

    along with respective license

    * RaftLib version of bzip compress is also included

    * Datasets: 

        * large - (1<<30)

        * medium - (1 << 20)

        * small  - (1 << 10)

* streamcluster - part of the PARSEC benchmark suite, original

papers and information is available [here](https://parsec.cs.princeton.edu)

    * Still in progress (as in not finished yet for the RL version)

    * TBB and Pthreads versions exist and are included as well. 

* dedup - todo

    * designed to be pipelined so should be easy 

* ferret - todo

    * designed to be pipelined so should be easy 

* x264 - todo 

    * likely pretty easy to parallelize

* fluidanimate

* stringmatch

    * not sure if this is the best benchmark given

    it almost always ends up being RAM speed limited

    in the RL case vs. coompute limited...but, maybe

    not a bad idea to add. 

# Porting notes

## M:N queue specifics 

* For specifying M:1 queue behavior, use 

the reduce kernel, the runtime can remove

this one and insert a M:1 capable Queue

* For 1:N queue behavior, use a split kernel

which the runtime will remove and insert a

1:N capable queue. 

* For M:N behavior use a mux kernel which 

takes M inputs, distributes them to N outputs. 

The runtime will take this kernel and replace

it with an M:N capable FIFO when available.