package bechamel-js

  1. Overview
  2. Docs
HTML generator for bechamel's output

Install

Dune Dependency

Authors

Maintainers

Sources

bechamel-v0.1.0.tbz
sha256=d84cb7fef1bc871830154ff09055fb8ccf6304c070f591e8d7cc51f8e4ce6238
sha512=2f528fea9bdbd2016bf443c1154dd6e8375e85580f145f9ac9a07f5c7a67dfb1376823e4597706cf79732ca5f2d73d21af7289511cc61b7777df9773617cbbd8

Description

A simple tool to generate a standalone HTML page which shows results from bechamel's benchmarks.

Published: 09 Oct 2020

README

Bechamel - Agnostic benchmark tool in OCaml

Bechamel is a toolkit to do a micro-benchmarking on your functions. It able to be extended with your measures. Indeed, Bechamel provides only GC measures. Then, from your target, you can choose to use perf (with bechamel.perf) or something else available on your target.

The main purpose is to able to use Bechamel on GNU/Linux, Mac OSX, Windows and MirageOS. Monotonic clock is available in all of these platforms.

Then, we provide 2 algorithms to analyze datas:

  • OLS (Ordinary Least Square)

  • RANSAC (Random Sample Consensus)

These algorithms come from core_bench (which uses a REALTIME clock), and operf-micro (which is available only on GNU/Linux).

The main goal of bechamel (instead to be a yet another benchmark tool) is to let the end-user to manipulate measures. We provide a not-easy-to-use API which give an access to collected informations (main difference with benchmark). From it, we can produce a JSON output - and allow an other analyze with an other tool From it, we can produce a JSON output - and allow an other analyze with an other tool (main difference with core_bench).

And finally, because some counters can be available in some specific platforms (like perf), we allow the user to define their own counters and use them in bechamel. bechamel.perf is one of them (it needs a fork a ocaml-perf however - a dunification of this package).

Measure

To define your measure, you need to make a new module which respect interface Measure.UNSAFE. Main function is to blit/set your value (which can be anything) from result of your counter.

For example, perf provides some counters like Cpu_clock, then, we need to set an int64 ref to what the counter returns. We save it, we execute your function some times and ask again to get value of your counter. Finally, we do a diff between your first value and your second one and cast it to a float (to be able to analyze it).

At the end, we generate plenty of values which can be analyze by an OLS algorithm or RANSAC.

Printer

bechamel has a little notty backend to be able to print results in a fancy way. An example is available in the binary - image shows you the output.

Of course, notty is not a part of the core bechamel, so user can print results like he wants - of course, it's a little annoying work already done in bechamel.notty - but feel free to put an other backend.

JSON

Finally, at the core, bechamel has converter to JSON and can generate JSON output from collected values. From them, you can process a smarter analyze than OLS or RANSAC for example - this is the main purpose of bechamel, be able to manipulate results and check some assumptions.

Purpose

eqaf, for example, needs to check if the equality function computes arguments in a constant time - so it's not a benchmark strictly speaking but the main loop to collect results and analyze algorithms are needed to check this kind of assumption.

In other way, because eqaf want to be available in many platform, a monotonic-clock in Mac OSX, Windows and GNU/Linux is needed too.

Dependencies (7)

  1. fmt >= "0.8.4"
  2. jsonm
  3. ocplib-json-typed >= "0.7.1"
  4. rresult
  5. bechamel = version
  6. dune >= "2.0.0"
  7. ocaml >= "4.08.0"

Dev Dependencies

None

Used by

None

Conflicts

None

OCaml

Innovation. Community. Security.