Considerations about bcrypt porting to N-API

[NickNaso]

Hi, everyone!

as you know recently I worked at porting of bcrypt to new N-API. The module isn't published on npm yet, so I created a repo based on my work here: https://github.com/NickNaso/bcrypt-napi and used it in web application in production here: https://www.lexgenda.com
I wrote an article about the described activity here: How I ported bcrypt to new N-API . Another developer translated the article in Russian and you can find it here: Russian version of: How I ported bcrypt to new N-API

Performance estimation

The first activity that I did about the estimation of performance was to execute all the tests and calculate the average of the execution time. The N-API version of the addon was found to be about 2% faster.

To verify and reinforce this preliminary results I created a repo https://github.com/NickNaso/bcrypt-perf where I conducted some experiments about bcrypt performance.
At the end in the various situations the N-API version was faster than NAN version and the gain in percentage was between 0.8% - 4.19% and the gap is to high in the synchronous api.

[mhdawson]

Thanks for writing up that article, tweeted out here https://twitter.com/mhdawson1/status/948562156127055872. Can the rest of the team do the same or retweet if you have notalready.

Looks like you already have a good number of people who have read and 'clapped', good job :)

I'll be interested to hear a bit more about your perf experiments in the next N-API meeting.

[NickNaso]

Hi everyone,
today I want announce that I just finished my work on bcrypt module. In this thread I want recap all that I did to arrive at use the N-API version of the addon in production on my web application.

• Fork the original project https://github.com/kelektiv/node.bcrypt.js and refactor all the functions and methods with the API provided by the C++ Wrapper of N-API

• Execute all tests on local machine and once all of them passed successfully I created a PR and after the module passed all tests on CI my work was merged on the napi branch https://github.com/kelektiv/node.bcrypt.js/tree/napi

• Until I attempt that mantainers publish the module on npm I conducted some experiments about performance of N-API version against the previous implemented with classical V8 API using NAN. All my works and results about that are on this repo https://github.com/NickNaso/bcrypt-perf. Substantially the N-API version is little faster in avarage 2% than the NAN version. On my tests I considered the execution time of the most common operations accomplished by bcrypt (password hashing and comparing password)

• Start testing bcrypt on web application before on development environment and after on production. I created a repo that contains my work and use the fit reference on package.json of my application.

• Use published n-api tagged version of bcrypt (bcrypt@n-api) in production on web application Lexgenda
  Lexgenda is a web application where user after signup and signing a subscription can execute research over the judgments issued by italian courts, essentially it's a search engine.
  Below I report some screenshots:
  
  

[mhdawson]

@NickNaso thanks for the good summary. Great work!

[mhdawson]

This was a great FYI but I think it can now be closed. Closing, please let us know if you think that was not the right thing to do.