Benchmark: Async vs Neo-Async
What is Neo-Async?
Neo-Async is meant to be used as a drop-in replacement for the Async library. It implements the same API, and adds extra functionality. I have been trying to make it faster and safer than Async.
https://github.com/suguru03/neo-async
It doesn't mean I don't like Async. I would like to improve both libraries and to showcase some performance optimizations that might be implemented by other Node.js libraries. I especially admire @aearly for their good work.
Benchmark
Recently, Async has been released v2.0.0-rc.6 and I released Neo-Async v2.0.0-rc.1.
So I would like to check performance using benchmark and func-comparator.
How to check
I made a repository to measure performance between Async and Neo-Async.
https://github.com/suguru03/async-benchmark
$ git clone git@github.com:suguru03/async-benchmark.git $ cd async-benchmark $ npm install $ node . // It might take more than one hour...
Environment
- Ubuntu v12.04
- Node.js v6.2.1
- async v2.0.0-rc.6
- neo-async v2.0.0-rc.1
- benchmark v2.1.0
- func-comparator v0.7.1
Result
Neo-Async is 1.27 ~ 10.7 times faster than Async.
The value is the ratio (Neo-Async/Async) of the average speed.
Collections
| function | benchmark | func-comparator |
|---|---|---|
| each | 3.71 | 2.54 |
| eachSeries | 2.14 | 1.90 |
| eachLimit | 2.14 | 1.88 |
| eachOf | 3.30 | 2.50 |
| eachOfSeries | 1.97 | 1.83 |
| eachOfLimit | 2.02 | 1.80 |
| map | 4.20 | 4.11 |
| mapSeries | 2.40 | 3.65 |
| mapLimit | 2.64 | 2.66 |
| mapValues | 5.71 | 5.32 |
| mapValuesSeries | 3.82 | 3.23 |
| mapValuesLimit | 3.10 | 2.38 |
| filter | 8.11 | 8.76 |
| filterSeries | 5.79 | 4.86 |
| filterLimit | 4.00 | 3.32 |
| reject | 9.47 | 9.52 |
| rejectSeries | 7.39 | 4.64 |
| rejectLimit | 4.54 | 3.49 |
| detect | 6.67 | 6.37 |
| detectSeries | 3.54 | 3.73 |
| detectLimit | 2.38 | 2.62 |
| reduce | 4.13 | 3.23 |
| reduceRight | 4.23 | 3.24 |
| transform | 5.30 | 5.17 |
| sortBy | 2.24 | 2.37 |
| some | 6.39 | 6.10 |
| someSeries | 5.37 | 4.66 |
| someLimit | 3.39 | 2.84 |
| every | 6.85 | 6.27 |
| everySeries | 4.53 | 3.90 |
| everyLimit | 3.36 | 2.75 |
| concat | 9.18 | 9.35 |
| concatSeries | 7.49 | 6.09 |
Control Flow
| funciton | benchmark | func-comparator |
|---|---|---|
| parallel | 7.54 | 5.45 |
| series | 3.29 | 2.41 |
| waterfall | 5.12 | 4.27 |
| whilst | 1.96 | 1.95 |
| doWhilst | 2.07 | 1.96 |
| until | 2.10 | 1.99 |
| doUntil | 1.98 | 2.04 |
| during | 10.7 | 7.09 |
| doDuring | 5.98 | 6.03 |
| queue | 1.83 | 1.75 |
| priorityQueue | 1.79 | 1.75 |
| times | 3.84 | 3.65 |
| race | 1.45 | 1.27 |
| auto | 3.23 | 3.50 |
| retry | 9.43 | 6.78 |
Future Prospects
I'm not done with performance optimization; the control flow functions are my main focus right now. I will release v2.0.0 when satisfied with performance.
And I'm keeping in mind this issue.
I believe the asyncFlowF library doesn't implement the same interface as Async, but I would like Neo-Async to be beat it benchmark anyways. ;)
Even as many JavaScript users switch to Promises, callback style is far from dead, so I want to keep improving Async and Neo-Async.
Next time, I will write about the difference between Async and Neo-Async. :)
