Node-crc32c

Basic C modules for NodeJS with crc32c (Castagnoli) implementation for Linux. The implementation uses the native Linux library "AF_ALG". It is compatible with node 0.10 and 0.11!

It supports strings, string objects, buffers, numbers! Works well with mongoose. Just do a toString on the entity to hash!

This module is not meant for secure hashing but really for something like ETags or anything that is easier to compare using a hash than the full string.

##Building

make build

or

npm install crc32c

##Usage

For small number of computation needed

You have the function compute, which takes only one argument, the string to hash.

From JavaScript:

var crc32c = require('crc32c');

// Works with strings!
var toHash = "HELLOWORLD";
console.log(crc32c.compute(toHash));

// Or buffers!
console.log(crc32c.compute(new Buffer('{"jsonString":"property1","jsonArray":["arr1","arr2","arr3"]}')));

// Or String Objects!
console.log(crc32c.compute(new String('A superb string!')));

// It also supports numbers, if you really need to compute on a single integer/float!
console.log(crc32c.compute(3.141592654);

For batch computing

With >100 iterations I get a 3x to 5x performance improvements. It really shows up at more than 10K iterations though.

You have to create an object called a Batcher. This object then has 3 methods: openSocket, closeSocket, and compute.

From JavaScript:

var crc32c = require('crc32c');
var Batcher = new crc32c.Batcher(); // You can create as many as you want. Every instance will use a single socket.
var Batcher.openSocket();
console.log(Batcher.compute(new Buffer('{"jsonString":"property1","jsonArray":["arr1","arr2","arr3"]}')));
// ... Iterate on many strings/buffer/etc.
Batcher.closeSocket();

From cli:

crc32c <filename>

It currently supports only one file at the time.

Want more examples?

SOON See the example sections!

##License

The plugin is under MIT license, please see the LICENSE file provided with the module.

Tests

You can run the test by doing make test. Currently the test only contains successful use case, but error handling test cases will be added soon.

Benchmarks

Run the script by doing make benchmarks!

I think that pure times are not representative of reality, since every setup will get different results. This is why I've put the times in ratio using AF_ALG batch as the base (1).

• AF_ALG batch: This test has used this library in single socket (batch) mode.
• AF_ALG std: This test has used this library in multi socket (standard) mode.
• SSE4.2: Using the SSE4.2 Implementation by Voxer, which is using x86 asm implementation.
• Pure JS (table): This is using the pure JS CRC32 implementation, using a pre-baked table.
• Pure JS (direct): This is using the pure JS CRC32 implementation, without using a pre-calculated table.

For the original times see benchmarks/results.txt

| Test | AF_ALG batch | AF_ALG std | SSE4.2 | Pure JS (table) | Pure JS (direct) | | ---- |:------------:|:----------:|:------:|:---------------:|:----------------:| | TEST_STRING_1024 | 1 | 4.4 | 0.3 | 60.4 | 97.4 | | TEST_STRING_2048 | 1 | 4.6 | 0.4 | 119.8 | 186.9 | | TEST_BUFFER_1024 | 1 | 5.7 | 0.4 | 8.9 | 45.5 | | TEST_BUFFER_2048 | 1 | 5.2 | 0.5 | 95.0 | 86.5 | | TEST_STRING_OBJECT_1024 | 1 | 3.5 | N/A | 50.6 | 115.8 | | TEST_STRING_OBJECT_2048 | 1 | 3.3 | N/A | 103.9 | 208.6 |

N/A means that it is not available because not supported.

Interesting things

• The library closely always takes the same times to execute even though the data is bigger. The main bottleneck is the data unboxing.
• The single-socket can improve the performance by a lot when needing to do a lot of calculations
• Yes the assembly implementation is faster, and that is not really surprising. Why taking crc32c? Because it's a cross platform implementation, that uses a core linux library (so very robust implementation). Also, this library support string objects.
• The pure JS library is waaaaaaaaaay slower with 1024 bytes strings and gets slower and slower when the string is bigger. The C implementation is pretty stable, and the only thing slowing it in the unboxing from JavaScript to pure C.