Fast file hashing of large files

Question

I'm using sjcl to hash files client-side so that I can check if they exist on the server before commencing a full upload.

However, it seems a bit slow. It takes about 15 seconds to hash an 8 MB file. I'm not sure if that's because the library is slow, JavaScript is slow, or the algorithm is inherently slow. It's using sha256 which is probably a bit overkill for what I need. Speed is key -- cryptographic security and collisions aren't particularly important.

Is there a faster way to do this?

$(document).on('drop', function(dropEvent) {
    dropEvent.preventDefault();
    _.each(dropEvent.originalEvent.dataTransfer.files, function(file) {
        var reader = new FileReader();
        var pos = 0;
        var startTime = +new Date();

        var hashObj = new sjcl.hash.sha256();

        reader.onprogress = function(progress) {
            var chunk = new Uint8Array(reader.result).subarray(pos, progress.loaded);
            hashObj.update(chunk);
            pos = progress.loaded;

            if(progress.lengthComputable) {
                console.log((progress.loaded/progress.total*100).toFixed(1)+'%');
            }
        };

        reader.onload = function() {
            var endTime = +new Date();
            console.log('hashed',file.name,'in',endTime-startTime,'ms');
            var chunk = new Uint8Array(reader.result, pos);
            if(chunk.length > 0) hashObj.update(chunk);
            console.log(sjcl.codec.hex.fromBits(hashObj.finalize()));
        };

        reader.readAsArrayBuffer(file);
    });
});

Edit: Just discovered SparkMD5 as per this answer. Initial tests have it running in under a second for the same 8 MB file, but it's still slower than I'd like.

@JasonLeBrun: I'm trying [xxHash](https://github.com/pierrec/js-xxhash) now. It won't take an `ArrayBuffer` as input, which might be problematic. — mpen, Jan 04 '14 at 08:49

mpen · Answer 1 · 2014-01-04T10:32:33.823

xxHash gives 32-bit hashes. It seems to be about 30% faster than SparkMD5. It, however, does not seem to work with HTML5's ArrayBuffer, so the file has to be read as text.

var blobSlice = File.prototype.slice || File.prototype.mozSlice || File.prototype.webkitSlice;
var chunkSize = 1024 * 1024 * 2;

$(document).on('drop', function (dropEvent) {
    dropEvent.preventDefault();

    _.each(dropEvent.originalEvent.dataTransfer.files, function (file) {
        var startTime = +new Date(), elapsed;
        var chunks = Math.ceil(file.size / chunkSize);
        var currentChunk = 0;
        var xxh = XXH();
        var fileReader = new FileReader();

        var readNextChunk = function() {
            var start = currentChunk * chunkSize;
            var end = Math.min(start + chunkSize, file.size);

            fileReader.readAsText(blobSlice.call(file, start, end));
        };

        fileReader.onload = function (e) {
            console.log("read chunk nr", currentChunk + 1, "of", chunks);
            xxh.update(e.target.result);
            ++currentChunk;

            if (currentChunk < chunks) {
                readNextChunk();
            } else {
                elapsed = +new Date() - startTime;
                console.info("computed hash", xxh.digest().toString(16), 'for file', file.name, 'in', elapsed, 'ms');
            }
        };

        fileReader.onerror = function () {
            console.warn("oops, something went wrong.");
        };

        readNextChunk();
    });
});

I think blobSlice will make a copy of the file, which I'm not super keen on. Nor do I particularly like treating binary data as text. I created this alternative version that works with the ArrayBuffer API by digging through the source of xxHash -- turns out only one method is missing to make HTML5's Uint8Array work like a Node.js Buffer.

/**
 * Hack to make Uint8Array work like a Node.js Buffer
 *
 * @param {Buffer} targetBuffer Buffer to copy into
 * @param {Number} targetStart Optional, Default: 0
 * @param {Number} sourceStart Optional, Default: 0
 * @param {Number} sourceEnd Optional, Default: source length
 * @see http://nodejs.org/api/buffer.html#buffer_buf_copy_targetbuffer_targetstart_sourcestart_sourceend
 * @see https://developer.mozilla.org/en-US/docs/Web/API/Uint32Array
 */
Uint8Array.prototype.copy = function(targetBuffer, targetStart, sourceStart, sourceEnd) {
    targetStart = targetStart || 0;
    sourceStart = sourceStart || 0;
    sourceEnd = sourceEnd || this.length;
    for(var i=sourceStart; i<sourceEnd; ++i) {
        targetBuffer[targetStart+i] = this[i];
    }
};

$(document).on('drop', function(dropEvent) {
    dropEvent.preventDefault();
    _.each(dropEvent.originalEvent.dataTransfer.files, function(file) {
        var reader = new FileReader();
        var pos = 0;
        var startTime = +new Date();
        var xxh = XXH();

        reader.onprogress = function(progress) {
            var length = progress.loaded - pos;
            var arr = new Uint8Array(reader.result, pos, length);
            pos += length;

            xxh.update(arr);

            if(progress.lengthComputable) {
                console.log((progress.loaded/progress.total*100).toFixed(1)+'%');
            }
        };

        reader.onload = function() {
            var arr = new Uint8Array(reader.result, pos);
            xxh.update(arr);

            var elapsed = +new Date() - startTime;
            console.info("computed hash", xxh.digest().toString(16), 'for file', file.name, 'in', elapsed, 'ms');
        };

        reader.readAsArrayBuffer(file);
    });
});

Unfortunately, they're pretty much identical in terms of speed, and it's still doing a copy. However, it this runs in about 270ms on the original 8 MB file which is much better than 15s.

score 2 · Answer 2 · answered Jan 04 '14 at 07:07

SparkMD5 is quite a bit quicker:

var blobSlice = File.prototype.slice || File.prototype.mozSlice || File.prototype.webkitSlice;
var chunkSize = 1024 * 1024 * 2;

$(document).on('drop', function (dropEvent) {
    dropEvent.preventDefault();

    _.each(dropEvent.originalEvent.dataTransfer.files, function (file) {
        var startTime = +new Date(), elapsed;
        var chunks = Math.ceil(file.size / chunkSize);
        var currentChunk = 0;
        var spark = new SparkMD5.ArrayBuffer();
        var fileReader = new FileReader();

        var readNextChunk = function() {
            var start = currentChunk * chunkSize;
            var end = Math.min(start + chunkSize, file.size);

            fileReader.readAsArrayBuffer(blobSlice.call(file, start, end));
        };

        fileReader.onload = function (e) {
            console.log("read chunk nr", currentChunk + 1, "of", chunks);
            spark.append(e.target.result);                 // append array buffer
            ++currentChunk;

            if (currentChunk < chunks) {
                readNextChunk();
            } else {
                elapsed = +new Date() - startTime;
                console.info("computed hash", spark.end(), 'for file', file.name, 'in', elapsed, 'ms'); // compute hash
            }
        };

        fileReader.onerror = function () {
            console.warn("oops, something went wrong.");
        };

        readNextChunk();
    });
});

Fast file hashing of large files

2 Answers2