|
Title: [PATCH] jsMiner: perf improvement Post by: mrb on May 30, 2011, 10:15:32 PM I spent 15min, for fun, looking at possible optimizations for jsMiner (https://github.com/jwhitehorn/jsMiner). My patch simply inlines some functions, pre-computes constants, and caches array values, and ends up increasing its performance by ~1.7x under V8 (benchmarked within nodejs 0.1.97) on my Core i3-i380UM (1.33GHz), from 8.3 kHash/s to 13.9 kHsh/s.
This means jsMiner is "only" between 1/38th and 1/59th the performance of cpuminer 1.0.1: * 'cryptopp' algo: 525 kHash/s * '4way' algo: 825 kHash/s See http://pastie.org/1995279 Title: Re: [PATCH] jsMiner: perf improvement Post by: 1bitc0inplz on May 30, 2011, 11:07:25 PM Wow!
Thank you. I have not yet looked at your patch, but that is awesome that you were able to increase the performance like that. I will look over this patch and post back up... But, I did want to post sooner rather than later to let you know that I saw your message. Thanks again. Title: Re: [PATCH] jsMiner: perf improvement Post by: AngelusWebDesign on May 30, 2011, 11:37:35 PM 1bitcoinplz:
Have you looked into various techniques for Javascript optimization? I did a lot of that a few years ago when I wrote a Warcraft II clone that plays in the browser. I wanted to squeeze every CPU cycle I could out of the Javascript engines/CPUs of the time. I read up on various things you can do -- I think it helped. You can even "compile" your code using a PHP/C/VB.net program to replace constants with actual numbers. That way, your code is readable/understandable, yet you're not using "variables" when you don't have to. I wrote my own "obfuscator" as well, which reduced the file size and made it hard to follow. Title: Re: [PATCH] jsMiner: perf improvement Post by: 1bitc0inplz on May 31, 2011, 06:04:58 AM I've been looking over the patch. It looks like you found some performance tweaks in the SHA implementation. That is, one area of our code where we just took a literal approach to the algorithm. So, I cannot say I'm surprised it can be cleaned up a bit.
mrb, did you want to setup a pull request on github? I'd like to see about incorporating your changes. AngelusWebDesign, I was under the impression that browsers like Chrome and Safari did JIT compilation of Javascript... I wonder if inlining constants has as much an effect in those browsers? Either way, I have not looked too much into such techniques, though I should. On a related note, we have been doing a lot of the backend stuff with node.js, and I must say... the V8 runtime & node.js are purely amazing! The turn around time on HTTP requests is super low, even with high volume. Having used ASP.NET and Rails on numerous other projects I think it'd be safe to say the CPU time and memory footprint of node.js is half of what they normally get. Title: Re: [PATCH] jsMiner: perf improvement Post by: mrb on May 31, 2011, 06:40:26 AM jsMiner actually performs even faster in Chrome than in node.js!
I am not suprised either by the performance gain I got. Some people in #bitcoin-dev wrongly claimed it was already optimized, when I knew it was not. Don't have time to set up a pull req., sorry. Title: Re: [PATCH] jsMiner: perf improvement Post by: Stefan Thomas on July 01, 2011, 01:20:46 PM Would this sha256 function be safe for us to use inside Webcoin? Or is it "mining-specific" in some fashion? I can't see how it would be, just making sure.
Also, do you want to have a go at optimizing Webcoin's JavaScript ECDSA stuff (https://github.com/bitcoinjs/bitcoinjs-lib/blob/master/ecdsa.js) as well? It's the bottleneck in Webcoin at the moment. Any speedup to the elliptic curve point multiplication or big integer math would be a godsend. |