This Lesson Taught Me How To Do Better Benchmarks

Key Takeaways

left pad an 11 line piece of code broke the internet and just recently had a second incident that involves me so I might have made a mistake where I thought I could program a faster version of left pad live and the results uh were not faster it goes to here left pad with this one okay so it looks like it's looked like mine is slower you hate to see it at first it inspired a meme then Ember hex made an actually fast version then the creator of crab laying through his hat in the ring even faster than Embers it might is still so slow a Renaissance of left pad happened and other people were joining in and then a competition was born who had the fastest left pad submissions were made more submissions were made then corrected and meanwhile I was hating life on YouTube I've officially cured of imposter syndrome too spicy it even stack Overflow had their own bespoke solution and it was recursive not only that but there's also a native left pad solution provided in modern JavaScript and we need to figure out which left pad is the fastest and find out do micro benchmarks lie yes they do so this competition has been about micro benchmarks but here's the thing is that some of you might not even know what micro benchmarks are well they're as simple as this you write a function you wish to test you've run it a bunch of times taking a bunch of time differentials and you see which function performs better than the other you have different implementations and you hope one of them is going to perform way faster then you've run it a whole bunch and you see that some of them are much slower say mine or the stack Overflow one and some of them are way faster like Embers or the native one or the special case one that I also have in a right and then you call that a day right you say hey my code is fast well here's the deal it's actually not fast you actually don't even know if they're fast in fact you just might be lied to but why well first off you're probably running them on your computer which if I just go P grep node I don't even know dude I must have a Bitcoin miner on my machine I do have Zig running hey calm down ladies and our choosers okay I'm taken tests also often hide implementation details such as garbage collection when you run these micro benchmarks they don't necessarily trigger all the things that could happen when you're running in production and finally it depends on what you're actually doing in your function the thing about JavaScript is that strings aren't strings they're actually rope data structures and there's a lot of optimizations around concatenation and depending with a heavier penalization of iterating over each character so that means if we're just appending a bunch of strings together it may look good on paper but when you actually use the string to do something say send it across the internet or print it out it has to iterate over a more complicated structure to get that string back out so you pay none of the cost of using your string you only pay the cost of producing it which could be cheaper thus giving you misleading results so how do we determine which left pad is actually faster well first I created a server set up Apache benchmark script and ensure proper sleep between each and manually execute garbage collection on the server if no requests have happened within one second send 50 million requests that took about 36 hours to complete I'd personally like to thank lenode for giving me an instance to run this on and not shutting me down after 50 million requests thanks Andrew create a parser to parse the Apache Benchmark results and calculate the average of medians import results into Google spreadsheets charts before we dive into the charts let me kind of explain what the data is you're looking at so each one of these points is the average of the median so what I did is I do like twelve and a half thousand requests and I'd take the median time of that and then I do it again and I do it again and I do it again and after a bunch of those requests I'd have a several amount of medians and then I take the average of those so each one of these represents somewhere between 50 to 100 000 requests each and then of course down here I just simply summed all those numbers across every size of requests and then I ordered these columns according to who had the largest sum all right so this chart is probably the most important chart to look at is the summed of the median averages on the y-axis we have milliseconds and on the x-axis we have the different left pads summed averages of the medians so worst place came in as the stack Overflow tail recursive version which I'm a bit surprised by then came left pad commanding a strong second worst place then look at this I'm talking about you Tom it hurt but the fastest really came down to left pad Travi and Native they're really within a margin of error of each other so what is the takeaway of this the takeaway is use native methods they're way better you don't have to be clever you don't have the right code that looks like that whatever what is this and the best part is is that as new techniques come out as things happen over time you don't have to make these weird changes to your code where you did these weird incantations to figure out what codes actually faster than the other code instead you know the platform is motivated to make a fast product so stick with syntax or vendor implemented options where possible and you will mostly get the best performance but if you ever feel spicy and you want to do power of two string concatenation and bit shifting well go ahead give it a try which is funny because that's actually how string dot repeat works is power of two string concatenation and bit shifting in V8 I don't I have no idea why that's fast I literally have no idea hey if you like this press the like button press subscribe come on give me some signals okay if you want more of this send me the signals make comments do all the stupid things you know you're supposed to do already okay because YouTube highly favors that so if you don't do it it makes me feel like this is a failure and then I feel like a failure and then my wife thinks I'm a failure don't do it to the wife of Jen this is for you Travi