PHP UK Conference 2013

No puedes optimizar lo que no puedes medir

Juozas Kaziukėnas  · 


Extracto de la transcripción automática del vídeo realizada por YouTube.

let's talk is about the optimization so first of all have a very strong name very strange name so people call me Joe or Joe hawk that's a new thing that's me I kind of live in New York at the same time I don't really live in New York it's

been half a year I'm actually literally homeless so if interested in any of that you can check me out right here as a job I think I'm probably best described as a professional idiot I just usually break things and then try to fix them in the thing

i do every day is I try to grow businesses using technologies and the reason I'm giving this talk is that as you grow certain pieces of technology quite large you kind of have to measure them I was always in u.s. actually up to thursday then last week

I was driving this car does anyone know what it is yes fucking amazing like I had this fire heart I'll do it again so this is a tesla model s then the latest Deathlok are probably the only properly built electric car out there like previous is a horrible

car if you're driving a Prius you probably want to leave this room decided like I don't like please drivers this is a test it's actually like this is the performance version I drove this and it like it goes like a train it just pulls straight up

what happened recently is New York Times published this article and once we did this v road actually it's pretty shit and how we tested that is we took the car for a drive went across the East Coast in years from from New York somewhere to all way to delaware

and kind of document it that did a time log wrote about how it's actually terrible at mileage and all those things and everyone kind of said wait a second but actually test should be pretty good like i was saying Tesla's very good khaibar one York

Times is claiming that's actually bad even musk the founder and the CEO of dust obviously kind of worrying position to defend that so why it failed like why the new york time is driver when he drove that car which is very very cool and very very fast and

it lasts forever it does like figure in the miles easily when one charge why it failed the only way for them to defend their position west you base it on data it is like going back talking about card it's very common in a car industry to give specifically

prepared cars for car testers so for example manufacturers like ferrari would give you a car but we've usually 100 more horsepower than the production car and we actually would bring that car to a test track before you test it to knit properly make sure

it works on that track and i'll give it to you to test so obviously it's fake what Tesla do after the top gear review because top gear did a review a year ago of the roaster and they kind of claim it it only lasts like 50 miles after that actually

Tesla suit top gear and not actually one and kind of the BBC apologize for that but what does not do now is actually they record every single thing about the car when they give it to two people to test so what happens is after that article they be graphed

this is the distance traveled and this is the average speed and these are certain like a bunch of facts the article claimed and why they're wrong so for example one of the fact was like oh I actually drove at 45 miles an hour and that is the data says

that actually it was never 45 miles an hour it was always faster and it talks about two different different speeds another thing is the article was claiming that they would actually recharge the car a few times but the data shows that they actually never did

that we endeavour vent to a super charging station and up and reach one hundred percent of charge the only charge for a while so obviously person fishing going on there's obviously something wrong another thing is the driver actually been to a parking

lot and just did circles in it just to try to kill a battery as it shows from this like he drove like 0.6 miles at the speed of ten miles an hour you're just driving in circles so what happens is that data doesn't lie what does elysian is they were

collecting metrics every single every second about every like every part of the of the car the charge level the speed the position of the car so when New York Times claim that's actually a bad car we are like wait a second we have actual proof that it's

not true because we call me notice its data so if data says is 0 it's 0 like there isn't there is no subjective in that and people at people are people people are idiots for a lot of reasons one of them is really make subjective decisions so the guy

obviously driving the car was like well I don't really know what I'm doing maybe I don't even know what cars do and I'm just like not going to charge it and then claim my chart so I was wrong the bug in production so what the what the reason

talking about Tesla is what we were doing is with debugging the current production so all the software running a car was already in production like it wasn't a development environment but still very debugging the car they were collecting data they can

use to debug the stories so when a journalist claim something they can go back and say wait a second like you're wrong because we know it's not right so they knew you went to one of the Facebook development dogs I think it was at the PHP okay last

year and how we do things how we deploying facebook they would actually take their code once once it's ready for production and deploy to a certain list of servers first and then go obviously group eventually deploy to all of them but the point of deploying

to smaller set of servers is that actually once we deploy it they look at how the behavior changes of all sorts of different metrics for example if you deploy a certain patch to Facebook and all of a sudden people went from uploading hundred million pictures

at an hour to 1 million pictures an hour probably wear something wrong it's very likely that the behavior of your users is always consistent and as you grow to the size of facebook at that size it's it's very like it's very likely that if something

changes drastically there's something wrong what currently I mainly work with the commerce systems and we would usually get a couple thousand orders a per hour if all of a sudden my metrics say that we didn't have any orders in an hour probably we

broke something like which it doesn't just doesn't happen unless the whole us exploded which is not going to happen Russians are safe now but the problem is actually I've know it's a lot like usually when something breaks in the work the places

at work for it's usually my fault and what they notice is that a lot in a lot of cases when things break at that kind of behavioral level it's very hard to notice them deploying your coat everything works no pitch barrows no nothing what all of a sudden

[ ... ]

Nota: se han omitido las otras 3.430 palabras de la transcripción completa para cumplir con las normas de «uso razonable» de YouTube.