Scaling Ethereum to 100k TPS by Sam

Transcript

mons now from Shar many he

[Music] anticipates Prett much and that that's like one yeah so this is unlock but there are other problems that [Music] come theem this is my all the differents

what transtion for second [Music] you'll see here that between 42 so all the combin they usually sit arounds Mark that sal sal alone does 2000 between all the robots 40 of um now I just want to

make the ethereum landscape is very very focused on scaling horizontally and what this means is unlocking probility this is the IDE that can have transaction transaction between on say AR we swap on in one transaction from transaction um this is a really tough problem it's really limiting factor for at the moment it's going to get Sol

eventually but so that's why indidual is if we so perance

yes on exp have have bus intelligence this is where you can see all the users The Experience users are anytime problem you get information

that see they went de and then that downstairs was like you had want all centralized exchanges

entral they are inherently more is efficient so and then on social a big of these so more

performance but really what we're here for really want to unlock M that's really important one thing I want to highlight here is so massive nwork Builders and users kinduct is far [Music] more interesting

that perance this is reason this is this is the anatomy of the roll our best to try con this so our transaction [Music]

contion extion comp then request lo lo that's generated andow is exed and then to the chain and then where what you do need to know is we have this

is where performance this is Da avability import how many how much data we can send to whatever VA service is to UNS and the other part interesting execution so this is you have like say if you want 100K transactions in 100,000 transactions you have to on H you have in one so that's just a hard thing to do the hard this us

extion this is reading the data from Storage so so as I said just some num this is that's use the entire so that's just LS have a li of

64 this where Celestia we've got 2 [Music] we get here we get very close to 100K TPS transfers we theoretically

95k would actually 100K but I didn't have time to encode 110 lights of the transfer so [Music] compress this is might exp simply this extion happens at the moment that you just walk through it one transaction next one to the next one this is very

slow so this is where performance the idea is say got example transtion we split them up and we send them to different so sequences You2 Plus send the transactions to those sces and they can execute parall um and then you validate them good that's the of concept in thises really work you have dep so say someone

transfers one USD from me to La and then someone else and then L also SS the USD to um say down that because there's dependency there those two transactions have so that's why so actually gets interesting it's antip will this um it's this is state so how it's actually

if you have the time to look into trees or sorry sounds really complicated I but the problem with this is so how it works you have this this tree of noes and leaves that and essentially you have this to if you have that you can have confidence that all the other data in the tree is what it should be so we have

two systems they just need to share that one rout same which is um the problem is in SSD device execution the um I one thing that we're doing to make this faster is we doing smart up this tree Three Trees two of those trees are memory memory r a lot faster you access power and the stuff that's less access

less often in storage good thing about having the tree outside whens any question okay so this is

what just have each of these green ones are Lea the leaf you have SM ESS you got a whole bunch of these Tre main the top one

two and a question yeah um that's right actually the um