Updating Gas Cost Schedule based on reproducible benchmarks by Jacek Glen | Devcon SEA

[Music] hi my name is yek and I'm here to present our project which uh looks at the gas C schedule and uh uh try to answer the question uh how accurate are actually the gas cost that we associated with uh each of the up codes so uh what we did we uh created some methodology which evaluates each op code and uh more importantly we want to share those tools so anyone can reproduce our uh uh results and then at the end we of course want to look at the existing Gasco schedule and uh update it updated so um what we did uh for each of the up code we created a series of uh bite code programs where we test op code under different conditions and uh different number of times we execute each each up code so we can see that by increasing the the time times we execute each up code the total time to execute U increases as well and then by using some statistical to tools we can figure out for each up Cod more or less what time it takes and as we can see the results are pretty consistent for each op code but Val differences between op codes which also indicates differences between um uh gas that each up coat should be associated with now the same Pro this gives us a good view of a single uh evm client the same process we replicate for uh number of clients that's uh seven in the total uh most popular that we found and uh this is where the fun begins because these results are made on different technology Stacks different uh languages uh uh and uh everything so what we want to do is to remove any noise and bring them to a common ground so we scale all the um results so they can be compared on a single graph and we can see differences and how actually uh each of the op code uh consume time and um uh efforts on the on the executing machine now what we do it's a huge data set what we do with this uh data set we want to analyze it and say where our existing Gasco schedule can be updated so we came up with two options first option option is more conservative we take the existing Gasco schedule and we try to fit the data that we got from our benchmarks into this Gasco existing Gasco schedule and only update what is necessary so in this case we can see that the Mo mode was underpriced we need to increase the price of it on the other hand exp was behave behaves much better and we can lower the gas uh associated with exper uh uh going forward uh we also measured the pre compiles in this case ecover as uh many people points out is much more expensive than the the current gas cost so that's a conservative way of updating uh the gas cost C um now the other way of doing this is more radical when we forget the current Gasco schedule and we start from clean page so we take the least expensive uh op code assign it with gas cost one and then gradually we increase gas cost as we go to the more um expensive uh up codes which is an interesting approach because it's completely new uh gasos schedule it's a massive change but there are some benefits of doing it if we do it that way we can um uh lower overall cost of each up codes because we start from one and we know that uh evm clients are getting better the machines are getting better so it does reflect the reality by lowering everything and by keeping the cost of State changes and the same level and lowering the computational cost we increase the gap which might in the future promote more effective use of the data in smart contracts um but of course does require a big change on uh evm clients um anyway uh all our methodology the tools that we use the Benchmark the decisions that we took on the way it's all explained uh in our uh repo um the EIP is coming with the uh proposal to change the gas cost schedule once I finalize it I will uh link it in Theo as well so thank you all right let's hear it from Mr Glenn ladies and gentlemen thank you so much sir appreciate it gas gas gas always the biggest conversation in the room how can we improve our gas how can we make it more efficient now we are waiting for questions to come in so if you don't mind helping me scan the QR code I don't want to block you in my hair but you can scan it and ask some questions now I do understand because the format of this is lightning it's just 5 minutes sometimes you need a bit of time to digest the content from the speaker and I I totally understand Mr Glenn I want to ask a few people were actually taking some pictures of your slides because I think they want to go home and they want to read it if they come and meet you personally later would you be able to give them a copy of the slides is that okay with you maybe if people want to understand a bit better they can ask you more of course of course of course that's what I want to hear so I know a few people were taking out their phones and snapping pictures don't worry our speakers are here because they want to share share their knowledge with you the format of this talk is a bit short yes I do understand and it can be a bit challenging but don't worry if you have any questions or you want to copy of the slides just check with our speakers they're so friendly we've got one question on the screen thank you so much let's look at it together did you find or recommend new useful op code to reduce gas costs with your experience it's a it's a vast uh topic uh whether we want to add some new up codes or uh um from the new upcodes that we already add that uh the M Copy is definitely a winner but uh whether we want to add any other new up codes uh yeah there are lots of things uh in the pipeline coming like evm Max for example and other stuff but uh having our methodology we can come and measure it correctly and assign it with some gas cost in the future oh just as you answered first question we've got a second one are the proposed new gas costs based on some scientific method or are they manually chosen definitely not manually chosen I I uh recommend you going to our Depot and look all at all the uh methodology and statistics that we used in order to come up with a with the proposal very good a science-based approach is always one of the best I'll wait another 10 seconds to see if anyone else has a question they'd like to ask and remember the only way to collect your your card to show proof that you attended the session is by scanning the QR so you have to scan the QR and then when you enter the Q&A choose collect card okay uh we've got a few more seconds I'll wait for another perhaps 10 if there any questions now again like I said earlier our speakers are very generous they're here to share their knowledge of course a format of 5 minutes is a bit tricky