Ethereum Cat Herders PEEPanEIP, 15 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: PEEPanEIP#84: Akula and TheMerge with Artem Vorotnikov

Description

Topics covered
----------------------------
3:15 - At which point the Erigon team decided to move to a full client implementation?
7:06 - Akula different from Erigon & Silkworm
10:35 - Can you talk a little bit about a modular client design? Why was selected for the full client development of Akula.
16:00 - Database, Sync time and other advantages
18:40 - Akula EVM and EVMODIN
21:08 - Is Akula ready for users?
22:35 - Particular user target base
23:15 - Plan of adding mining feature
24:15 - Akula in Merge devnet/event
27:18 - Sepolia post-merge upgrade
28:10 - Best CL client combo?
30:10 - Future of Akula - Client Diversity
31:45 - Best place for resources
32:38 - What Telegram?
34:55 - Any message for the community?

Resources:
-----------------
https://akula.app/

TheMerge playlist - https://www.youtube.com/playlist?list=PL4cwHXAawZxqoLxXqZqT4hcYhoHoP6w12

KnowYourClient - https://www.youtube.com/playlist?list=PL4cwHXAawZxoruie1hmYWJfiakpYivXft

PEEPanEIP playlist - https://www.youtube.com/playlist?list=PL4cwHXAawZxqu0PKKyMzG_3BJV_xZTi1F

Check out upcoming EIPs in Peep an EIP series at https://github.com/ethereum-cat-herders/PM/projects/2

Follow at Twitter
-----------------------------------------------
Artem Vorotnikov @vorot93 | Pooja Ranjan @poojaranjan19

Check out upcoming EIPs in Peep an EIP series at https://github.com/ethereum-cat-herde...

Contact Ethereum Cat Herders
---------------------------------------------------
Discord: https://discord.io/ethereumcatherders
Twitter: https://twitter.com/EthCatHerders
Medium: https://medium.com/ethereum-cat-herders
GitHub: https://github.com/ethereum-cat-herde...
Website: https://www.ethereumcatherders.com/

A

Ethereum blockchain is close to the merge upgrade introduction of a new client at this point is a good step towards adding client diversity to learn about the new ethereum client implementation for execution layer continue watching.

A

Welcome to p paneer episode 84. I am puchar ranjan back with another episode on new york line series.

A

Today we are joined by ethereum client developer, arthion wattenikov, who is currently working on development of akula, a project by the aragon team about the client after the official website description written in rust, akula is a next generation implementation of ethereum protocol. We will learn more about akula in this episode with arthur. Welcome to the show atu.

B

A

Thank you for joining and because this is your first time joining us on peep and eep show. May we start with a brief introduction and how do you find your way to the ethereum community.

B

Sure thank you puja for having me today, and my name is artem. I've been in the ethereum core development community for around two and a half years at this point previously starting january 2020 and until august 2020 I was the lead developer of open ethereum, basically maintaining it and keeping it alive after parity handed off the project to gnosis.

B

I was a part of doses back then, and I is coordinated, led the development of open ethereum and, as I said, made sure that it stayed with sync with the rest of the network and we supported our users who relied on exclusive at the time. Parity features like tracing api, which is by the way available both in aragon and akula, but we'll come to that later.

B

uh After august 2020, I left open ethereum to join aragon. Since the time I have been on the roster as the evergone core developer. Basically, I spent several months significantly. Improving aragorn and participating in the research uh and aragon is basically the premier platform for research into ethereum nodes into ethereum protocol potential protocol improvements.

B

And after that I decided to basically maybe do my own thing and write in the language that uh that I want to write in rust. Eragon has written go while akula takes all the best things from aragon in terms of features and in terms of innovations and just implements them in rust, which is the shiny, new and very fast programming language with us. Since 2015 open ethereum was written in it and now akula is the rust implementation of the ethereum protocol.

A

That is very interesting. Thank you for sharing this, so we will get into details of how akula was developed and maybe an interesting feature about it. But let's start with a little bit of link motivation, I read somewhere that akula was originally conceived as a helper library for the edibon database.

A

So at what point the team decided or you decided- to move to a full client implementation.

B

This is actually a very good question and kind of shows uh where roads can lead you to like the the dark depths of the core development.

B

The thing is that, uh basically, over the past two years, I slowly built up the required the new required components for building the new implementation. So originally I built century it was like known at the time like down century is just part of akula, but at that time there was an independent project called ethereum century, which was about making a peer-to-peer node, which would connect to the ethereum peer-to-peer network and just listen to it like get incoming blocks incoming transactions and just provide users with an option of what to do with it.

B

So we can do any block processing, it wouldn't build any state, it's just a appear and that's it, but uh it was there. It was developed, it works uh and uh after that, for my personal reasons, I decided like why not make a library to access aragon database, especially since I was working on aragon, so that library was called heathdb uh rust, ethereum db, like that's the the precise repo address on github to basically uh access uh aragorn's, mdbx database and read everything that ergon stored, like all the what two terabytes of data.

B

But after some time after like I had uh etp, I understood that if you implement the data access layer, you basically implement the storage model, and this is like the arrogance key innovation. This is the changed storage model, which is a lot more effective than than the way that the classic way nodes are built, combine that with the php layer and build an execution on top of it, and you already have uh you could have an mvp for the node.

B

So at that point I decided like why not make my own note like it seemed crazy at the time like like no it's making your own cloud ethereum client, it's just so much work like like parity ethereum. It's like what was like worth. 200 000 lines of code, it's huge or like look at go ethereum. It's also many many many thousands of lines of code but like why not maybe we could try that it's also research like it's. It fits the aragon's project purpose very well. It's also research.

B

It improves in some ways upon aragon in different prospects and also it provides the client diversity, the venerable client diversity that is required for for ethereum to if you want to advance any kind of innovation.

B

So at that time I decided to just try it and after one and a half years of work here we are, it works it syncs. It's a viable full node implementation.

A

That is interesting, of course, after parity or the open ethereum there was a big gap and decline. Diversity is a big challenge at present for the execution layer, especially so having an addition of new client gives option to users to maybe try something new, a question that is like related to the arigon team itself. I'm aware that ericon is also having a c plus plus implementation uh known as silkworm, so how is akula different from aragon and silkworm.

B

So uh basically, just like silkworm akula is built from fresh from ground up. It's not a fork of any existing uh node implementation. It is uh not. It doesn't have any strings attached both in terms of force and also in terms of licensing, so it doesn't didn't inherit any copyleft uh from go ethereum or in our case from parity theorem at the same time, how it's different from silkworm uh from silkworm uh compared to silkworm it's in a different language, it's in in rust, not c, plus, plus, basically rust.

B

In my personal opinion, it's uh reimagine c-plus plus, as if c plus plus, was developed not in 1970s but in in the middle of 2000 2010s, basically using uh 40 years of in-between of programming, language theory, innovation to improve actually the design of the language, and it's uh much less of a hodgepodge that c, plus plus, has become because, like c plus plus has changed, a lot of times has gone through many many concepts. It's tried out a lot, but as a result, it's a much less cohesive language compared to rust.

B

So rust is much cleaner. It's much more convenient to work with it has the best of the industry tooling, like cargo package manager, for must it's. It is considered to be the gold standard of all package managers for all languages, and the situation here is polar in c plus, plus that c, plus, plus and c. Also they are considered to be the languages with the worst uh instiller slash package manager situation, so uh reusing libraries is generally a big pain point in c plus plus.

B

So because of that, because of of this points, it's actually much easier and much faster to develop a node in in raster than c plus plus, and it's most important. It gives comparable performance if not better performance, because rust is as low is as low level as c plus, plus. Basically, it was designed as essentially a c-class plus replacement.

B

So all this boils down to that. uh There are two languages uh rust and c, plus plus there are also two nodes: a cool and silkworm. Currently akula is already a full node implementation. It works it syncs, while silkworm it has the execution module and it can build some of the indexes, but it's it does not have the the complete uh syncing pipeline. So it it's not yet a full node, but I'm sure it will be. It just takes longer, if not much longer time to develop.

A

Interesting very interesting, I'm in one team coming up with three different client implementation: providing opportunity for users to maybe try different things differently and with uh some different features. I'm talking about like designs and features. um I understand the architecture of akula suggests a modular approach for the client development. Can you maybe talk a little bit about modular client design, and why was that selected for the full client development.

B

So the general idea about modularity um there are two layers to modularity. Basically, the first one is that in like aragon itself and uh optionally, akula too, they are like they can be both monolithic, full nodes that provides all the services, that is, the general sync transaction pool century. That is the the beer on the peer-to-peer network and in the future, like also the consensus like it's part of currently it's ingrained in a cola and silkworm and aragon, but it will also be able to be split potentially, but it's not necessary and rpc demon.

B

Most importantly like providing the ethereum json rpc, uh you can take out these components. You can run them as separate processes outside of the main akula or aragon binary, and you can run, for example, many instances of rpc demon and connect them to the same database.

B

This is one layer to modularity like it provides basically the ability to horizontally scale serving cpu intensive rpc requests like if call for example or tracing that is one layer layer to modularity. Another thing is that the core itself, which which is basically so something that I have not described yet so everything else is in core, but the core itself, like this, the syncing part, the syncing process, where which downloads the blocks processes them, builds the ethereum world state compute state route.

B

On top of it, it itself has a very unique um sinking pipeline, which is not yet seen anywhere in the blockchain space in general.

B

It's called stage sync, so, unlike other ethereum node implementations and not just ethereum any other node in general of any blockchain network, we do not provide. We do not structure everything as like a set of actors.

B

Instead, what we do is uh everything the whole the whole thinking thing. This whole thinking pipeline, it's just an infinite loop, which just calls uh several functions, sequentially like one after another after another after another, like uh basically, the difference is that we do not do processing by block, but we do processing by function, basically download all headers from genesis to current tip, download all blocks from genesis to current tip. Then, uh for example, retrieve all senders from transactions from genesis to currency then execute actually all transactions from current tip and build state.

B

Then compute uh the ethereum uh state route at the at the current time. So we don't do like everything for one block like the way it's usually done. I do everything for one block like uh download one block instantly process, it that says, execute it. uh Compute state routes, uh build indexes, then do the next block. We do it the other way around. So we do uh one thing for all blocks, and uh this leads to two benefits. Most important benefits.

B

First, is that one person one developer can take his one thing, one one index and not touch uh basically not touch the whole pipeline, so he can do just his one function. Do it well, do it in with in parallel, for example, if if it allows- and some things are actually parallelized compared to to the classic way nodes are built, for example, retrieving uh address of the sender of the transaction from the signature. This is a very cpu intensive operation it and if there are a lot of transactions, it actually takes a lot of time.

B

uh But if you do it everything by block instead of by one thing, then you cannot paralyze it. You cannot use the multiple cpu cores, so it actually takes a lot of time a long time and you have to paralyze it if you want to speed up. So our approach allows speeding this up.

B

This is one thing um so and, uh and uh that's uh basically, the advantage of the of our stagesync pipeline.

A

That is awesome. um I assume that processing by header and not by blocks may possibly lead to processing time. That can be an additional advantage, as you are saying already so um now. Let's maybe talk about uh the database, the same time and the execution speed of akula with respect to any other client. Those are available at the moment.

B

So, basically, the aforementioned advantages in how we do sync, they basically and also the innovations in the data model. They mean that you can again sync the archive node actually, and it will, it will work like it won't take you months. It won't take you weeks. It will take you at worst days, if not hours, depending on how on how fast your computer is and on how good your internet connection is.

B

So basically, there are two implementations that work with aragon, ideas that is arrogant itself and akula, but the difference between them is that akula has it's a clean slate implementation with many of the legacy stuff removed or like not present, because aragon inherited a lot from goethe, which is not really compatible with arrogant ideas. So this is actually getting cleaned out over time, but it still takes time, and another thing is that rust is just a much more low level at the same time, more convenient language than go.

B

It doesn't have garbage collection, it's generally more efficient in terms of cpu cpu cycles, and the current bottleneck, both in aragon and in ecuador, is the execution of the transactions of approximately.

B

I think it's just under 2 billion already transactions in ethereum, since 2015. so executing all of them from genesis to current tip. It actually takes a lot of time. It's actually the most the longest stage it uh and in a cooler it takes approximately takes what 18 hours 17 hours, depending uh on your cpu like. If you take like the best of the line cpu, that is, I think, uh 12 900 k or 12 900 ks from intel. This would take less than 24 hours but close to it.

B

This is the most intensive stage, and this is where our evm written in rust really shines. So this is like the uh the performance figure like there are also other stages, but in general, like execution, takes more than 50 of the time.

A

Interesting- and I believe you briefly touched about the uh the center part, the main machine that is working behind it- I understand akula evm is not the usual ibm that is being used for any other client. So maybe, if you would like to talk a little bit about evm modem or the apollo evens.

B

So basically, uh akula originally used edm audience.

B

It was also another project that I didn't mention uh another project built separately from separately as a separate component, just like sentry and ethdb evm audience uh key thing is that it follows the evmc interface which which allows for for switching out so-called hosts hosts is an entity that actually provides for for uh the storage of of accounts of storage slots and which also calls interpreter when, when basically, when the interpreter reaches a call into another, smart contract like it goes back to the host and host calls the the uh evm interpreter.

B

So if your model is not strictly an evm, it's an avm interpreter. It's like one half of it. uh It's resumable like this is like the key advantage of it. Is that it's resumable, so it's you can when it reaches like anything it it needs from host. Like give me storage, give me account it stops. It stops execution and it it hands. The control back to the programmer and programmer can do whatever the programmer wants.

B

However, this feature of resumability it's uh it has. It presents significant penalty in terms of cpu cycles. In terms of how long execution takes for that reason, uh akula does not use evm audit directly anymore. Instead, we we modified evm audim and uh it's now called akula evm. It's part of akula, it's not resumable, so it's just an edm interpreter. It's a very fast evm interpreter one of the fastest on the market.

B

uh It's integral part of akula and uh well, that's it. It works. It's fast.

A

That's great and thanks for clarifying okay, uh so one question now like is akula ready for users.

B

Well, the there. That's um that's an interesting question. There are like it depends on the user. Currently because akula works, a cooler, syncs cooler things fast, as you probably test it or our user. Our listeners could have tested already. It can be used.

B

As an archive note like if you are, for example, a business and you need tracing or you need logs for transactions or you just need some viewing to blockchain into ethereum, then you can definitely use a cooler one thing that we don't have yet we do not have a transaction pool and mining that is or validator mode. That's the same thing. We don't have them yet. This is work in progress.

B

It will take some time not long, but it will take some time for it to appear in a cola and be tested. So I'm not sure if you can use metamask already with a paired with a coolant mate. The answer is, maybe not, basically, the parts not related to the transaction pool, but as the archive node, not the transaction pull node but erica arnold. It fully works and we do have actual users who actually use a google to do their own thing.

A

Well, you partially answered my next part of this question like is there any specific target user base? You mentioned that validator mode is work in progress. That is fine. um Are you planning to maybe target hobbyist or just general ethereum node runner.

B

uh Just general ethereum node runner like uh I'm, not like we're, not cutting off anyone from akula. Obviously anyone will be able to use a cola for any purpose. So, like absence current absence of transaction, it's it's not a feature. It's a block that we are slowly fixing.

A

Right- and um there was a mention of like mining feature, it is not available, but is there a plan to have that available? Considering that ethereum will be out of proof of work.

B

So when I say mining, I don't mean necessarily the proof of work mining like money or validating or to be more precise block creation. So we are working on block creation, block creation and the ability to give away the block template. Basically, so if you can give a giveaway block template, then it doesn't really matter like what?

B

uh What kind of consensus you have? Is it proof of work or is it proof of stake like the key feature? Here is not money itself, not the crunchy numbers, but the creation of the block or at the side of the node like this is what we are working on.

B

If we yeah I'm, not sure we will have et hash at any point, maybe we will remove it after uh after ethereum maintenance uh does the merge. Maybe we will completely remove the traces of it hash algorithm, but it doesn't mean that mining will go away because mining is just the slang for block creation.

A

Right that make a lot of sense, so I was looking into uh the website the official website and it states that the akula client at the moment uh supports network out of the box, and some of the networks mentioned here is sepolia godly robston. Obviously, rink b is deprecated, for you know like eating the main chain. So let's talk about all these test nets and uh we know that merge is around the corner.

A

So um how soon do you think akula will be able to participate in devnets for march or maybe even after the march.

B

uh So akula does not participate in the in the merged devnets, which actually tells the emerging event and, most importantly, we do not have the plan to support the merged event itself. That is when the consensus does the switch.

B

uh The rationale is that we are not like we're not the backbone of the network yet and uh likely will we will not be by the time the merge hits mainnet so and there are no validators running akula, so it doesn't just make any sense for the complexity. So what will happen instead? Is uh we do implement post merge?

B

Basically, akula can run paired with uh with consensusly clients, it does run sepolia, it does run robs10. So what happens? Is that we just edit when the merge heads we just added the configuration of the network in the source code and just manually just change the consensus from from uh et hash to to beacon, merge consensus with and fill. I fill in the required parameters like the actual block where the merge gets.

B

So we do not have terminal total difficulty supported, but we do support this transition post facto and that's it, but we do support post merge and I expect that will be the same way on mainnet. So our users, when the merge event hits uh most likely we'll have to shut down their node. We will we'll do an emergency update on the point of the merge it will take. I don't know five minutes.

B

uh They will download the new uh version of the node and run it. It will be, as I said, five minutes of downtime- uh maybe it's unfortunate, but it uh allows us to not implement the merge which, which is arguably complex.

A

Right um so today there was an announcement of upgrade uh which is expected somewhere around august 17 for uh sepolia testnet, considering uh akula supports aponia. Are we planning to participate in this.

B

We we are planning to to to participate. Of course, like we have not looked, and we have not implemented this upgrade. Yet. uh I guess we have two weeks to do so, uh but we will continue being with the network, with the mainnet and with all official testers.

A

That's awesome, and uh after gauri uh and prata testnet march, I hope that similar upgrade will be planned for godly platter as well and yeah. That would be another testing ground for a cooler to have post-merge upgrades.

B

A

We briefly talked about like its readiness with a combination, cl client and have it like a support of validated node. Is there any favorite cl um client for a cooler, considering a lighthouse written in rust? Do you think it will be the most compatible or akula is designed in a way that it can be run with any consensus, their client.

B

Technically uh we are cooler implements not uh not like not any flavor of the of the consent of the of the merged protocol too, but implements like the the minimum amount necessary to get to be all with the network. So, as I said, we do not support validation. Yet so akula doesn't support the whole merge protocol, yet only the required minimum to be with the network to make sure that the consensus layer runs and the cooler months. So it was a conscious decision to optimize resources like it will be fixed just give it some time.

B

As for the exact consensus layer, clients, I'm testing with lighthouse personally, uh I have not tested all combinations, never had the time to do it. So lighthouse is guaranteed to work because well I'm testing it, but uh but if it doesn't work with other consensus layer implementations again, it's a bug not a feature, and if someone discovers an issue, we will probably fix it.

A

Right yeah, generally speaking like I have, I have talked to so many clients both on el ncl side and because they were participating in interrupt sessions and they were testing on different client combination. We definitely have like uh information that it works. The general consensus and execution specs works with each other, irrespective of the language of clients. So hopefully it works for a cooler too, and we don't find much bugs there. um Let's talk a little bit about client diversity.

A

Everyone knows that uh the execution layer is not so good. Looking at the moment, how do we see the future of akula?

A

um Maybe after the merge, when it is when you're ready for general usage, and do you think that akula will be helpful to get the network share of parity.

B

Most importantly, uh I think uh it's um it's a bit. uh It's a bit late to talk about uh parity share of the network, because parity does not exist on the network anymore. Only the nodes that didn't care to upgrade, but they have been kicked off with the latest uh difficulty bomb switch. So that's uh that's one thing: there is no more parity share on the network anymore.

B

I do believe that those who need to tracing this switch to aragon at this point and the tracing works and everyone works well, and it's like the recommended uh client to switch to, but akula may as well take some of that arrogant, share and a share of other implementations. I do not preclude that. That is very much possible again. It all depends on us. Everything is in our hands how fast we will develop, how fast we'll keep up with others.

B

It's a competition, but I just think we are handling this competition well enough.

A

That is awesome talking about like managing the client or maybe providing features and continued development. Where is the best place to find resources and how a community may be able to contribute if they are interested.

B

The general answer to this question is come to our telegram and we'll talk that uh that there uh there are always stuff to to help out with uh there is always stuff to discuss in private how or what can be developed uh in a cooler or if someone needs something special in a cola which is like something on standard, how we can collaborate.

B

This is uh this is all discussable. We do have an open telegram and all community members are encouraged to join it and come visit. Us.

A

Sweet, a follow-up, I'm not sure if it is appropriate here or not, but just out of curiosity most of the clients are on discord. Then, why telegram.

B

The there are two things to it. uh First of all, um telegram is it's a lot. It's much more convenient to use as general use messenger compared to discord like it's. Actually, the source of uh I'm I'm russian. I live in russia and uh in russia, telegram is, in general, the source of information source of news like the the newsline for everyone. It's also the private chat like what's up for your friends and family and uh it's uh it works fast. It works well.

B

It works in any remote location, much much better than discord, so this court is it's only good if you like need to build a huge community with a lot of channels. It's not currently the thing with a cola. We only have one channel one discussion room and that's it and uh this preference for telegram. It's also validated by the fact that many ethereum communities they are actually on telegram, like lobster dao, for example, like the most known example.

B

So I wouldn't say that discord is the universal choice for for communities in and at some point triple gap which ergon was called at that time. It was also on telegram until like there were, there were many many more channels, so it switched to this port, but it started on telegram as well. So.

A

B

It's a convenient starting point, at least, and maybe not not just a starting point. It's a it's just a very good messenger. To use super convenient. It works anywhere.

A

I would agree, I personally believe that telegram is helpful in many many many things like, especially when it comes to uh talking in terms of file sharing and transfer. It's very convenient than discard.

A

It comes like every every app comes with its own trade-off, so yeah, whatever works best here, all right. um It's probably time to wrap up anything that you would like to share with the community.

B

You're welcome to to use a cola you're welcome to test it, as I said it works and if you have any questions, uh feel free to visit our website. The I'm sure the link will be there. uh You're also welcome to visit our telegram, which is open to to anyone and say hello and ask any questions to the to the developers and to the community and uh akula is getting better and better day by day, getting there much faster than any other implementation before it.

B

In my personal opinion, and it will become, I believe, the backbone of the ethereum network at some point.

A

Awesome, these are nice words and I hope this encouraged new users or even existing user, to try akula atem. We appreciate you taking our time to talk about akula an efficient client node running option for ethereum users.

A

I hope, with the release of this talk, we see increased adoption of a cooler and a changing number of client diversity stats, especially for execution clients.

A

On this note, thanks to our channel subscriber for watching and listening to episode of know, your client series check out description for related link and twitter handle of guests to follow, hit like and share and increase the reach. Should you have any question on this or any other talk? Let us know at ethgatherthis discord we'll be back soon with another interesting talk till then keep watching listening and keep sharing your love with ethereum cathedrals, cheers.