Diseconomies of Scale: Strengthening Ethereum’s Decentralization

Transcript

hello everyone my name is Tony I'm with the applied research group at the airm foundation and today I'm want to talk about a topic that is very important to me and I guess also very important to the broad etherum ecosystem very important to many of you which is decentralization before I get into my talk just a rough outline what I'm going to talk about I want to start with a quick recap on attestations why do we need attestations when it comes to Air's consensus what are attestations and why are we attesting second I want to quickly outline one of today's problems and finally quickly introduce an EAP EAP

7716 why we need it and make the case for its inclusion in a future hog so let's get started as said a quick recap on attestation so validators on airm blockchain attest to blocks in order to kind of um coordinate on the on the canonical blockchain to agree on what is the latest block and what is the block to finalize and technically an attestation contains of three things a source vote a Target vote and a head vote where the source and the target vote are mainly used to um finalize certain checkpoints in time and the head vote is used to agree on what is the latest block that we see as um canonical

as confirmed and that the next proposal Builds on top of validators have an incentive to attest because they receive rewards for correct and timely attestations and on the other hand if you if validators don't attest they receive a penalty here a quick um image of how testing works so as the example here shows if you're the tester in the first slot of an Epoch then you basically um place your your Source vote to the um previous Epoch your target vote goes to the first slot in the current Epoch and your head vote is is um pointing to the latest block to the latest slot that you

see as valid of course this assumes um no reor no Miss slot but it's good enough um for Simplicity here right so let's get into the problem and the problem is solo stakers are struggling today why is that the case so what you can see here on this um bar chart is a comparison of professional node operators like K like eifi Lio and so on and we on the other hand of the spectrum we see solo stakers and what the chart shows is the expected um the expected missed head votes over time compared to the market sh of the respective valid data and if

we focus on the very left hand side of the chart we can see that very that those professional node operators they have a great performance so they don't miss attestations and for most of the times they ATT test correctly on the left hand side ah sorry on the right hand side we can see solar stakers and solar stakers are basically the from a performance perspective the worst right solar stakers struggle with a testing correctly and timely and this is kind of expected because solo stakers are known to be those small entities that run the Run uh some Noe from their home base they might have power outages have no backups they might have to fix their Noe at some point they might go on

holiday so there are many reasons while it is expected the solo stakers perform bad on the other hand solo stakers are super important because there are situations let's for example a 51% attack sensorship and so on where solo stakers are super important so while solo stakers are performing form in worse than professional node operators they are very very important um to to allow eum to achieve its Mission so what is the problem here staking comes with centralizing forces and this is kind of inherit to affirm today there are those centralizing forces that we continuously need to

fight against and those centralizing forces they originate from one very in from one very um hard to defend thing which is economies of scale so economies of scale many of you might heard of of that phrase is basically if you can reduce your costs just by scaling up the operation and of course solo stakers cannot do that professional node operators they can and when we look um onto ethereum staking today then we get something like this picture where the more valid datas that you run the lower you can push down your costs and if we had to um if I had to show you where solo stakers are they are

on the very left hand side of this curve while the professional node operators are on the right hand side of this curve so the professional node operators have it can push their costs down while the solo stakers cannot really leverage econom economies of scale this um on this table you can see a quick example how economy of scale play out in ethereum today just comparing solo stas with uh for example a coinbase um for example with coinbase and we can see that then even though they both run one device one machine which is in this case uh just as an example and Intel knock um we can see that you can run thousands of valid

datas on one machine right this of course comes with increased costs for example you have to pay higher bandwidth you might need more electricity but in the end what we can see here is um the coinbase box so the coinbase machine would make thousand times more profits than the solo staking box on the other hand the costs are not thousand times um more and this is exactly what I mean with economies of scale coinbase can operate much more efficiently than uh the solo STA and of course having your valid datas run from one machine is not the only source of economies of scale you

can use one cloud service provider isps the geographical location client software but also the maintenance responsibilities are um um are sources for economies of scale for example a solo Staker might um one solo Staker might manage one valid data whereas a professional node operator might have a team of 10 to 20 people managing hundreds of thousand thousands of validators so the question is what can we do against that what can we do to improve the decentralization that is kind of um in staking um in the situation where kind of those centralizing forces push us

towards centralization one thing that is very important here is economies of scale don't come without correlations which means if you put thousands of validators on one box those validators behave in a very correlated way for example if you have a power outage and the Box goes offline all your thousand validators will be offline too right so you have a high correlation between those validators and this is very important to understand you cannot really leverage economies of scale without also creating High correlations between your validators and the goal of eum must be

to reduce those correlations um for example if Google goes down or if AWS goes down their film Network um should not suffer right we should have a uncorrelated network that ensures that even if something happens um that um is an external Factor like some cloud service provider going offline we want air to still continue as normal so the goal must be to reduce correlations on every front and then what we've get is basically on this picture you can see we will go on onto the left path so we want to go towards diversity when it comes to

the staking setup we don't want to have homogeneity when it comes to staking because then we um open ourself to those external threats that are very bad right we don't want to have all node operators in one country because if something happens in that country that effect sters then things can go bad very quickly and now this is the potential solution EAP 7716 anti-correlation attestation penalties the EAP itself is very new so as you can see in the date created it's two months old now um it was um created by De line withal me we improved it um

in the E Berlin hecon so all of the ideas I'm presenting now are still very new so keep that in mind the EOP itself is called anti-correlation at the station penalties and the goal of it is basically to introduce um an increased penalty when you're correlated a lot with other validators and decrease the penalties if the correlation is is low and getting back to our economies of scale curve this should kind of be the outcome that we want to achieve through that Improvement proposal we want to get that economies of scale curve down flatten it so that the solo stakers has the same economic benefits than the big

node operators here is a quick example we see two boxes this can be two machines one of them runs a lot of valid datas the other one has only one single validator and the thing is if the box with the many validators goes offline then we see a high correlation which means we want to see a high penalty for all of those validators in that box on the other hand if the box with only one validator goes offline then the correlation is very low and we can also um set the penalties very low so the EAP basically proposes to scale penalties based on the observed correlations within the

network here you can can see some initial results from some analyzes um we did a few months ago and what we can see here is that introducing eap1 uh 7716 would lead to small entities getting less penalized while big entities will be penalized more and this is exactly how to counter um economies of scale so here you can see the small size clusters which include solo stakers they would have had less p penties then um compared to the current situation while the large siiz clusters they would face larger penalties and this is purely because they are leveraging economies of

scale thus creating High correlations within their setup and then getting higher penalties this brings me um to the open question as said the EAP is still very very new so there are still many open questions that we have to answer before we can ever ship that EIP first of all One open question is do we want to apply anti-correlation penalties only to the source Vote or do we want to apply them to all parts of an attestation so also include Target votes also include head votes there are some arguments for only doing it for Source votes because for Source votes you're very independent from external factors for example the

slot index within an Epoch on the other hand I in my opinion I think we should it for all parts of the attestation just because um even if there are some external factors like timing games and so on we in the long run we should see um the positive effects of anti-correlation penalties playing out no matter if it's only the source V or we also including the Target and the head next the exact me mechanism and the constants if you look up the EAP today you will see it's in a very very drafty stage still so there are are some constants that we will have to set and I invite everyone of you to engage in that discussion help us find the right values

find the right mechanism how to introduce those anti-correlation penalties and finally of course there's a question when can we ever ship such a change um it might be too late for the pure hard Fork by now but the goal has definitely the goal must be to ship it in the fork after Petra but in the end it's of course the community um every one of us who participates in the discussion who voices that such a change would be important should be prioritized that um in the end pushes the EOP towards being included in some future hard for finally let's together make if you're more decentralized please engage

on Twitter voice your concerns if you're a Sol Staker tell us how how you're doing if it's hard to keep up with um with all your duties and also engage in the discussion on Twitter farcaster on social media and so on because if it's not the community that says we want to have such a change then such a change will simply not happen second um I invite everyone to contribute you can find um below um on Eve magicians for example you can find all the links on our GitHub so my GitHub is no relation you will find um many links there you can find lious um implementation you can find implementation into the ethereum um

consensus layer specs and also on E research if you want to analyze the whole thing please engage reach out to me I'm happy to help you and together we can then analyze the whole thing how would it change how um which effects would it have on solar sters and in the end kind of yeah add some weight to the whole proposal by showing the potential impact it would have yeah and here you can also see my Twitter name it's the same on telegram please reach out if you have any questions I'm always happy to help you with that and this is it thank you very much