Does a production-level Plasma spec exist?


#1

There is quite a bit of confusion about Plasma recently. There are multiple Plasma flavors considered and researched. From reading this message board my humble conclusion is that there is not yet a production-level spec for Plasma. There are research-level specs or proof-of-concept specs but not a detailed production-level spec. This is totally fine imho since people are trying different things, approaches, zksnarks etc.

What is really strange though is there are multiple companies claiming that they do production level Plasma. I am really confused what are these guys developing if there is no spec yet. Investors are investing zillions of dollars in Plasma. What are they investing - into research/prototype specs that still need to be polished, or in a production-level spec that has all security subtleties perfected?))

So let me ask a question on behalf of many people interested in Plasma - is there a production level Plasma spec, security of which is approved by @vbuterin and other people at ETH foundation?

Imho it would be nice if ETH releases a public statement regarding Plasma, something like “we have prototype specs, and expect a production level spec 6 months from now”, or “we have a production level spec, and here is a link to it”.

It cheaper and faster to build an insecure system than to build a secure system. If there is no official guidance on Plasma, it almost certain that insecure systems will be first to market.


#2

Can you name some of those companies/projects?

To the best of my knowledge, no.

I think the later makes sense (once we are there), but remember that there will never be a single Plasma spec, Plasma is more of a design philosophy/set of guidelines.

Yes, and it’s kind of concerning if one of those projects actually goes live (although I think chances are slim, they just want to grab naive investors’ money). Then it could make some mess and damage the reputation of Plasma. That’s why I think the Eth community will/should discuss and review such projects (one of the reasons I’ve asked you to name them).


#3

All Plasma designs have vulnerabilities or at least specific UX tradeoffs that need to be addressed. Some companies put a PoS model on top and claim a kind of “security by obscurity”. These companies need to be called out.

I’d leave Vitalik and EF out of it, as their involvement has allowed for the Plasma-hype in the first place. Rather I suggest to define the requirements that a spec would need to satisfy to be considered ready for “production”.

Ultimately, any Plasma design will have to prove itself in production deployment.


#4

Congratulations for your upcoming fund raising. Your SKALE Network seems like EVM-ecosystem scaling solution. And I personally guess the withdrawal procedure will be done by ParityBridge-like somehow trusted(federated) withdrawal module.

But I’ve seen this article and it says “SKALE will be the first implementation of the Ethereum Virtual Machine (EVM) on a Plasma chain”.

So there is quite a bit of confusion that SKALE is EVM Plasma, but sounds like it has exit mechanism. Do you plan to use expensive EVM inside EVM construction? Or, is it Plasma?


#5

Not really production level, but something I’ve been workshopping a little lately: https://www.learnplasma.org/en/resources/#plasma-mvp-specification

Feedback would be very appreciated. Definitely needs more detail on the child chain side of things and client interaction.


#6

@sg - thank you for your message and congratulations on the fund raise! There has been admittedly a bit of confusion in the way we got described in the press. We do not consider ourselves Plasma. In my humble opinion Plasma is a very specific thing which includes a Plasma operator that operates an UTXO chain as well as a set of exit algorithms that are executed if the Plasma operator becomes malicious.

At Skale we are building a network of ETH-compatible side chains that use fast PoS-like consensus. Miners will be able to add servers to our network and stake Skale tokens to make money. Out of this large network of servers we will be able to build side-chains on demand by randomly picking servers, and then rotating them from time to time. A dapp developer will be able to pay in Skale tokens and get a side chain for her application.

Nowadays for some people (including investors) Plasma is more of a generic term everything that speeds up transactions :slight_smile: If they want to keep calling us Plasma, we do not really argue too much … ))


#8

It looks like OmiseGo is the leader in terms of Plasma implementation on the moment. For me OmiseGo are the good guys, in a sense that they are trying to make things secure.

So in this spec, what happens if I start with a $100 UTXO, and do a 100,000 transactions sending 1 cent to myself (to my other address).

I then randomly pick out of these transactions 1000 spent UTXOs and try to fraudulently exit all of them.
How does this work in case of OmiseGO? Who is going to be the party challenging the exits and how is this party going to be paid?


#9

User must put up a bond on each exit, challenger receives that bond for a successful challenge. Slight oversimplification of the solution but that’s the basic idea. Incentives should be aligned such that people will just challenge stuff because it’s worthwhile to do so.


#10

Where’d you get $900 from for the other 90,000 transactions? :stuck_out_tongue:

With plasma cash, you can create 10,000 tokens from this $100 and then you just submit proofs of inclusion / exclusion when it comes time to exit. Anyone can challenge an exit, but in the case of you sending coins to yourself, there doesn’t seem to be any reason for anyone to challenge given that you’re only touching your own collateral.


#11

Understood - thank you. I think the next step is to provide a formula for the value of the bond. It should probably be proportional to the amount being exited.

Thats the problem - you are creating lots of spent UTXOs that you can try fraudulently exit by claiming them unspent - what Kelvin says is that for every exit the person exiting is supposed to post a bond which serves as an incentive for bounty hunters to catch fraudulent exit attempts.

There is another difficult problem, which also needs to be addressed somehow. Lets say the operator becomes malicious and everyone attempts to exit the system.

Due to performance restrictions of the main chain, one will probably be able to have no more than one exit per second without significantly affecting the gas prices.

This means 3600 * 24 * 7 = 100,000 exits maximum in a week.

Therefore, each Plasma operator should(?) be limited to having 100,000 coins maximum unless batch exits are implemented.

May be a way to solve this is to require multiple Plasma operators and limit each of them to 100,000 coins. Then there has been

It is a hard problem, I hope it will be solved somehow …


#12

Is this relevant? (by @gakonst)
In the conclusion it reads:

A reference implementation which is used in production is provided.



#13

Does this include an exit bond similar to OmiseGO implementation? I think it does, but does not specify the value of the bond. It should probably be linearly proportional to the amount exited.

In the case where exits from self-transfer transactions are challenged, the challengers need to potentially have a copy of the entire Plasma chain, which may be terabytes of information. This needs to be discussed I think. The Plasma operator needs to provide to the bounty hunters a way to continuously download the entire chain.

In the example where Alice creates a chain of 1M self-transfer transactions and attempts to fraudulently exit some of these transactions, there are no counterparties that will care about storing the proof of Alice spending the transactions.

Therefore, the only party that will be able to challenge Alice will be the Plasma operator itself, or the bounty hunters. But bounty hunters will need to have an up-to-date copy of the entire chain.

The funny thing is that if, say, there are 10 bounty hunters that have to maintain a copy of the entire chain, it is almost like having a PoA chain where there is one block producer and 10 slaves.


#14

Unfortunately the paper does not go into detail on that topic and it’s not discussed as a possible attack. However, I could imagine that this part also solves for the attack you describe:

Starting an exit for a coin requires providing the transaction that gave the exitor owner- ship of the coin signed by the previous owner in the coin’s history, as well as a direct ancestor of that transaction.

I haven’t had much time to think about it, but could it be forbidden to exit a coin that you sent to yourself in the parent transaction? :thinking:


#15

Why do you believe bond pricing should depend on the amount being exited?

Creating a chain of 1m self transfer transactions in Plasma Cash is just Alice having a certain coin and spending it over 1m blocks to herself. At any given point if she attempts to exit at an earlierpoint than the latest spend, her exit can be challenged with a direct spend of it. No need to store the whole tx history. The challenge is just 1 merkle branch (+signatures obviously) which proves that the coin being exited has been spent.

Plasma Cash is not vulnerable to any of the attacks you describe, exactly because each coin is unique and independent from each other.


#16

Actually we are discussing Plasma MVP and not Plasma Cash ) OmiseGo spec we started to discuss above it Plasma MVP. The same is as a understand true for Loom Network.

So for Plasma MVP, if I have a long set of transactions sending money to myself, then no-one except myself cares about these transactions.

This means that if I try to fraudulently exit some of the spent UTXOs, I will have no counterparty to challenge this, since I was both the sender and the receiver.

So the only one to challenge me will be “bounty hunters” who just want to make money from my deposit.

Do other people on this thread agree to this point of view?

If I am a bounty hunter, I need to have a copy of the entire chain since fraudulent exits may come from anywhere in the chain.

At 1000 transactions per second, a chain will produce 6 TB a year. This means that a bounty hunter will need to run a Hadoop or Spark cluster to verify exits. You wont be able to do it on a PC.

Getting the economics right for the bounty hunters will be super important. If they are only paid when they find fraudulent exits, the question will be to have enough fraudulent exits to support the hunters.

Lets say I want to become a full time bounty hunter and run a Spark cluster to catch bad guys. I will probably want to make $200K a year and also pay for the cloud costs. The question is, how I am I going to make this $200K a year if I am only paid when someone does fraudulent exit attempts and I catch them ? The better job I do, the less fraudulent exits will be attempted, the less money I will make.

May be the Plasma operator needs to do some mockup fraudulent exits to feed the bounty hunters?))


#17

That’s why mass exits should exist.

I’m not sure how you arrived at this 1txps exit rate?

This is trivial to game, they only need to collude.

This is a valid point (the numbers can vary, though).

And this as well. And on top of everything, users need to trust these entities (no one watches the watchers), which is not in the spirit of Plasma.