Hanging blocks are essentially like blocks on a blockchain, but instead, they register a Merkle root in an Ethereum contract, and promise the corresponding file was made available. Checking on the promise, and creation of blocks require people to have some kind of stake in the system.
A client making a block registers the Merkle root in the hanging block contract, and spreads around the file. Block makers must also put in stake that can be used to punish not providing the actual data to the block, or invalidity of the block.
The criterion to allow people to make a block can be arbitrary, it could be
random based on RANDAO and
msg.sender and the stake. Or it could rotate
through a list, and the nearest to next one is allowed to create the block.
‘Proving’ that the block data is available
If the files are not being shared, people cannot point out invalid facts in said files as discussed in the next paragraph. To ensure it is available, there is a vote about it.
This is why a system where people have stake is necessary, otherwise it is not possible to defend against many fake accounts. (Sybil attack)
Validity of blocks
Entire blocks, and their successors can be made invalid by when the combination of chunks, as proven to be in a block with Merkle paths, are shown to be invalid. The hanging block contract logic would contain the ability to check this.
The size of the transaction to do so, if the hanging has size n, is ∝log(n) -the length of path to prove a chunk is in there-, plus the size of the chunk itself.
What can you do with it?
You can do anything with it where you can use a few chunks that are not inpractically large to show invalidity. What does this imply? For instance, what if there is a program f with states Si and input Ii?
f(Si,Ii) = Si+1 †
This can be done! Store triplets in the block: (pi, Ii, Si), where pi is the previous change of state. We need to check two things:
- pi as reported is correct: Do this by searching the hanging blocks for one that is actually between the claimed new one and the old one. If there is, it is a invalid transaction.
- End state reported is wrong: This is the requirement at †, and it can easily be checked by the client and verified to be wrong by the hanging block contract, by providing the two relevant blocks.
Of course, f needs to be available to the Ethereum contract one way or another.
The program here is limited by the entire state needing to be offered up to to the hanging block contract. So this state cannot be too large. Furthermore, barring things like cryptographic commitments, the different programs running on the hanging blocks cannot interact.
This can easily be fixed by allowing contracts to depend both their own and other contracts' state, and simply have more backward state connections that need to be checked to be correct, and can be shown to be wrong if needed.
Communicating with contracts on regular Ethereum
Contracts cannot know about the data in the hanging block themselves. However, a client can prove what value chunks of it have via the hanging block contract; it is just the same process as proving invalidity, except the chunk is send to the other contract. So it is somewhat limited, but useful interactions are possible.
The ‘depth’ of a chunk doesnt increase security the same way as mining, but the passing of time improves the chance that inconsistencies were reported, and decreases the chance that a vote to consider it missing will take place.
Amazingly, if you look at the above, you can see that this does not actually require all full nodes to compute everything about the hanging blocks. But it does require a lot of clients responding to inavailability the data by voting the block to be missing, and some clients checking everything, and reporting invalid actions.
It is not quite clear what the implications are, or if i am missing some problems. If it is a good idea, likely the best approach is to try figure out how to make developing for hanging blocks similar to developing for Ethereum itself, so if people know how to do the one, they can do the other.
The data the Merkle tree are about do not need to be all explicit in memory/harddisk. For instance, the client could keep track of the state like Ethereum, and just calculate the intermediate states as needed. (Essentially, this just compresses it)
Furthermore, considering not everything needs to be run, you might not even need all the data either, just the checksum of other parts the merkle tree on some branches. However, due to the need for those votes, most of the clients will need to be able to tell when any part of it is not available, because a vote to falsely claim everything is available needs to be countered. Of course you might try have a “raise the alarm if a bit can’t be found”, but an attack of spamming such alarms would have to be ineffective in that case aswel.
A thread on /r/ethereum.c