Thread by @paddykcl

next up is Ronan Oliynykov to tell us more about DAGS.

Looks like he'll talk about Ghost, Spectre, Pahntom, Graphchain, <not be named cuz trolls>, Avalanche

#onthechain
Live stream: facebook.com/BDTechCommunit…

He is part of @InputOutputHK yay

He is going to talk about dencetralized consensus protocols.

No trusted parties, peer-to-peet flat network, each connected node can faily arbitrary. No list of malicious of faculty nodes available to honest parties, that list is dynamic too.

There are present network failures (arbitrary delays or message drops).

So he is considering worst-case scenarios, and our cryptocurrenies should work in this environment.

What are the requirements?

Reliable operation - where honest nodes converges to the same state as any other honest node in a limited number of steps.

Honest nodes only follow known formal consensus protocols (no human interaction or extra info besides p2p network).

Assumptions:

1/2+ majority, or vast

majority 2/3+ of honest participants.

Redundancy is accepted, all honest nodes may independently execute the same algorithm.

Where do we use such protocols? Cryptocurrencies!

Highly reliable computer systems (data bases, computing clusters) and critical systems (avionics,

spaceship systems, nuclear power stations, etc).

He has a picture of global with transaction propagation. Point is that there might be an inconsistent transaction sequence across all nodes (because of delivery delay).

Delivery delay due to latency + validating tx

So the first solution was the blockchain of course, everyone just follows the longest proof-of-chain chain.... (slows down confirmation, but provides total ordering of all transactions). I'll not explain this part - since most people who follow this tweet storm should know it lol

Disadvantage of PoW blockchain - slowdown of throughput, security based on physical limits, energy consumption. Security limitations based on block generation rate vs block propagate rate. Long time for transaction approval and trends towards centralisation due to mining pools.

First solution - a tree based consensus protocol -> GHOST Protocol:

it doesn't look at the "longest chain", but the the heaviest branch. He has a nice diagram.

Now into the Spectre world.

A block includes links to all previous unreferenced blocks (leaves) in the observed DAG.

He has a picture from the original DAG paper.

We can have conflicting tx in the DAG. Which tx do we accept?

It sounds like there is a voting mechanism... if some block B sees a transaction in their past in block C (but not the competing block A with t), then this block B votes for the past block C. (perhaps someone can explain this better than me on twitter lol) You take into account

the DAG that voted that tx X came before Y.

OK so now he is going to explain Phantom. The authors changed how to reach a decision on which transactions are in the block.

When I mine block, I see the past of all mined blocks. My blocks will be seen by future blocks.

I see blocks I reference in the past, my block will be referenced by blocks in the future.

At this moment, several blocks will be mined at the same time, and I won't see these blocks. So they won't be referenced by mine.

The parallel blocks to "intercone?" - if we consider

honest parties, then the size of the intercone - it should be small - since not many blocks are generated in parallel. Why is this situation different? If someone would like to organise attack - attacker will try to build sub tree and needs to overcome sub tree already created.

OK so lots of blocks will belong to the attacker, but since they were not public - likely not to have any weight since all the honest parties will reference their blocks. So there is a linear numbering of the blocks.

@hashdag @yoadlew is there a concise and tweet-size way to describe the transition from GHOST to SPECTRE and SPECTRE to PHANTOM?

He is providing a summary about Graphchain, <not to be named cuz trolls> and Alavanche. But not a lot of detail about it (guessing cuz of time).