Short Version: GHOSTDAG allows us to make a chain out of the parallel blocks in the BlockDAG, which results in the same security guarantees provided by Nakamoto consensus (bitcoin), but with increased throughput and lower confirmation times.
Long Version: The fact that Kaspa allows parallel block creation via a BlockDAG is just the start. Now we need an ordering mechanism that sorts the transactions into a linear stream of events – otherwise, we won’t know if the sender actually has the required funds to send a transaction, or if they were already spent in another parallel block.
And this is exactly what GHOSTDAG does. And it also takes care of:
1) In case of a conflict between two transactions that spend the same funds, the protocol will eventually choose one of them.
2) An attacker with less than 50% of the hash rate that wants to reverse a transaction has chances that are diminishing exponentially as long as more time passes since the transaction was sent.
3) Under normal circumstances, any transaction will be confirmed after 10 seconds. Double spending a transaction after that time requires a very high hash rate.
In addition to GhostDAG, Michael Sutton and Yonatan Sompolinsky are working on a new ordering mechanism called DAGKNIGHT where instead of the 10 seconds confirmation time guaranteed by #3, the confirmation time adapts automatically by the internet speed, which means, any improvement in the network conditions of the miners and any optimization in the block validation algorithm will also speed up the confirmation time. The DAGKNIGHT consensus mechanism was shared with the world on October 31,2022 at the Crypto Economics Security Conference at UC Berkeley. Watch the video below to hear Yonatan’s presentation and subscribe.
The basis for DAGKnight are ideas Michael and myself came up with to solve the pruning problem in DAGs (Medium post pending) but I was never directly involved with DAGKnight So what makes DAGKnight so exciting? Simply put, it is the *perfect* PoW based consensus algorithm. It satisfies all points in the desiderata.
Recall that the huge advantage of GHOSTDAG over all other PoW algorithms is that it removes the security constraints on throughput.
In all other techs, increasing the block rate directly increases orphan rates whereby killing the security. Some techs allow sharding the network and then increasing the number of shards while retaining slow blockrates on each separate shard, whereby increasing the blockrate but retaining slow confirmation times as well as many other issues consequential to sharding such as load balancing and data availability issues. However, they still do not allow increasing the block rate in each shard.
GHOSTDAG is the first PoW to allow reducing block rates on a non-sharded network, which is the reason it can achieve unprecedented confirmation times. However, GHOSTDAG is still limited in the sense that it is *non-responsive to network latency*. That is, we still need to hardwire an upper bound on network latency (which we can assume holds 95% of the time), and the rest of the properties of the network (in particular, confirmation times) are derived from this bound. This means that the performance does not improve as latency improves, and worse, that the security is compromised if the network latency deteriorates. This is true for *all* existing PoW algorithms, with the only exception being SPECTRE, another algorithm conceived by @hashdag, which he once described to me as his “most beautiful creation”.
However, SPECTRE has a different flaw: it does not provide a linear ordering, and blocks can switch places long after they were confirmed (though never in a way which invalidates transactions). This is fine for a distributed ledger, but abysmal for smart-contracts, which is why it was decided that Kaspa would implement GHOSTDAG and not SPECTRE.
And now we have DAGKnight, which achieves *both*, and is the first consensus protocol to have it all: 1. Nakamoto consensus security independent of block rates (like GHOSTDAG and SPECTRE), 2. rapidly converging linear ordering (like GHOSTDAG), and 3. responsiveness to *actual* network latency (like SPECTRE).
Simply put, it is a PoW consensus algorithm which has no speed limitations beyond hardware, suitable for smart-contracts, AND *scales itself* as network latency is improved.
This is the true epitome of what PoW can be, and should make PoW proponents very excited!
And a bonus feature is that the combinatorial similarities of GHOSTDAG and DAGKnight make it that many of the utilities required to implement DAGKnight are already present in Kaspa!