#219 Mance Harmon: Hashgraph – A Radically Novel Consensus Algorithm

Support the show, consider donating:
BTC: 1CD83r9EzFinDNWwmRW4ssgCbhsM5bxXwg ( )
BCC: 1M4dvWxjL5N9WniNtatKtxW7RcGV73TQTd ( )
ETH: 0x8cdb49ca5103Ce06717C4daBBFD4857183f50935 ( )

Hashgraph is a new consensus algorithm that radically differs from proof-of-work as well as proof-of-stake consensus algorithms. While work on Hashgraph begun in 2012, it's design is radically different from today's blockchain architectures. The Hashgraph team claims that it has found an optimal consensus algorithm design that will be impossible to significantly improve upon.

We were joined by Mance Harmon, who is CEO of the Swirlds, the company developing Hashgraph. Our conversation covered the origin story of hashgraph, how it compares to existing consensus algorithms and how Hashgraph works.

Topics discussed in this episode:
– Leemon Baird and Mance Harmon's long history of building companies together
– What motivated Leemon Baird to start working on Hashgraph in 2012
– The existing categories of consensus algorithms and their problems
– How Hashgraph consensus combines voting and gossip protocols
– The performance characteristics of Hashgraph
– How a public Hashgraph network could look like

Links mentioned in this episode:
– Hashgraph Homepage:
– Hashgraph Whitepaper:
– Hashgraph Consensus – Detailed Examples:
– Sybil Attacks in Hashgraph:
– Hidden Forces Podcast Episode on Hashgraph:
– Lemon Baird's Talk on Hashgraph at Harvard Business School:

Watch or listen, Epicenter is available wherever you get your podcasts.

Epicenter is hosted by Brian Fabian Crain, Sébastien Couture & Meher Roy.

22 Comments on #219 Mance Harmon: Hashgraph – A Radically Novel Consensus Algorithm

  1. More hand-wavy technical abstraction. A very complex solution in search of a problem. “Radically novel”, they swear it; Trust them.

  2. One point of clarification… When we discussed finality, I misspoke at one point. Cryptocurrency as an application doesn’t require a total order on transactions. A partial order is sufficient. Blockchain does result in a total order. Bitcoin blockchain is not BFT because the network never achieves 100% certainty on the order of transactions. However, it does result in a total order on transactions. I hope that clears up my earlier comments.

  3. When will the first public hashgraph launch? What will be the first hashgraph cryptocurrency to trade on an exchange? Will there be a crosschain distributed ledger coin between this type of DAG and different blockchains? This episode raises more questions than it answers.

  4. Very good interview – loved it: as far as I understood, they’ve not fully defined the public network part (or they are not ready yet to announce it). Hence: Is it possible to imagine *multiple* public networks preventing sybil attacks via PoS *or* even *PoW* and still have the mentioned gains like high TPS, finalty etc.?

  5. Another point: does the hashgraph allow a complete local reconstruction of the gossip network? If yes, would that not be a problem, because people would be able to locally calculate optimal attacks to partition the network?

  6. Look, at the end of the day it comes down to whether your project is open-source or not. No one of consequence is going to simply trust that your code runs smoothly in an adversarial environment. The world won’t. It needs to be examined openly and freely. So, until that happens does this really matter?

    • I would assume that once they start implementing a public ledger they will make it open source. You can say “until that happens” for just about any ground breaking idea. The point is whether it will or won’t happen and when. “There’s an arbitrary amount of time until this car hits me so it doesn’t matter…”

    • Yes, a serious red flag to permissioned Digital ledger technology. If it’s no permissionless, then its just another form of centralization or control, simple as that.

  7. Man Brian, I really appreciate you ask some great questions. but can you just stop yourself from interrupting people when they are describing the idea in depth ref- 37:00

  8. So, the nodes need to gossip with each other, whether they contain transactions in their packets or not. Is there a strategy for keeping the percentage of transactionless packets as low as possible, so that bandwidth efficiency is maximized?

Comments are closed.