The road so far

Fantom | 01.20| 139

By Michael Kong

On 15 February 2018, the founders of Fantom had an idea. An idea for a different approach to consensus.

In the years before the Nakamoto consensus, we had very little reason for practical fault-tolerant systems in an adversarial network. We did have fault-tolerant systems, however, and had near perfected distributed consensus mechanisms such as RAFT.

Then came along Nakamoto consensus, which suggested a mechanism to be able to reach consensus in an adversarial setting. Before this, no one had practically implemented the idea of trying to create an adversarial designed system, let alone financial ones. And Nakamoto consensus pioneered and changed the landscape we have today.

But as with all technological evolution, or perhaps, in this case, revolution, there are multiple iterations until we reach the pinnacle.

We have managed to design practical byzantine fault-tolerant systems, and in my opinion, tendermint is the first-class example.

Fantom had another idea, though: what if we could improve on practical byzantine fault tolerance?

Small changes, such as asynchronous communication to increase connectivity fault tolerance. And knowledge graphs to decrease message overheads. These changes may be small, but small changes can have profound impacts.

That was the goal set out to be accomplished on 15 February 2018.

An asynchronous, optimized byzantine fault-tolerant protocol that could function in not only adversarial networks but also loosely connected ones, while decreasing message complexity and overhead.

The new internet won’t be a spoke model, but instead a mesh model, and for it to function, it needs to have loosely connected assumptions, as well as very low message overhead. This proposed consensus is not only an advancement but will become a necessity in the future.

We spent the rest of 2018 resourcing a team capable of not only sharing this vision but being able to execute on it. At the end of 2018, we released the first viable proof of concept. It was unoptimized, didn’t scale past 30 participants, and had significant finality drop-offs as participants increased. But it proved the baseline. From here, we could identify the problems, replicate their environments, and propose solutions.

And through all of 2019, that’s exactly what we did. We proposed a total of 7 academic papers, all focused on the issues we occurred — each addressing a new flaw that came up in the design.

It has taken a team of 9 researchers in 4 different countries to identify and propose solutions to these problems. It has taken a team of 22 engineers in 4 different countries to try to build out these solutions.

And we did it.

On 27 December 2019, we invite everyone to publicly participate during the genesis event of Opera, the first showcase of fully powered Lachesis consensus. That dream that started in early 2018, conceptualized in late 2018, and implemented through struggle in 2019, was actually possible.

An open-source, scalable, asynchronous byzantine fault-tolerant solution for loosely connected networks with low message overheads in an adversarial setting. Or as we call it Lachesis.

Now, for the everyday user, this might not mean much. If we do our jobs right, you will never know that Lachesis is powering the app you are using on your smartphone. But you will be using it. As we move to an interconnected mesh peer to peer world, we will be there, ready to guide.

Official Links for Fantom:

Official Email Address:

Official Website Link:

Official Telegram Discussion Group (English):

Official Telegram Announcements Channel:

Official Subreddit:

Official Twitter:

Official Medium:

Official Github:

Official YouTube Channel:

The road so far was originally published in Fantom Foundation on Medium, where people are continuing the conversation by highlighting and responding to this story.

Comment 0


Are you sure you want to delete this post?