Hashgraph: How does the next-gen post-blockchain tech stack up?
It's apples and oranges. Hashgraph is very similar but completely different to other non-blockchain DLTs.
Update 28/3/2018: This section has been updated to include feedback on specific sections from Paul Madsen, technical lead at Swirlds and key Hashgraph developer.
Distributed ledger technology (DLT) refers to the system that underpins cryptocurrencies. This is the decentralised ledger that records all transactions, forms consensus and keeps the network secure. Blockchain is used by bitcoin and many more, and is undoubtedly the best-known type of DLT.
Blockchain systems store transactions as a string, one after another. This chain is secured by being observed by "nodes," who essentially keep a close watch on the blockchain and make sure they're all looking at the same thing, and that no one's trying to do any funny stuff to the network. These nodes are usually miners.
This is very robust and highly secure, but inherently slow and only able to handle a certain amount of transactions at once. The original bitcoin blockchain, for example, tended to hit its limit at only 7 transactions per second while Ethereum capped out at around 15.
This inherent limitation is often called the "scaling problem," and solving it is often seen as one of the main obstacles for cryptocurrencies to overcome.
Cryptocurrencies have been exploring different ways around this. Some of the more popular approaches are:
- On-chain solutions – Fit more transactions onto the existing chain. Bitcoin Cash did this by simply increasing bitcoin's block size from 1MB to 8MB. It's generally regarded as more of a bandaid than a permanent scaling solution though.
- Sharding – Break up the blockchain into independent sections that can be processed by different nodes. Ethereum plans to implement this alongside layered solutions.
- Layered solutions – Run some transactions off the main blockchain to ease the load on the main chain. The bitcoin Lightning Network is a layered solution.
- Non-blockchain DLTs – Move away from blockchain technology entirely and look for brand new solutions. This is a tough nut to crack, but Hashgraph has joined the other cryptocurrencies making big strides in this area.
A handful of cryptocurrencies are moving away from blockchain systems, creating entirely new types of DLT in the process. The most common blockchain alternative is a directed acrylic graph (DAG) system.
In simple terms, this is basically just a blockweb rather than a blockchain. By breaking free of the rigid chain system, there's room for a lot more transactions and a potentially permanent solution to the scaling problem. These systems tend to entirely remove the concepts of blocks, block size and miners from the equation, and mostly aim for instant zero fee transactions.
All the most prominent DAG systems are highly complex and entirely unique.
- Byteball Bytes – Blockless Design. The Byteball DAG removes blocks and instead connects individual transactions along a main chain that gravitates towards prominent users.
- Nano – Block Lattice. This system essentially gives each individual account its own blockchain.
- IOTA – The Tangle. This is a giant web of transactions, using a randomness-based system where every one transaction will verify two others.
Hashgraph is an entirely new DAG-style system, and it's attracting a lot of attention by claiming to achieve much better security and higher scalability than the others.
How does Hashgraph work?
Hashgraph uses a random gossip protocol. This essentially means each transaction in the network talks about itself to two randomly chosen nodes, which then pass it on to other nodes until enough nodes on the network can know exactly what's happening.
Consensus is reached when enough nodes are sufficiently informed to vote on the current state of the network and reach consensus. This entire process happens almost instantly.
This is an important element of DLT. As the word suggests, consensus refers to the entire network being in agreement on the current state of the network. So who's transferred funds where, how much money is in each connected address and everything else.
This is one of the main concepts that keeps DLTs secure. If someone can beat or trick consensus, then they might be able to control the network.
For example, if someone can beat consensus then they might be able to commit a "double spend" attack, which lets them spend the same money twice. On a poorly secured network, this might be done by attacking it with DDoS attacks, botnets or by controlling enough of the nodes to control the entire network.
How does Hashgraph compare to other DAGs?
By the numbers, Hashgraph might be one of the more impressive systems around. It's extremely fast, reportedly able to handle over 250,000 transactions per second, and has been mathematically proven to be highly secure and asynchronous byzantine fault tolerant.
Byzantine fault tolerant
Byzantine fault tolerance (BFT) is essential for cryptocurrencies. This refers to being able to mathematically prove that a network:
- Will reach consensus.
- Will know when it's reached consensus.
- Knows that enough nodes will reach the same consensus to call that consensus the truth.
The mathematical proof will need to make some assumptions about different variables. One of them is what percentage of nodes on the network are bad actors, having technical difficulties or otherwise can't be trusted. This assumption is based on factors like the number of active nodes on the network, how decentralised it is and similar.
Asynchronous byzantine fault tolerance (aBFT) assumes that there are going to be problems, and an aBFT network has been mathematically proven to still function even when there are issues.
Hashgraph is aBFT. This is a significant achievement that's eluded some other DAGs so far and is important because security has been a tough problem for DAGs to solve as their complex networks are also more fragile than blockchains in some ways. In this respect, Hashgraph might have some clear advantages over other DAG systems.
However, it had to make certain sacrifices in order to achieve this, says Yaoqi Jia, head of technology at Zilliqa.
It's important to note that Hashgraph is operated by a for-profit company called Swirlds, and it decides who gets to be a node. This naturally makes it much easier to make certain assumptions about network security. For example, it can safely assume that no hostile entity will generate multiple nodes in an attempt to influence the network and so doesn't need to bother securing against this. Mathematically defending against these so-called "Sybil" attacks is a major drain for public cryptocurrency networks, and not needing to worry about that might go a long way towards explaining Hashgraph's exceptional speed.
Swirlds owns the patent for hashgraph and uses the corresponding license in two distinct ways
1) we sell hashgraph to enterprises to run in permissioned ledgers, eg a bunch of credit unions, or a supply chain etc. This is like lots of other enterprise software
once we sell the license to an enterprise , we have zero ability to influence who they give the software to for running a node.
2) we grant to the Hedera council the sole & irrevocable license to build a public ledger based on hashgraph.
Swirlds in involved in bootstrapping the Hedera ledger and as such have been involved in picking the initial members of the Hedera governing council. We also expect that it will be these 39 enterprises running the nodes in the early days . In this sense, Swirlds has some early influence in who will run the nodes on the public ledger
But if one entity does manage to get enough nodes approved by Swirlds, or if someone manages to take control of enough already-approved nodes, they might be able to hijack the entire network. Plus, Swirlds as a for-profit business has absolute control over the network which causes some major trust issues. This is probably a deal-breaker for any company that's after a very long term DLT solution.
See above, for permissioned networks we have zero ability to control nodes. For the public Hedera network, we have some influence in early days (but no more than the other 39 governing members) and in the long run, will have zero influence because anybody can run nodes
And at this stage, its software is also closed source, which makes it difficult to verify some of its claims, and raises the likelihood of there being other vulnerabilities that Swirlds hasn't discovered. This arrangement also means it's going to be relatively centralised, with dozens or hundreds of different nodes rather than the potential millions or billions that public DAG networks like IOTA and Nano are aiming for.
The node software will be open review, allowing anybody to download the code, and check for any such vulnerabilities.
Consensus will absolutely NOT be centralized. We expect/want to have millions of nodes, with none of them having any undue influence.
In contrast to all of this, Byteball, Nano and IOTA are public, open source and largely decentralised systems. Although IOTA still has a central coordinator to give the network "training wheels" until it has enough nodes to achieve better throughput and security.
I think you are conflating two things. We can talk about centralization in both the actual consensus, and almost completely orthogonally, in the governance and licensing model.
IOTA is very centralized at the moment with, as you say, a coordinator (and a somewhat vague roadmap on getting rid of that coordinator). I submit you gave them too easy a pass here :-)
DPOS systems like EOS & Cardano are somewhat centralized in consensus, with specialized delegates being given the right to produce blocks
I will posit that hashgraph is the most decentralized in consensus of any system
Separately, we can talk about the governance & licensing
IOTA et al are open source, like BTC etc. And the market is incredibly fragmented because of the trivial ease with which anybody can launch a new coin or ledger.
Some are arguing the value of fully decentralized governance through an on-chain model (like Tezos) where anybody with stake can have a vote in deciding how the ledger will evolve. Possibly interesting but lots of potential issues, ie Ive seen reports that voter participation is extremely low, eg less than 10%. Remains to be seen how viable this is
The Hedera governance model is somewhat centralized, with 39 different known & credible enterprises given the right & responsibility to govern.
We look at the current chaos & volatility of the public ledgers & coins, with constant & rampant forks & clones - and think its time for some maturity & expertise to be brought to governance
Overall, despite the technical similarities, Hashgraph is functionally extremely different from other DAG cryptocurrencies. By the numbers, it's extremely good, but it's also fragile and can only operate in a very sterile environment. Dropping Hashgraph into the same environment as Byteball, Nano or IOTA would be like dropping a goldfish into the ocean, with a similar impact on its life expectancy.
an unfair characterization. hashgraph is very non-fragile & resilient. Ask IOTA how resilient they are to the coordinator being DDOSed.
In a public ledger, we will use proof of stake to inhibit
"If and when used in a public setting, Hashgraph will face the same issues that other public blockchains are facing today and may not be able to maintain its security and performance," Jia concluded.
of course, we know we have to prove ourselves.
These downsides won't be an issue for many applications, and companies who are happy to enter into a business relationship with Swirlds and purchase the use of Hashgraph as a service on which to create high speed DLT applications. For example, Intiva Health is using it to create a system to more efficiently manage the credentials of medical practitioners.
But other brands can't risk putting all their eggs in one basket by entrusting critical systems to a third party, especially for life and death products, and will need to use public DLTs instead, which aren't quite ready for prime time yet. For example, Volkswagen is a key partner of the IOTA project and aims to one day use the Tangle for control of autonomous vehicles.
Disclosure: At the time of writing the author holds ETH, IOTA, ICX, VEN, XLM, BTC, XRB