Today, we are excited to launch Celer Network out of stealth!
TL;DR: Celer Network is the first off-chain operating network with coherent technology and economic architecture to bring Internet-level scalability to blockchains. Celer Network is horizontally scalable to billions of off-chain transactions per second, trust-free, decentralized and private. It encompasses a layered architecture with significant technical innovations for each layer:
- a channel construct suite with flexible support for generalized off-chain dApp state transitions beyond simple payments, sidechain-like channels with minimal fund lock up, and more;
- first-ever provably optimal state routing algorithm with 15X higher transaction throughput than state-of-the-art solutions;
- an off-chain operating system that drastically simplifies the development and usage of off-chain applications on various platforms.
In addition, Celer Network proposes a principled off-chain crypto economics design to balance tradeoffs made to achieve scalability.
In this blog, we focus on Celer Network’s technology stack. We will put out a separate blog dedicating to Celer Network’s crypto economic mechanism design.
Celer Network’s Layered Technology Architecture
Inspired by the layered abstraction principle that made the Internet hyper-successful, we design Celer Network as a cleanly layered architecture that decouples sophisticated off-chain operating network into hierarchical modules. This architecture greatly reduces complexities in the system design, development, and maintenance, as well as enables rapid evolution of each individual component. We call the collective specification of each layer’s interfaces as cStack, which consists cChannel: generalized state channel and sidechain suite; cRoute: provably optimal state routing protocol; cOS: development framework and runtime for off-chain enabled applications; and cApps: application ecosystem running on Celer Network. Below we highlight the technical challenges and features of cChannel, cRoute, and cOS.
cChannel: Flexible, Generalized and Out-of-Box Features
cChannel is the lowest layer of Celer Network interacting with different blockchains. It provides to the upper layer the common abstraction of up-to-date states and bounded-time finality. cChannel with two integrated components of on-chain contracts and off-chain protocols has the following key features:
- Generalized dApp support beyond simple payment. cChannel enables generalized conditional state dependency DAG on top of networked state channels and side chains. This makes Celer capable of supporting not only simple payment but also a large category of applications such as gaming, online auction, insurance, prediction market and decentralized exchanges. There are awesome concurrent research works on this front in Ethereum community from the pioneering L4 Research team, recently Perun Network and other teams. They are working towards similar generalized state channel design and implementation as Celer does in this part of cChannel. We believe it would be extremely beneficial for all Ethereum state channel researchers to work together and define some open standard for generalized state channel core abstraction and interfaces.
- Multi-hop state relay. All off-chain applications can ride on the shared underlying state channel and sidechain networks concurrently with multi-hop state relay support. cChannel extends Sprite to support atomic time lock not only on payment relay but also on generalized state relay.
- Out-of-box conditional payments. cChannel contract suite also comes with a rich set of “pre-compiled” features to facilitate common off-chain application patterns such as boolean circuit conditional payment (e.g. Alice pay Bob if Carl wins the game), conditional state settlement (e.g.pay Bob X/N eth if Carl wins and N is the total number of moves Carl made to win) and more.
- Reduced liquidity locking with “sidechain-like” channels. cChannel goes beyond the scope of traditional state channel by introducing a special kind of sidechain channel as an option to reduce liquidity lock up to the level of fraud proof bond. These sidechains are further connected as a network.
- Dependency DAG traversal off-chain protocols: cChannel defines off-chain state transition, that models any off-chain application state updates, as topological traversals (either forward or reverse) of conditional state dependency DAG. On each traversal, dependency edges are created, updated or removed. Communication protocols are further specified to implement this abstract state transition model.
- Combined formal specification and verification of on-chain contracts and off-chain protocol is an integral part of Celer Network’s technology stack.
- We planned to elaborate many other features such as dynamic deposit/withdrawal, cross-channel deposit swap, on-chain settle congestion avoidance in future posts.
cRoute: First-ever Provably Optimal State Routing
The need for state routing (“payment routing” in the case of simple unconditional payments) is apparent: it is impractical to establish direct state channels between every pair of nodes due to channel opening costs and deposit liquidity lockup. Therefore, it is necessary to build a network consisting of cChannels, where state transitions should be relayed in a trust-free manner. State routing is crucial for off-chain payment networks because it directly determines the performance of the network, which is how fast and how many transactions can flow on a given network. Existing solutions all essentially come down to “shortest path routing”. This principle makes sense in traditional computer networks because their link capacity is stateless (not affected by past transmissions) and their topology is relatively stable. However, the link capacity in a payment network is stateful (i.e. determined by past payments), which leads to a highly dynamic network where link capacity and topology are constantly changing. This key difference makes the shortest path routing algorithms hard to converge, and thus yields low throughput, long delay, and even outages.
In Celer Network we propose cRoute, the first-ever routing algorithm specially engineered for the off-chain operating network. We highlight some key features of cRoute:
- Provably optimal. cRoute is provably optimal in the sense that it can support all feasible payment processes on a certain state channel network topology with bounded delay. In our simulation, it can achieve 15X higher transaction throughput (close to optimal) as compared to algorithms built upon the shortest path principles (the principle used in Raiden and Lightning’s routing algorithms).
- Transparent channel balancing. “Keeping channels balanced for high throughput” has been an intuition since the initial proposal of Lightning Network. However, existing attempts are all heuristics that require heavy on-chain or off-chain coordination with no optimal balancing guarantees. cRoute’s algorithm transparently embeds the channel balancing process along with routing and constantly balances the entire network without requiring any additional coordination.
- High failure resilience. cRoute is designed to be resilient to failures of peering nodes because of its adaptive and multi-path nature. It automatically breaks down payments to avoid depleted channels and ensure “data-plane” connectivity via distributed link reversal algorithms.
- Fully decentralized. With all these features, there is still no need to have a path finding services. Peers would only need to talk to its neighbors in the state channel network topology.
cOS: Friendly Developer and User Experience for Off-chain dApps
An On-chain dApp is relatively easy to build: it has a frontend connecting to a simple uniform data+compute backend (the blockchain). Off-chain dApps, though with great potentials for high scalability, are much harder to build and use. Just to name a few challenges: a. juggling out the correct state dependency between arbitrary off-chain and on-chain states; b. developing communication protocols for dependency DAG traversal for every dApp; c. tracking states and performing resolution and state dispute; d. handling payment failures transparently; e. handling multiple concurrently running off-chain dApps. The list goes on.
cOS in Celer Network targets exactly those challenges and provides both a development framework and runtime for scalable off-chain dApps. At development time, cOS provides the development framework for developers with common design patterns that capture the conditional dependency Directed Acyclic Graph (DAG) of states. At runtime, cOS handles the storage and tracking of off-chain states, constructs and relays conditional state transitions and automatically settles and disputes states with counter-parties.
The ambitious scope of Celer Network requires a full stack and highly diverse expertise in distributed systems, networking, formal verification, game theory, mobile OS, and smart contracts. We are happy to introduce our core developers who have strong academic and engineering track records and deep connections to talents in exactly those areas.
Dr. Mo Dong received his Ph.D. from UIUC. His research focuses on learning based networking protocol design, distributed systems, formal verification and Game Theory. Dr. Dong led project revolutionizing Internet TCP and improved cross-continental data transfer speed by 10X to 100X with non-regret learning algorithms. His work was published in top conferences, won Internet2 Innovative Application Award and being adopted by major Internet content and service providers. Dr. Dong was a founding engineer and product manager at Veriflow, a startup specializes in network formal verification. The formal verification algorithms he developed is protecting networking security for fortune 50 companies. Dr. Dong is also experienced in applying Algorithmic Game Theory, especially auction theory, to computer system protocol designs. He has been teaching full-stack smart contract courses. He produces technical blogs and videos on blockchain with over 7000 subscribers.
Dr. Junda Liu received his Ph.D. from UC Berkeley, advised by Prof. Scott Shenker. He was the first to propose and develop DAG based routing to achieve nanosecond network recovery (1000x improvement over state of art). Dr. Liu joined Google in 2011 to apply his pioneer research to Google’s global infrastructure. As the tech lead, he developed a dynamic datacenter topology capable of 1000 terabit/s bisection bandwidth and interconnecting more than 1 million nodes. In 2014, Dr. Liu became a founding member of Project Fi (Google’s innovative mobile service). He was the tech lead for seamless carrier switching, and oversaw Fi from a concept to a $100M+/year business within 2 years. He was also the Android Tech Lead for carrier services, which run on more than 1.5B devices. Dr. Liu holds 6 US patents and published numerous papers in top conferences. He received BS and MS from Tsinghua University.
Dr. Xiaozhou Li received his Ph.D. from Princeton University and is broadly interested in distributed systems, networking, storage, and data management research. He publishes at top venues including SOSP, NSDI, FAST, SIGMOD, EuroSys, CoNEXT, ToN, and won the NSDI’18 best paper award for building a distributed coordination service with multi-billion QPS throughput and ten microseconds latency. Xiaozhou specializes in developing scalable algorithms and protocols that achieve high performance at low cost, some of which have become core components of a few widely deployed systems such as Google TensorFlow machine learning platform and Intel DPDK packet processing framework. Xiaozhou worked at Barefoot Networks, a startup company designing the world’s fastest and most programmable networks, where he led several groundbreaking projects, drove technical engagement with key customers, and filed six U.S. patents.
Dr. Qingkai Liang received his Ph.D. degree from MIT in the field of distributed systems, specializing in optimal network control algorithms in adversarial environments. He first-authored over 15 top-tier papers and invented 5 high-performance and highly-robust adversarial resistant routing algorithms that have been successfully applied in the industry such as in Raytheon BBN Technologies and Bell Labs. He was the recipient of Best Paper Nominee at IEEE MASCOTS 2017 and Best-in-Session Presentation Award at IEEE INFOCOM 2016.
About why the name Celer.
“As for c, that is the speed of light in vacuum, and if you ask why c, the answer is that it is the initial letter of celeritas, the Latin word meaning speed.”
— — Isaac Asimov in “C for Celeritas (1959)”
This blog reads long, but this is just the tip of an iceberg.
Please stay tuned for our cOS alpha release and more detailed ELI5 and technical blogs on each technical and crypto economics front in the coming weeks. Please subscribe to the following channels to stay tuned: