Accelerating and Scaling Blockchains via Responsibility Decomposition
Abstract:
Blockchains are rapidly being adopted by mainstream applications and users. To meet the demands of growing workloads, blockchains must be both efficient and scalable so that they can support large and diverse participant sets, which are essential for resilience against correlated failures. In a blockchain system, a set of nodes reaches consensus on transaction order and replicates execution over shared state. To store the large volumes of data associated with blockchain history and application assets, blockchains rely on decentralized storage networks (DSNs) that use peer-contributed storage resources. In this thesis, we identify three key challenges in improving blockchain efficiency and scalability. First, blockchains must achieve consensus with both high throughput and low latency. Second, blockchains must keep state storage overhead affordable, which requires scalable state storage across nodes. Third, DSNs must be able to simultaneously scale up storage volume and fault tolerance as the number of nodes increases. We review prior work on these topics and highlight the limitations of existing approaches.
We propose the principle of responsibility decomposition: separate blockchain responsibilities and engineer specialized services, each responsible for one function. Our key observation is that current blockchains rely on mutually untrusted nodes that jointly handle all responsibilities. By extracting selected responsibilities from nodes and assigning them to specialized services, each service can be optimized independently, improving overall blockchain performance. Following this separation-of-concerns approach, we design a series of novel services that enhance blockchain efficiency and scalability.
Our first system, NeoBFT, delegates consistent transaction ordering from nodes to the network. We introduce authenticated ordered multicast, a primitive that delivers messages to multiple receivers in a common order while enabling authenticity verification. By leveraging network-level ordering, NeoBFT achieves both high throughput and low latency. Our second system, BigBFT, adopts a state-execution decoupling architecture in which a distributed storage layer maintains blockchain state on behalf of all nodes. In BigBFT, this storage layer scales independently and enables each node to execute transactions without storing full state locally, thereby preserving the security model while decoupling state storage from execution. Our third system, Entropy, is a DSN that uses verifiable-randomness-based sampling as the service responsible for data placement. This service enables node-centric data placement, which is key to achieving both scalable storage volume and scalable fault tolerance.
These systems demonstrate the effectiveness of our principle: keep blockchain nodes focused on application replication and offload additional responsibilities to specialized services that are better suited to handle them. By decomposing responsibilities across dedicated services, we improve blockchain efficiency and scalability while preserving fault tolerance guarantees.