Blockchain technology has now been widely used in a variety of distributed applications such as cryptocurrencies, insurance, healthcare, and supply chains. In this thesis, we study two types of blockchain systems depending on the number of blockchain involved, and we focus on characterizing and optimizing their respective performances. First, we consider blockchain systems that utilize a single blockchain, whose performance usually suffers due to serialized transaction confirmations and long consensus latency. Sharding is commonly used to enhance the performance by parallelizing transaction confirmations. However, existing sharding solutions incur a significant number of cross-shard validations, which may offset the performance improvement. Payment channel networks (PCNs) utilizing off-...[ Read more ]

Blockchain technology has now been widely used in a variety of distributed applications such as cryptocurrencies, insurance, healthcare, and supply chains. In this thesis, we study two types of blockchain systems depending on the number of blockchain involved, and we focus on characterizing and optimizing their respective performances. First, we consider blockchain systems that utilize a single blockchain, whose performance usually suffers due to serialized transaction confirmations and long consensus latency. Sharding is commonly used to enhance the performance by parallelizing transaction confirmations. However, existing sharding solutions incur a significant number of cross-shard validations, which may offset the performance improvement. Payment channel networks (PCNs) utilizing off-chain for financial transactions, enable constant transaction confirmations. Yet its fundamental performance characterization is largely unknown. Second, we examine blockchain systems that involves multiple heterogeneous blockchains, where transactions are confirmed across different blockchains, referred as cross-chain transactions. A third-party blockchain such as a relay chain can help to facilitate cross-chain confirmations. However, existing solutions cannot ensure confidentiality and atomicity at the same time while also largely ignoring the existence of failures when reading or writing data across blockchains (i.e., r/w failures). Moreover, the third-party blockchain may well become a performance bottleneck during cross-chain confirmations. We collectively address the above research problems in this thesis.

The first work focuses on improving the degree of parallelism in sharded blockchain systems by eliminating cross-shard validations. This turns out to be challenging due to the transaction dependency and imbalanced sharding distribution in term of the number of transactions in each shard. Such transaction dependencies affect cross-shard validations and lead to imbalanced sharding distribution that reduces the confirmation parallelization. We derive a innovative sharding formation by capturing the inherent relationship between transactions and smart contracts. We propose an inter-shard merging algorithm and an intra-shard transaction selection mechanism to minimize the imbalanced sharding distribution.

The second work characterizes the fundamental performance limits in payment channel networks (PCNs). We develop a novel mathematical model capturing the PCN performance, and we study the impact from a number of key factors including channel capacity and type of transactions. Through rigorous mathematical analysis, we derive the gap between the theoretically optimal performance and the performance achievable in practice, which helps to characterize the design space in PCNs for scheduling transactions.

The third work addresses the atomicity and confidentiality for cross-chain confirmations under different failures. When failures occur during reading or writing data, we consider two scenarios depending on whether data is the latest or not. We design a four-phase-commit protocol (4pc) and a smart contract-based solution (SSC) respectively to ensure atomicity and confidentiality under different scenarios.

The fourth work aims to improve the cross-chain confirmation performance by optimizing the throughput of a relay chain through sharding techniques. To eliminate cross-shard validations, we capture transaction dependencies on the relay chain and place transactions with dependency into a single shard. To ensure a balanced sharding distribution, we mathematically formulate the transaction distribution problem into an integer optimization problem and design an efficient solution.