Date of Award


Publication Type


Degree Name



Computer Science


Blockchain, Chaos engineering, Consensus algorithms, Performance measurement, Stochastic modelling, Distributed ledger technology


W. Anderson


B. Boufama




Over the past decade or so, blockchain and distributed ledger technology (DLT) have steadily made their way into the mainstream media. As a result, new blockchain platforms and protocols are emerging rapidly. However, the performance of the resultant systems, and their resilience in hostile network environments is as yet not clearly understood. This thesis proposes a methodology to compare these platforms (specifically permissioned platforms) - and analyze the role of consensus protocols in determining system performance. It studies system performance in the face of network faults and varying loads, and also provides a qualitative analysis of each shortlisted platform. The four platforms - Ethereum, Hyperledger Fabric, Hyperledger Sawtooth, and Cosmos-SDK - are shortlisted on the basis of the consensus protocols they offer, i.e. Clique, Raft, PBFT, and Tendermint respectively. The following chapters discuss our selection criteria, the performance metrics used for comparison, and the steps followed to build a blockchain application on each platform. Considering the prominence of modelling techniques in the existing literature, we build stochastic models for each shortlisted protocol, and measure the same performance metrics as in our applications. Ultimately, this research aims to determine what factors affect the performance of blockchain systems, and what is the best way to measure their performance characteristics - by building applications or by building stochastic models? The experiments show that both methods of performance measurement have their pros and cons. They also highlight the importance of platform architecture in the determination of system performance. Selecting consensus protocols and blockchain platforms are critical decisions for any blockchain system. However, different choices shine in different settings. To recognise the best choice for a given use-case, it is crucial to first compare the protocols - and this thesis does that on the basis of performance.