COMPARING BLOCKCHAINS: PERFORMANCE, ENERGY, AND ECONOMIC EFFICIENCIES

Blockchains are distributed ledgers that let mutually distrustful parties agree on an append-only transaction history without relying on a central authority. By combining cryptographic hashing, digital signatures, and consensus mechanisms, blockchains provide tamper evidence, auditability, and agreement among nodes. Modern blockchain systems significantly vary in consensus design (e.g., Proof of Work, Proof of Stake, Proof of Author- ity, and Byzantine Fault Tolerance mechanisms), access model (open vs. permissioned), and execution layers (from simple asset transfers to expressive smart-contract virtual machines). The related architectural choices shape decentralization, fault tolerance, and the attainable latency-throughput envelope. As blockchain deployments expand to payments, tokenization, decentralized finance, supply chain traceability, and digital identities, comparing these systems has become an urgent necessity. Unfortunately, rigorous blockchain evaluation remains difficult. On the one hand, measurements are confounded by fluctuating network conditions, heterogeneous infrastructures, and rapidly evolving software. On the other hand, results are too often collapsed to a single number (such as transactions per second) without dispersion or methodological details; economic assessments of crypto-assets lack a unified and interpretable index that captures the balance of core economic parameters and their trade-offs (usage, liquidity, stability, and security) rather than market price sentiment; and experimental studies rarely address the dimensions of experimental repeatability (same setup, same results) and performance predictability (stable expectation). The consequence is an evidence gap: how to assess and compare the efficiency of blockchains – spanning performance, energy, economics, and result stability – in different scenarios? This dissertation aims at reducing this gap with a coherent yet modular approach that combines topology-controlled benchmarking with an orthogonal, entropy-based economic analysis, delivering four contributions. First, it introduces Lilith, a system-agnostic benchmarking framework that couples workload generation with network emulation to run controlled, repeatable experiments under explicit overlay topologies (i.e., the logical peer- to-peer connectivity graphs) and link properties such as latency, bandwidth, and packet loss. Lilith orchestrates deterministic deployments (pinned artifacts, controlled boot order, and CPU core pinning and memory binding), integrates power probes, and provides a uniform client interface; this underpins a comparison based on typical performance metrics. Second, Lilith is employed to quantify blockchain energy consumption under realistic conditions. Third, Lilith is adopted for a network-controlled, multi-run measurement campaign to produce a public dataset. By combining dispersion metrics (e.g., worst-case deviation) with analysis of variance and intraclass correlation, we quantify run-to-run variability and performance predictability across blockchains, topologies, workloads, and node-set sizes. Fourth, in addition to Lilith, the dissertation introduces the Entropy Balance index (EB-index), which aggregates heterogeneous on-chain indicators into a single, interpretable score of economic efficiency. ii As for the first three contributions, we set up the experimental baseline by considering five network topologies (fat-tree, full mesh, hypercube, scale-free, torus) and five industry- grade blockchains (Algorand, Diem, Ethereum Clique, Quorum IBFT, Solana), exercised with transfer transactions and smart-contract workloads (DDoS, FIFA, GAFAM, gaming, PayPal, VISA) across two node-set sizes (10 and 40). In the performance study, the network topology emerges as the primary factor de- termining throughput and latency. Full mesh, hypercube, and torus deliver higher performance under heavy load. The performance of Algorand and Diem is stable with respect to topology changes, while Ethereum is less sensitive but remains slower. In the energy study, fat-tree and full mesh turn out to be the most energy-efficient topologies, especially at high load. Algorand and Diem exhibit the lowest energy per transaction, Ethereum Clique the highest across topologies; Quorum IBFT and Solana become costlier as workload intensity and network size increase. The experimental repeatability and performance predictability study shows low per- formance variance (transactions per second, block latency, energy consumption) for Algorand and Diem and pronounced sensitivity for Solana and Quorum IBFT, especially as workloads, node-set size, and geo-latency conditions vary. The released dataset and the accompanying analysis templates, which are based on clusters instead of public-cloud testing, enable thorough checks that go beyond point estimates by quantifying dispersion and confidence in comparative results. Finally, in the economic study, the EB-index aggregates heterogeneous on-chain indicators – such as user activity (transactions, active addresses), token distribution (balance concentration), and supply turnover/velocity – by using the normalized Shannon entropy and its weighted Beliş-Guiaşu variant. When applied to the capitalization-based leading crypto-assets Bitcoin, Ethereum, Ripple, USD Coin, Dogecoin, and Cardano, the EB-index separates volume-driven bursts from structurally balanced ecosystems and reveals differences that price, total value locked, or raw activity may blur. Overall, this dissertation delivers a topology-aware blockchain benchmarking frame- work, empirical evidence that network structure materially affects performance and energy, a public multi-run dataset together with analysis templates that promote experimental repeatability and performance predictability, and an entropy-based index for assessing economic efficiency.