# The Blockchain Execution Dilemma: Optimizing Revenue XOR Fair Ordering

Crypto.world

> Modern blockchains have separated consensus from execution, which moves the performance bottleneck to the execution layer and inserts a sequencing step that decides which transactions go to which…

**arXiv**: 2604.23266

**Published**: 2026-04-25

**Authors**: Artjom Pugatsov, Can Umut Ileri, Jérémie Decouchant

## AI summary

## Overview

**Research area:** Blockchain execution-layer design — transaction sequencing between consensus and parallel execution, with an explicit treatment of Maximal Extractable Value (MEV) and order-fairness.

**Technical level:** Advanced. The paper combines a formal scheduling model, a genetic-algorithm optimizer, and an empirical evaluation; the intuition is accessible, but the notation and scheduling constraints require familiarity with distributed-systems and combinatorial-optimization concepts.

**Scope (one sentence):** The paper formalizes transaction sequencing for "lazy" blockchains with continuous, overlapping block execution and proposes an anytime genetic sequencer that maximizes validator revenue while optionally enforcing consensus-determined fair ordering, evaluated on real Sui data and synthetic Ethereum-based workloads.

## What This Paper Is About

Modern blockchains have separated consensus from execution, which moves the performance bottleneck to the execution layer and inserts a *sequencing* step that decides which transactions go to which parallel worker and which get postponed to a later block. Today's sequencers (e.g., Sui's) order transactions greedily by gas price, which can let one resource-heavy transaction block many cheaper, parallelizable ones, while proposed fair-ordering protocols forbid reordering almost entirely. The paper asks whether a sequencer can be smarter than greedy gas-price ordering — and at what cost fairness comes — by letting the sequencer search for high-revenue orderings, with fair ordering as an optional constraint rather than a hard architectural assumption.

## Key Contributions

1. **First formalization of sequencing under continuous execution.** The authors model single-shot sequencing (deadline `d`, `m` workers, per-transaction object read/write sets, execution times, gas prices) and extend it to multi-shot continuous sequencing where successive blocks overlap, including an extended deadline, worker-busy and object-freed constraints, and a deferral cap `θ` after which a transaction is cancelled. The model is also extended to express order-fairness policies as optional execution constraints.
2. **An anytime genetic sequencer.** The algorithm optimizes validator profit (the sum of execution time times gas price over scheduled transactions) using gas prices, object sets, and predicted execution times, and returns its best intermediate schedule if it must stop early — a requirement for continuous operation. It is augmented to enforce fair ordering only for interdependent (conflicting) transactions, so independent transactions can still be reordered concurrently.
3. **Implementation and comparative evaluation.** The sequencer is compared against several heuristics (gas-price ordering as the baseline, plus randomized order, historical arrival order, and shortest-execution-time-first) on real-world Sui data and synthetic Ethereum-based workloads, including a study of perturbed execution-time estimates.
4. **First empirical evidence on the cost of strict fairness.** The paper quantifies that enforcing strict fair ordering can impose a 50% to 60% profit penalty on validators during peak congestion, suggesting rigid fair ordering may nullify the gains of advanced parallel sequencing layers.

## Main Findings

- **Revenue gain:** The genetic sequencer increases validator profit by approximately 15% (stated as "up to 16%" under sustained congestion) relative to the evaluated baselines.
- **Faster decongestion:** During sporadic transaction spikes, the approach accelerates congestion relief by up to 58%.
- **Fairness is expensive:** Fair-ordering constraints reduce validator profit by 50% to 60% during periods of high congestion, the paper's headline result and the source of its "Profitability XOR Fair Ordering" framing.
- **Greedy gas-price ordering is structurally suboptimal:** A single highly congestive transaction can be scheduled first and defer many smaller transactions (the authors' Figure 4 example defers transactions 2–5 under greedy ordering, while an alternative defers only transaction 1 and raises parallelism and gas revenue). Swapping in a different single ordering criterion does not fix this.
- **Fair ordering suppresses parallelism:** Strict fairness severely limits the reordering of independent transactions, which is precisely the mechanism parallel execution relies on.
- **Overhead profile:** On the continuous Sui workload, median per-round wall-clock time stayed under 300 ms for all heuristic sequencers, versus roughly 1.0 s for "GE10" and 4.4 s for "GE50" (the genetic configurations).
- **Prediction error matters (partially reported):** The experiments deliberately inject inaccuracies into execution-time estimates to evaluate their impact; the specific perturbation results appear in the evaluation section, which is truncated in the supplied content.

## Methodology in Plain English

The authors abstract the blockchain's middle layer into a clean optimization problem. Each transaction is described by three things: which state objects it reads and writes, how long it will take to execute, and its gas price. A fixed number of parallel workers must finish work before a deadline, conflicting transactions cannot run at the same time, and any transaction that does not fit is deferred to a later block. The sequencer's job is to choose the *order* in which the scheduler considers transactions.

Crucially, they do not try to solve the scheduling problem optimally — they keep a fast, deterministic, greedy scheduler that always places each transaction at its earliest feasible slot on a worker and even fills gaps in existing schedules, indexing those gaps in a B-tree for speed. The optimization effort goes entirely into the ordering, which the greedy scheduler then converts into a concrete assignment.

For the ordering, they use a genetic algorithm: candidate orderings are chromosomes, running the scheduler and computing revenue is the fitness function, and evolution proceeds via order crossover, insertion mutation (probability set to 60%), and elitism (a (μ+λ) survival strategy). The initial population is seeded with the gas-price ordering, so the algorithm refines a strong baseline rather than searching globally. Because it is an anytime algorithm, it can be cut off when the next block must be fed to workers and still output the best schedule found so far. Reported parameters: `pop_size = 100`, budget of 1000 or 5000 evaluations, `m = 4` workers, and `d = 62,500` for Sui versus `1,250,000` for ETH.

For continuous operation, each round inherits only the workers' busy status from the previous round, and other effects of overlap are pushed into the parameters: an extended deadline derived from the previous round's slack and a constraint that a transaction cannot start before the objects it touches are free.

For the fair-ordering variant, transactions are constrained by a dependency graph derived from the fairness order. The seed becomes a gas-price-prioritized topological traversal of that graph, crossover only picks transactions whose dependencies are met, and mutation relocates a transaction only within its neighborhood of non-conflicting transactions. The fair scheduler additionally defers everything conflicting with an already-deferred transaction.

Finally, they compare the genetic sequencer with the heuristics on Sui data and synthetic Ethereum workloads, measuring realized profit and how quickly congestion clears, rather than algorithm runtime.

## Why This Matters

**For research:** The paper reframes the MEV-versus-fairness debate as a *scheduling* problem with measurable costs. It provides a formal bridge between two literatures that usually ignore each other — parallel execution engines and order-fairness protocols — and supports the claim that fairness proposals must be evaluated against execution-layer performance, not just consensus-layer guarantees.

**Real-world applications:**

- **DEX and DeFi ordering:** quantifying the revenue forgone by exchanges or rollups that adopt arrival-order fairness, and enabling conflict-aware fairness that still permits parallel reordering.
- **High-throughput object-model chains (e.g., Sui, Solana-style pessimistic execution):** upgrading the sequencer to use execution-time and conflict information instead of a single gas-price sort.
- **Validator economics and congestion management:** converting a 15–16% revenue improvement and up to 58% faster decongestion into tangible operator incentives and better user experience during spikes.
- **Layer-2 / rollup sequencer design:** the model is deliberately blockchain-agnostic, so the same trade-off analysis applies to centralized and decentralized sequencers.

**Industry relevance:** Sequencer policy is a live design decision for chains that separate consensus from execution; the paper's central practical warning — that strict fair ordering can erase the benefits of advanced sequencing — is directly actionable for protocol designers choosing between revenue, spam resistance, and credible neutrality.

## Future Directions

- **Conflict-aware fairness that is cheaper.** The paper shows strict fairness costs 50–60% of profit at peak load; a natural follow-up is fairness schemes that constrain only genuinely conflicting transactions and quantify the residual penalty.
- **Better execution-time prediction.** Seeding and evaluation depend on runtime estimates, and the authors explicitly inject prediction error; more accurate or adaptive predictors (and robustness bounds under miscalibration) remain open.
- **Real deployment and timing.** The authors state they do not estimate real-world execution time and do not report algorithm runtime, measuring effort in sequencer iterations instead; deployment on a live or testnet system with real timing constraints is the obvious next step.
- **Broader datasets and parameter regimes.** Dataset sizes for the Sui and synthetic Ethereum workloads are not reported in the available text, and the study fixes `m = 4`; scaling to many workers, variable deadlines, and additional real chains would test generality. Comparing the genetic approach against constraint solvers or learned orderings is also left open (the authors reject constraint solving on efficiency grounds, citing Chahoki et al.).

## Target Audience

Blockchain protocol and distributed-systems researchers working on execution layers, sequencing, and MEV; validator and sequencer engineers at high-throughput or object-model chains and rollups; researchers on order-fairness and transaction-ordering fairness mechanisms; and graduate students in distributed systems or algorithmic game theory looking for a concrete, data-backed case study of a scheduling trade-off in a real system.

## Authors’ abstract

Successive consensus algorithm generations displaced the blockchain performance bottleneck to the execution layer. Recent works address this by parallelizing non-conflicting transactions. Historically, execution followed consensus-level ordering, a practice susceptible to Maximal Extractable Value (MEV) exploitation. Conversely, recent academic proposals introduce rigid fair-ordering protocols that severely restrict transaction reordering. Parallel execution frameworks…

[The full abstract on arXiv](https://arxiv.org/abs/2604.23266)

[Read the original paper](https://arxiv.org/abs/2604.23266)

---

Canonical: https://crypto.world/paper/the-blockchain-execution-dilemma-optimizing-revenue-xor-fair-ordering-2604.23266
