Predicting Lakehouse Performance in Clouds: An Empirical Exploration of Query Runtime Variance
2026-06-02 • Distributed, Parallel, and Cluster Computing
Distributed, Parallel, and Cluster Computing
AI summaryⓘ
The authors studied how the time it takes to run the same data analysis queries on cloud-based lakehouse systems can change a lot each time. They found that the runtime can differ nearly two times between runs, which makes predicting query speed tricky. They looked into reasons for this variation, like where data is stored and how busy the system is, and showed that fixing these issues can make predictions much more accurate. Their work also shows that better predictions can help reduce carbon emissions when scheduling workloads. Overall, the authors highlight the importance of understanding and managing runtime variability in these systems.
Query Performance PredictionLakehouse SystemsRuntime VarianceDistributed AnalyticsKubernetesData LocalityCo-tenant LoadCachingAdaptive Resource ScalingLow-carbon Scheduling
Authors
James Nurdin, Wei Liu, Richard Mccreadie, Lauritz Thamsen
Abstract
Data analytics increasingly runs on distributed lakehouse systems, where platform operators must optimise monetary, resource, and environmental costs. Query Performance Prediction (QPP) helps to balance these costs and supports workload management techniques, such as adaptive resource scaling and low-carbon scheduling. However, runtimes in lakehouses can vary substantially, and the impact of runtime variance on QPP accuracy and workload orchestration has not previously been systematically studied for lakehouse systems. This paper addresses this gap by investigating the runtime variance observed for distributed lakehouse analytical queries and its impact on QPP. First, we quantify the run-to-run variance using Kubernetes deployments across three public clouds and one private cloud, spanning multiple database scales and three analytical benchmarks. Our results demonstrate that repeated executions of the same query can vary in runtime by nearly twofold. Second, we conduct a factor analysis study assessing key sources of this runtime variance such as data locality, co-tenant load, and caching effects. Third, we examine how variance influences state-of-the-art QPP models, revealing that addressing key sources of variance can reduce prediction error up to 80%. Finally, we demonstrate the downstream implications for low-carbon scheduling as an example of a workload management technique that relies on performance prediction, showing that accounting for runtime variance can lead to a significant reduction in carbon costs.