Cross-Processor Execution-Time Prediction Dataset for Heterogeneous Computing Platforms
Year: 2026
Type: article
Abstract: This dataset is designed for research on execution-time prediction across heterogeneous computing platforms. It contains execution measurements collected from multiple computing platforms with different processor configurations and resource allocations. The dataset represents the execution behaviour of benchmark applications under systematically varied hardware and resource conditions. The data were generated by executing a set of benchmark applications on different computing platforms while varying the available computational resources, including processor operating conditions, the number of allocated CPU cores, and the amount of allocated memory. For each execution, the workload characteristics and the corresponding processor configuration were recorded together with the measured execution time. This enables the dataset to capture the relationship between application workload, available computing resources, and execution performance. The dataset consists of multiple CSV files corresponding to different portions of the collected experimental data. These files can be combined to form the complete dataset. In our study, the data from demo0.csv to demo6.csv were combined to construct the training dataset, while demo7.csv was reserved as an independent external test dataset. The external test data were kept separate from model development to evaluate the generalisation capability of execution-time prediction models on previously unseen samples. The dataset can be used for machine-learning-based execution-time prediction, cross-processor performance modelling, heterogeneous computing research, resource-aware scheduling, task offloading, and performance analysis. It is particularly suitable for investigating whether execution behaviour observed under one computing configuration can be used to predict execution time under another target configuration.
Institution: SASTRA University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7208818344