A Portable Compiler-Runtime Approach for Scalability Prediction

• A portable compiler-runtime framework based on the SYCL/Celerity programming model, which is capable of extracting static information from kernels as well as dynamic information from the distributed runtime system. • A machine learning-based model operating on static and runtime feature representations, which predicts the scalability of single-task applications. The proposed methodology is abstract to the hardware and is able to predict the scalability for both CPU and GPU clusters. • An extended scalability prediction model for multi-task applications, which uses the computational DAG provided by Celerity’s runtime system and the provided range mappers to carefully model the data movement. • An experimental evaluation demonstrating portability on two large-scale systems, JUWELS and Marconi-100 (equipped respectively with multi-core CPUs and GPUs), and including error model analyses, static and dynamic feature evaluation, and code correlation study explaining how the model automatically learns from the training data to predict the scalability of unseen applications. Highly scalable parallel applications can efficiently solve expensive computational problems when run on a large number of compute nodes. However, selecting the optimal number of nodes for a compute job of a given size is non-trivial, and allocating too few or too many nodes may not yield the expected performance. Knowing the scaling behavior of an application in advance enables us, for example, to make optimal use of the available hardware resources. We introduce a novel, portable approach to predict the scalability of parallel applications written in modern high-level programming models. We propose a predictive compiler-runtime framework based on Celerity, a task-based distributed runtime system that enables executing SYCL codes on clusters. The framework targets a broad range of computing systems, from CPU to GPU clusters, and proposes a model that combines machine learning, communication modeling and DAG heuristics. Experimental results on two large-scale clusters, JUWELS and Marconi-100, show accurate scalability prediction of unseen single and multi-task applications.