Concurrent Scheduling of High-Level Parallel Programs on Multi-GPU Systems

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Knorr, Fabian, Salzmann, Philip, Thoman, Peter, Fahringer, Thomas
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910874161119232
author Knorr, Fabian
Salzmann, Philip
Thoman, Peter
Fahringer, Thomas
author_facet Knorr, Fabian
Salzmann, Philip
Thoman, Peter
Fahringer, Thomas
contents Parallel programming models can encourage performance portability by moving the responsibility for work assignment and data distribution from the programmer to a runtime system. However, analyzing the resulting implicit memory allocations, coherence operations and their interdependencies can quickly introduce delays into the latency-sensitive execution pipeline of a distributed-memory application. In this paper, we show how graph-based intermediate representations help moving such scheduling work out of the critical path. In the context of SYCL programs distributed onto accelerator clusters, we introduce the instruction graph, a low-level representation that preserves full concurrency between memory management, data transfers, MPI peer-to-peer communication and kernel invocation. Through integration within the Celerity runtime, we demonstrate how instruction-graph scheduling enables a system architecture that performs this analysis concurrently with execution. Using a scheduler lookahead mechanism, we further detect changing access patterns to optimize memory allocation in the presence of virtualized buffers. We show the effectiveness of our method through strong-scaling benchmarks with multiple Celerity applications on up to 128 GPUs in a production cluster.
format Preprint
id arxiv_https___arxiv_org_abs_2503_10516
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Concurrent Scheduling of High-Level Parallel Programs on Multi-GPU Systems
Knorr, Fabian
Salzmann, Philip
Thoman, Peter
Fahringer, Thomas
Distributed, Parallel, and Cluster Computing
Parallel programming models can encourage performance portability by moving the responsibility for work assignment and data distribution from the programmer to a runtime system. However, analyzing the resulting implicit memory allocations, coherence operations and their interdependencies can quickly introduce delays into the latency-sensitive execution pipeline of a distributed-memory application. In this paper, we show how graph-based intermediate representations help moving such scheduling work out of the critical path. In the context of SYCL programs distributed onto accelerator clusters, we introduce the instruction graph, a low-level representation that preserves full concurrency between memory management, data transfers, MPI peer-to-peer communication and kernel invocation. Through integration within the Celerity runtime, we demonstrate how instruction-graph scheduling enables a system architecture that performs this analysis concurrently with execution. Using a scheduler lookahead mechanism, we further detect changing access patterns to optimize memory allocation in the presence of virtualized buffers. We show the effectiveness of our method through strong-scaling benchmarks with multiple Celerity applications on up to 128 GPUs in a production cluster.
title Concurrent Scheduling of High-Level Parallel Programs on Multi-GPU Systems
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2503.10516