Scheduling across Multiple Applications using Task-Based Programming Models

Task-based programming models have shown their potential for efficiency and scalability in parallel and distributed systems. With such a model, a parallel application is broken down into a graph of tasks, which are subsequently scheduled for execution. Recently, implementations of task-based models...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2020 IEEE/ACM Fourth Annual Workshop on Emerging Parallel and Distributed Runtime Systems and Middleware (IPDRM) s. 1 - 8
Hlavní autoři: Chung, Minh Thanh, Weidendorfer, Josef, Samfass, Philipp, Fuerlinger, Karl, Kranzlmuller, Dieter
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.11.2020
Témata:
cooperative task scheduling
coordinated scheduling
distributed memory systems
distributed system
Libraries
parallel application
parallel distributed systems
parallel programming
Programming
reactive tasking
resource allocation
resource management
Runtime
scheduling
scheduling tasks
Synchronization
task analysis
task migration
task-based framework
task-based models
task-based programming model
task-based programming models
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Task-based programming models have shown their potential for efficiency and scalability in parallel and distributed systems. With such a model, a parallel application is broken down into a graph of tasks, which are subsequently scheduled for execution. Recently, implementations of task-based models have addressed distributed memory and heterogeneous systems with accelerators. However, the problem of scheduling tasks as well as allocating resources at runtime is still a challenge. In this paper, we propose coordinated and cooperative task scheduling across multiple applications. The main idea is to exploit the application's idle time e.g. from imbalance to serve tasks from another application. The experiments use Chameleon, a task-based framework for reactive tasking in distributed memory systems. In various example scenarios, we show improvements in CPU utilization of 5% - 15% by coordinated scheduling.
DOI:10.1109/IPDRM51949.2020.00005