Efficient and Fault Tolerant Computation of Partially Idempotent Tasks
Edson Borin, Ian Liu Rodrigues, Alber T. Novo, João D. Sacramento, Maurício Breternitz, Martin Tygel · 2015
The popularization of multi-core processors on computational systems and cloud services enabled access to high performance computing infrastructure. However, programming for these systems might be cumbersome due to frequent system failures when using thousands of machines, poor load balancing and task scheduling. To solve these problems, we introduce a concept of partially idempotent tasks and discuss how their properties ease the implementation of fault tolerant mechanisms, load balancing on heterogeneous systems and dynamic provisioning of resources. Additionally, we propose a programming model, a scalable system and an API which executes these kind of tasks and we present an usage example of our system to a simple problem.