In this thesis, we investigate heterogeneous multicore architectures and their ability to process multimedia workloads. We first conducted several case studies with simple multimedia workloads, where we only optimized for a single architecture. Next, we investigated a complex multimedia workload with real-time requirements optimized for multiple heterogeneous architectures.
Using all the knowledge gained from our studies of simple and complex workloads, we designed and proposed a framework for processing real time multimedia workloads on heterogeneous architectures, called P2G. P2G use the workload's data, task, and pipeline parallelism to optimize the granularity of the programs, either at compilation time or at runtime. The P2G is designed to support distributed processing and uses a two-level scheduling approach. The fundamental ideas of the P2G were implemented in a prototype running on an x86 multicore architecture. We showed that we are able to express multimedia workloads in the P2G programming model and to scale performance when more processing resources are added to the system.
The thesis is written within the field of networks and distributed systems. The work has been conducted at Simula Research Laboratory and University of Oslo.
Prior to the defense, at 11:15, Håkon Kvale Stensland presented his trial lecture Introduction to compilation techniques for parallel architectures.
The adjudication committee
- Professor László Böszörményi, Institute of Information Technology, University of Klagenfurt
- Professor Shervin Shirmohammadi, School of EECS, Ottawa, Ontario, Canada
- Professor Xing Cai, Department of Informatics, University of Oslo
Chair of the disputation
- Dag Langmyhr, Department of Informatics, University of Oslo
- Professor Carsten Griwodz, Simula Research Laboratory
- Professor Pål Halvorsen, Simula Research Laboratory
Announcement of the PhD defense at the University of Oslo's web pages (in Norwegian).