Embedded Adaptive Streaming sYstems
Project Manager: Prof. dr. ir. Gerard Smit
Faculty of Electrical Engineering, Mathematics and Computer Science – EEMCS
Tel.: +31 53 489 3734
Streaming applications (e.g. for multi-media, medical image processing, wireless communication) are quickly becoming more and more important. They are often realized in the embedded systems domain and are subject to severe constraints concerning non-functional requirements such as time and energy consumption. A crucial element in these developments is to map streaming applications to possibily heterogeneous multi-processor platforms which may consist of hundreds or even thousands of specialized processors. A major challenge is to make the resulting systems adaptive such that they are able to react automatically to changing circumstances. To master that challenge, a necessary precondition is that the mapping of an application to a multi-processor platform is flexible in the sense that it can be changed at run-time.
Current programming paradigms are not well equipped to include non-functional requirements or mapping directives in the design of a streaming application. Hence, they do not support a flexible mapping mechanism.
This proposal aims at enlarging the adaptivity of streaming applications by developing means to include the aforementioned elements in the specification of a streaming application. In particular, the proposed research will
- use Kahn Process Networks (KPNs) as the model of computation for specifying streaming applications, thus building upon and exploiting the extensive prior work on KPNs,
- augment KPNs with language constructs to express non-functional requirements of an application,
- augment KPNs with directives to map applications to multi-processor systems in a flexible way such that adaptivity is supported while respecting resource boundedness,
- develop new metrics to compare the quality of the results of different mappings.
Project duration: April 2008 – April 2012
Project budget: 372 k-€
Number of person/years: 2.4 fte
Project Coordinator: UT
Participants: UT, University of Amsterdam
Project budget CTIT: 186 k-€
Number of person/years CTIT: 1.2 fte
Involved groups: Computer Architecture for Embedded Systems (CAES)