A low overhead self-adaptation technique for KPN applications on NoC-based MPSoCs

MELONI, PAOLO;TUVERI, GIUSEPPE
2013-01-01

Abstract

Self-adaptive systems are able to adapt themselves to mutating internal/external conditions so as to meet their goals. One of the challenges to be tackled when designing such systems is the overhead introduced in making the system monitorable and adaptable. A large overhead can easily compensate the benefits of adaptation. In this work, we are addressing this challenge within the context of KPN applications on NoCbased MPSoCs. In particular, parametric adaptations at the application level are considered. We present a low overhead technique for the implementation of the monitor-controller-adapter loop, which is present in selfadaptive systems. The technique is fundamentally based on an extended network interface which provides the ability to interrupt remote tiles on a NoC-based multiprocessor platform. Results from the MJPEG case study show that the proposed interrupt-based approach incurs an overhead as low as 0.4% without compromising the quality of the adaptation control. Our new technique provides an improvement of approximately 6.25% compared to another state-of-the-art technique that interacts with the application using KPN semantics (i.e., blocking channels). Moreover, the sensitivity of the overhead to the complexity of the adaptation controller is much lower in case of our interrupt-based technique as compared to the blocking channel based scheme.
2013
3rd International Conference on Pervasive Embedded Computing and Communication Systems, PECCS 2013;
9789898565433
IEEE
Dresden, Germany
262
269
8
3rd International Conference on Pervasive Embedded Computing and Communication Systems, PECCS 2013
contributo
Nessuno
February 2013
Barcelona, Spain
internazionale
4 Contributo in Atti di Convegno (Proceeding)::4.1 Contributo in Atti di convegno
Derin, O; Ramankutty, P; Meloni, Paolo; Tuveri, Giuseppe
273
4
4.1 Contributo in Atti di convegno
none
info:eu-repo/semantics/conferenceObject
Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Questionnaire and social

Share on:
Impostazioni cookie