Predictable matching of demands on networked architectures (PreMaDoNA) (EES.6390)

Project summary
The PROGRESS embedded systems roadmap 2002 [1] notes that its highest priority is to 'promote, develop and facilitate the reuse of IP blocks' as a major factor in an effort to reduce the design-complexity gap. We believe that the use of so-called Networks-on-Chip (NoC) [2], which are built-up from various core blocks (processors, memory, i/o, etc.) connected by an on-chip network, highly enable the reuse of IP blocks. Therefore the PreMaDoNA project takes NoCs as mapping target for multi-media applications, as design intervals reduce for these applications.

Research summary
One of the major problems when mapping applications to processing platforms like NoCs is dealing with real-time constraints; e.g., how to deal with them on an architecture that includes non-predictable elements like caches and shared buses. This problem is becoming even worse due to the increasing dynamism inside applications and due to the dynamically changing set of running applications (on a single platform); this especially holds for the video domain. Guaranteeing real-time behavior therefore requires dynamic adaptation of the video quality, without being disruptive, and still satisfying non-functional constraints, like latency and throughput constraints. As a result many design iterations are needed. This inspired PreMaDoNA's major research objective: being able to design NoC-based real-time systems in a predictable way, such that we can guarantee non-functional requirements, while being able to dynamically match quality versus available resources.
PreMaDoNA proposes a solution based on the removal and/or software control of unpredictable elements in the architectures in combination with a predictable mapping methodology that supports reasoning about throughput. PreMaDoNA is centered around three tracks, resulting in three connected research tracks: 1. The definition and modeling of a NoC architecture and its implementation. The architecture is de-fined in terms of services (APIs) performed by the system. These include both the communication and computation services that can be requested by an application. The expected results are the architecture and implementation, a resource manager, a computational model for mapping purposes, and a prototype realization on an FPGA board.
2. Methods for predictable design for systems targeted to NoCs. In order to satisfy non-functional re-quirements on e.g. latency and throughput many design iterations may be required. This track aims at reducing the number of design iterations by taking these requirements into account within the whole application modeling and mapping trajectory. Expected results are a mapping methodology supported by an analysis tool that is embedded within a complete design flow. 3. Demonstrating predictable mapping of highly dynamic real-time applications on a NoC. For this we need real-time applications that can deal with changing resource availability, by adjusting their perceived quality (so called QoS, or quality-of-service management). The main result of this track is a demonstrator application, video object-coding (MPEG-4), running on our NoC which shows three things:

1. That our predictable design methodology saves a lot of design iterations.
2. That NoCs can be used as an adequate target for real-time video applications.
3. That dynamically application demands can be matched with the available NoC resources. Video quality should gracefully reduce when resources are limited.

Utilization summary
Successful utilization starts with carefully chosen project partners. We used the following criteria to select partners. Partners should provide adequate additional case studies in order to verify and tune our predictable design approach. These applications should be different to avoid that our methodology becomes too specific. Partners should be implementation oriented. With this we mean they should be concerned about efficiency and real-time constraints, and be aware of the tradeoff between quality and required resources. This also ensures that implementation bottlenecks are identified and solved early in the project, and that the research team has a realistic view on the efforts required for mapping an application on the target platform. An equally important criterion for selecting a partner is the extent of his customer portfolio. The reason is that we aim at maximum impact of the research results in the SoC design community. We envision the following cooperation. Our partners provide and supervise the case studies necessary to prove that the research results are applicable to a set of applications representative of the multi-media domain, and to ensure that industrially relevant problems are encountered and solved in the course of the project. The latter is considered especially important, because it is a necessary condition for our partners to feel sufficiently involved with and to value the research results to a degree that they will integrate and support successful deliverables of the proposed project in their product portfolio for their customers. Although our direct cooperation is with the mentioned partners, valuable contacts have already been established with their customers in order to stimulate early involvement of the end user with the proposed project. Based on the above criteria we selected the following project partners.

• LogicaCMG (Peter de With) will contribute extensive knowledge on innovative video applications. LogicaCMG has a large client base. LogicaCMG is an extremely interesting partner because it develops for their clients many advanced software-intensive applications that have to tradeoff quality versus performance.
• Philips Semiconductors Nijmegen (Digital Car Radio, BL CarITS, Rene van den Berg) contributes with an FM radio application.
• Philips research (Bart Barenbrug) contributes on a graphics texture mapping application.
• Philips research (Marc Duranton) also contributes with their extensive knowledge on platform design. One of their customers, Philips Semiconductors (Digital Consumer Systems) is especially interested in applying our platform and design technology.

All of above partners have many customers, several of them are important to our project and have already been contacted.

Zie http://www.es.ele.tue.nl/premadona.

Er zijn nog geen gebruikers geïdentificeerd, interesse?