SAMOS 2008 Workshop Special Sessions Description
The workshop considers state-of-the-art research papers as well as papers on ongoing work with prom-ising preliminary results. Positional and in-depth review papers are also considered. Furthermore, the workshop covers additional topics compared to IC-SAMOS that are complementary to processing and necessary to build embedded systems, e.g., sensor systems. In 2008, we intend to have two special sessions focusing on important aspects of embedded systems. They are described below.
Special Session 1:
The rapid diversification of domain-specific processing architectures for signal processing systems, which now incorporates architectures as varied as streaming processor, massively parallel multi-core chips, and heterogeneous FPGA means that current and next-generation system-on-chip platforms for com-munications systems contain highly complex interacting components. This provides systems designers with potentially unprecedented levels of processing power. However, design tool and methodology sup-port to allow rapid architecture derivation, programming, and optimization of specific DSP systems on these general platforms is seriously lacking and is hindering the capabilities of designers to harness this power. This special session would feature expert speakers in the areas of system level modelling, rapid implementation and design space exploration for hardware, software and entire system architectures outlining their vision, proposed solutions and experimental evidence of how system design in these ar-eas should be carried out to establish the next generation of practical, productive design processes and tools.
Special Session 2:
Programming multi-cores . be they homogeneous or heterogeneous . are no different conceptually than programming high-performance supercomputers. If we look at the last thirty years of research in this area, the results are meager (at least in practice). The abstraction used by around 99% of HPC users comprises an API that allows tasks to be created and to send and receive messages with various types of synchronization. Surely, we can do better than this! The problem with this paradigm is that it mixes al-gorithmic and concurrency engineering and does little to promote safe and composable parallel pro-grams. At the HPC end of the computing market, researchers can afford to spend large amounts of effort in re-engineering code for each new generation of supercomputer. However, the same cannot be said for the embedded and commodity markets. Here, concurrency is also making its presence known in the form of multi-core and many-core chips . both homogeneous and heterogeneous. In the next ten to fifteen years processor will become like the transistor and the new processor will be the cluster, the question is whether we have the correct paradigm for programming these devices. What is not appar-ent in the above statistic is the large amount of research that contributes to the remaining 1% of users. There is no lack of contenders for concurrent programming languages. The question that must be asked is whether there are programming models that are ubiquitous enough to form the basis for a coordi-nated solution to the problems of programming MPSOC.
This special session aims to bring together a number of like-minded researchers working in the area of programming models for the next generation of MPSOC and to explore this possibility of developing programming models, languages and compilers to enable a paradigm shift in this important area. The aim of the session would be to explore solutions that enable the problems of algorithm design and con-currency engineering to be separated. This requires languages for algorithm design that express solu-tions generically but expose concurrency, e.g., functional languages, data-parallel languages or paralleli-zation from sequential code. It also requires languages to compose well-specified components together safely and to map them onto resources, either statically or increasingly dynamically, e.g. using streaming or coordination languages. Finally, it will require execution and resource models to uniformly bind these different aspects together into dynamic systems. These are challenging problems but significant progress has been made by a number of researchers in recent years and this session would aim to invite key players to present their work and provide an opportunity for accelerating this development.