Åóð ×ô Blockin Blockin¬ Blockin Blockinøøóò Òò Úú×ùùððþþøøóò Ò Øøø Ååëëáîî Ôöó Blockinø

Model spe i ation and visualization in the MASSIVE proje t Laurent iu Ni olae Ed Deprettere Bart Kienhuis Leiden Institute for Advan ed Computer S ien e Leiden University, Leiden, The Netherlands E-mail: fnld,edd,kienhuisg lia s.nl Abstra t| The MASSIVE proje t aims at developing methods and tools for the exploring of large-s ale embedded system designs at a high level of abstra tion. One of the key problems that must be addressed is mastering the omplexity of both systems and tools. For e e tive and eÆ ient exploration of su h system designs, one needs models of behavior and stru ture of appli ations and arhite tures and a fast simulator for measuring the performan e of the ombination of both. It turns out that existing frameworks have either good modeling apabilities or good simulator apabilities, but not both at the same time. In the MASSIVE proje t we have investigated the possibility of interfa ing the Ptolemy framework from Berkeley, and the System Simulation Framework from Dartmouth (DaSSF). Ptolemy is used to reate, visualize and manipulate stru tural models. DaSSF is used for fast lustered simulation of system level designs. Stru tural models are exported from Ptolemy to DaSSF and performan e numbers obtained from the simulator are imported ba k to Ptolemy for further evaluation and visualization. We have implemented the interfa ing of these two frameworks and evaluated the usefulness of the new environment by working out an illustrative exploration example. Keywords|embedded systems design, design spa e exploration, system level modeling I. Introdu tion The MASSIVE proje t is a resear h proje t that is fo used on the development of a methodology for high level design of large s ale embedded digital signal pro essing systems. As arrier for the proje t, a onrete ase is planned to be ta kled: exploring alternative system level implementations of the digital signal pro essing part of a Square Kilometer Array (SKA) type radio teles ope [1℄, [2℄, onsisting of 10,000,000 RF sensors, demanding multiple Terra-operations per se ond of pro essing power. The problem we are addressing is design spa e exploration for su h systems, using models for both the possible ar hite tures and for the appli ations that run on these ar hite tures. After de iding upon one model from ea h ategory, the designer hooses a mapping of the appli ation onto the ar hite ture and exeutes a simulation of the resulting system model. After the simulation is over, the performan e numbers from the simulation are sent ba k to the designer, who de ides whether he should explore other alternatives.