To harness the performance potential of today's complex computing platforms, applications have to be optimized in several dimensions, such as number of operations, parallelism, access to the memory system, communication among processing nodes. Computer science methodologies are fruitfully combined with domain specific knowledge in various areas of scientific and technical computing. Examples:

• Methodologies for establishing the minimum communication requirements of algorithms

• Highly parallel, low communication algorithms for the Finite Element Method in structural and material engineering

• Special purpose processor design, with application to Quantum Chromo Dynamics

• Alignment and assembly of DNA sequences

Homepage: http://acg.dei.unipd.it

People: Gianfranco Bilardi (contact person), Geppino Pucci, Carlo Fantozzi, Enoch Peserico, Andrea Pietracaprina, Francesco Silvestri

High-Performance Computing (HPC) platforms feature many processors sharing a fast communication medium and a deep memory hierarchy. Selecting a good abstract machine model for algorithm design on these platforms requires striking the right tradeoff between the largely conflicting goals of usability (ease of design and analysis), effectiveness (predictable performance on the actualÊ machines), and portability (effectiveness across machines). Some specific research themes pursued are the following:

• Analysis and Critical Comparison of Bandwidth/Latency Models

• Hierarchical Pipelined Memory Models

• Paging with dynamic memory capacity

• Distributed Implementation of Shared Memory

• Area-Universality for VLSI Design

• Automatic Translation of Submachine Locality into Locality of Reference

• Network-Oblivious Algorithms
• Cache-oblivious and cache-aware algorithms

Homepage: http://acg.dei.unipd.it

People: Geppino Pucci (contact person), Gianfranco Bilardi, Carlo Fantozzi, Enoch Peserico, Andrea Pietracaprina, Francesco Silvestri