parallel 2013 » Agenda »
// Keynote: The struggle to design software that fills the performance dimensions of modern CPUs
This talk will review the complexity of modern microprocessor-based compute-intensive servers and discuss the difficulty of designing and writing software that will utilise all the "seven performance dimensions" available in modern hardware. Whereas we used to rely on Moore's law not only to increase the transistor count, but also to exploit the reduced feature size for augmenting the processor's frequency and hence throughput, those days are long gone. During the last ten years (at least), the community has been forced to struggle with multiple complex issues, such as vectorization and multi-threading, to try to regain lost performance and throughput.
The talk will try to assess where we are, based on experience at CERN and elsewhere, and suggest ways to (continue to) improve the situation.
// Referent
// Sverre Jarp
is the Chief Technology Officer of CERN openlab, a joint collaboration with leading industrial partners in order to assess cutting-edge information technology for LCG, the Large Hadron Collider’s Computing Grid. CERN is the European organisation for high energy physics in Geneva. Sverre has been working in computing at CERN for over 39 years and has held various managerial and technical positions promoting advanced but cost-effective computing and data storage solutions for the Laboratory. In 2001-02 he spent a sabbatical year in HP Labs (Palo Alto, USA). Upon his return to CERN he was appointed CTO. His current fields of interest cover areas such as parallel programming and big data management for physics. Jarp holds a degree in Theoretical Physics from the Norwegian University of Science and Technology (NTNU) in Trondheim.