University of Illinois at Urbana-Champaign
255 Coordinated Science Lab, MC 228
1308 W. Main St.
Urbana, Illinois 6180
United States of America
- AG-66 (Lecture Theatre)
The next major innovations in computer science and engineering are likely to come from [UTF-8?]â€œintelligentâ€ deployment of human-centric systems that must optimally interact with other man-made and natural systems with a focus on seamless availability of dynamic decision-making capabilities. The compelling driver is our need to measure, model, communicate and manage our vast natural and social enterprise. New innovation and creative approaches are essential to be able to handle problems of scale and complexity envisioned in this future. This change is being driven by inexpensive technology deployed at previously unimaginable scales. Our planet itself, via a vast and varied network, is generating or has the potential to generate an unprecedented level of information that we were previously oblivious to and could ignore - we were not able or hear it, to see it or to capture it. The vast and dynamic information gathered with these information networks could be analyzed using powerful yet ubiquitous systems- that are yet to be invented. In this seminar Professors Iyer and Lumetta will discuss two projects at the University of Illinois that in their unique ways address this [UTF-8?]â€œbig [UTF-8?]dataâ€ problem.
Professor Iyer will address the issue of protecting a system from accidental errors and malicious attacks via a new approach based on [UTF-8?]â€œexecution under [UTF-8?]probationâ€. Via a combination of continuous monitoring and leveraging Virtual Machines, the method can both identify and sequester a potential attacker while ensuring that any consequent system corruption is contained.
Prof. Lumetta will discuss the development of combined software and hardware coherence protocols in the context of Rigel, a 1000-core accelerator chip. Dynamically shifting regions of memory between these two types of coherence both offers a mechanism to support the two models now dominant in the industry and simplifies the process of parallelizing applications for many-core execution.