Advanced Course on Multicore Scheduling and Timing Analysis

September 2-4, 2014 – Retis Lab

Abstract:
Semi-partitioned scheduling aims at combining the best aspects of partitioned and fully-migrative scheduling. Partitioning simplifies scheduling by breaking it up into multiple uniprocessor scheduling problems and has low overheads. But for efficient system utilization, task migration is required. Semi-partitioning employs limited and carefully managed task migration, for high utilization without excessive overheads. Algorithms from the two main flavors of semi-partitioning (server/timeslot-based and timed-migration-based) are discussed and compared.
In another research front, the immense raw processing power of GPUs is increasingly used for general-purpose extremely parallel applications consisting of thousands of identical threads, operating on different data. Techniques are needed for upper-bounding the time since the first thread starts executing until all threads have completed: this is the "GPU equivalent" of a WCET. The challenge is that we cannot control the intra-GPU scheduling, nor is it publicly documented.

Brief Bio:
Konstantinos Bletsas is a research associate at the Polytechnic Institute of Porto, Portugal, working on multiprocessor scheduling algorithms. He was born in 1978 and grew up in Volos, Greece. He has a first degree (2002) in Electronics and Computer Engineering from the Technical University of Crete (Chania, Greece) and a PhD (2007) in Computer Science from the University of York (England), supervised by Neil Audsley. In 2007, he joined the CISTER Research Unit (Porto, Portugal).

AttachmentSize
PDF icon Flier of Bletsas Course117.98 KB
Tags: