Title: "Utilizing underlying synchronization mechanisms for efficient support
of different programming models".

Abstract:  


Parallel programming is a very intricate, yet increasingly important, task as
we have entered the multicore era and more cores are available to the
programmer.  One of the biggest challenges is the synchronization of the
available threads. It needs to be done not only correctly but also
efficiently, as otherwise it can become a bottleneck and hurt performance
significantly. Throughout the years, many programming models have been
proposed with different approaches regarding the way threads are created,
synchronized and managed, as well as their role on speeding up the application
(symmetric worker threads, helper threads, speculative threads, etc.).
At the same time, different hardware synchronization mechanisms have been 
proposed and/or implemented (e.g. "lock-box" for SMTs, "full-empty bits" for
Cray multithreaded architectures, etc.).
This talk reviews our work on how to leverage efficiently underlying
synchronization mechanisms for different programming models. More
specifically, we will review our earlier work on utilizing the MONITOR/MWAIT
instructions (introduced with Prescott core) to synchronize asymmetric threads
executing on Intel's Hyper-Threaded processors. Then, we will present our
current work on combining the newest synchronization mechanism (Transactional
Memory) with helper threads that speculatively exploit optimistic parallelism.
This technique has been developed and applied for accelerating Dijkstra's
algorithm, a known inherently serial application. 

 


Bio: 


Nikos Anastopoulos received his Diploma in Computer Engineering & Informatics
from the Computer Engineering & Informatics Department of the University of
Patras, Greece. He is currently a PhD candidate in the School of Electrical and
Computer Engineering, National Technical University of Athens. His research
interests include parallel and high performance computer architectures,
multithreaded and multicore processors, programming models and software
techniques for the efficient interaction of scientific applications with
emerging multiprocessor architectures. He is a student member of the IEEE.