Skip Ribbon Commands
Skip to main content

COE

:

News: PhD Dissertation - Ayman Hroub - 16 Dec 2015

Title

PhD Dissertation - Ayman Hroub - 16 Dec 2015

Body

ANNOUNCEMENT – CSE DISSERTATION DEFENSE


Mr. Ayman Hroub, Full-Time PhD Student (CSE), will defend his PhD Dissertation on Wednesday, December 16, 2015 at 10:00 a.m. in 24-130. His PhD dissertation title is “A HYBRID SOFTWARE/HARDWARE SIMULATION FRAMEWORK FOR EARLY ARCHITECTURAL EXPLORATION OF CHIP MULTIPROCESSORS. His dissertation advisor is “Dr. Muhammad Elrabaa, Asso. Professor, COE Department”. You are cordially invited to attend.

 

Abstract:

“Simulation is the de facto tool for computer architecture performance evaluation. In the multicore era, processors became much more complex. They comprise large number of cores, complex memory hierarchies, and complex interconnection networks. Thus, the design space to be explored became much larger. Moreover, this kind of architectures has a voluminous number of concurrent events. Simulating these events sequentially would be too slow. In the last few years, researchers exploited the massive fine and coarse grained parallelism offered by FPGAs to accelerate such simulations. This is done via building fine grained parallel simulators such that the concurrent events of the simulated architecture are mapped to the fine grained components of this simulator and hence simulated in parallel. This approach has accelerated computer architecture simulation by orders of magnitude. In this dissertation, we propose HySim, a trace-driven hybrid software/hardware simulation framework for early architectural exploration of chip multiprocessors. It exploits the flexibility of software and the massive parallelism offered by the FPGAs.  HySim is faster than all existing simulators and in the same time its accuracy is in agreement with the existing accurate simulators. Its speed is up to 2.2 GIPS with 14% average absolute error relative to real hardware execution time. Since FPGAs have limited storage resources and the multi-threaded applications have very large execution traces, we have developed a technique to compress such traces. This technique compresses the execution trace on-the-fly and stores it in an executable format that can be interpreted by the timing simulator without a need for a decompression phase. The proposed trace compression technique achieved a compression ratio of up to 1862.20.

 

 ​Refreshment will be served​

Expires

1/16/2016

DisplayItem

1

Attachments

Created at 2/7/2016 11:31 AM by Webmaster of CCSE website CCSE
Last modified at 2/7/2016 11:31 AM by Webmaster of CCSE website CCSE