Fixing the Bugs and Performance Bottlenecks in Shared-memory Concurrency Programs

PhD Thesis Proposal Defence


Title: "Fixing the Bugs and Performance Bottlenecks in Shared-memory Concurrency
Programs"

by

Mr. Peng LIU


Abstract:

Shared-memory concurrency programs are pervasive in this multi-core era. In 
these programs, programmers use the threads to concurrently access the shared 
resource and use the locks to synchronize between the threads to guarantee the 
correctness criteria such as race-freeness.

It is very challenging to specify the locks that achieve both goals of the high 
performance and the correctness. For example, using the coarse-grained locking 
often leads to the performance bottleneck because multiple threads are 
unnecessarily prevented from concurrently accessing the shared data structure. 
On the other hand, using the fine-grained locking often leads to concurrency 
bugs such as data races and deadlocks, as the large space of possible 
interleavings make the manual reasoning difficult.

We first propose two techniques, which are built upon formal models, for fixing 
the concurrency bugs in existing programs, without introducing the high 
performance overhead. Then, we propose a technique, which is based on the 
static analysis of the data structure hierarchy, for fixing the performance 
bottlenecks by unleashing more concurrency for the data structure. We also 
present our implementation and the evaluation on heavily used real world 
applications.


Date:			Thursday, 3 October 2013

Time:                   2:00pm - 4:00pm

Venue:                  Room 4480
                         lifts 25/26

Committee Members:	Dr. Charles Zhang (Supervisor)
 			Dr. Lin Gu (Chairperson)
 			Prof. Shing-Chi Cheung
 			Dr. Sunghun Kim


**** ALL are Welcome ****