[Colloq] REMINDER: PhD Thesis Proposal - FRIDAY, AUG. 25 - Fabio Rojas

[Rachel Kalweit]

College of Computer and Information Science
presents
PhD Thesis Proposal by:
Fabio Rojas

Title: Predicting Garbage Collection Behavior Using Object Lifetime Models

Friday, August 25, 2006
12:00pm
366 West Village H

Abstract:
Garbage Collection is a technology for automatically reclaiming
dynamically allocated heap storage that is no longer needed by a
program. Garbage collection removes the burden of deallocating memory
from the programmer, removing a common source of programming errors, and
it allows for better interoperability between third party libraries.

There are several different garbage collection strategies, each with its
own advantages and disadvantages. Different strategies will collect
different regions of the heap, or treat newly allocated objects
differently. These decisions will impact the overall performance of the
garbage collected application, making the choice of garbage collection
strategy an important part of performance tuning. The objects allocated
by an application will exhibit different lifetime behavior than those
allocated by another application. This is one of the reasons why no one
garbage collection strategy is always best for all possible programs. It
is important to understand how different collection techniques will
affect the performance of an application. It is also desirable to be
able to predict how garbage collection will impact the runtime
performance of an application.

Object behavior is complex and difficult to understand. But it is
possible to build mathematical models of object lifetimes which
approximate the behaviors of an application's objects. While these
models cannot capture all of the details of object behavior, they have
been shown to be a useful tool in analyzing garbage collection
performance. Until now building these models has been a computationally
expensive task that must be performed offline.

The goal of my research is to use object lifetime models to predict the
impact that different garbage collection techniques would have on the
performance of an application at runtime. My approach will be to build a
system that gathers the information needed to model the behavior of a
running application with minimal impact on the application itself.  The
model will then be used to estimate the impact of several different
garbage collection strategies on the application. The system will use
these estimates to choose the best collection algorithm from those
available at runtime.

Advisor: Will Clinger