The a-kernel Project
OS code is notoriously complex, with many systems having millions of
lines of code containing subtle and intricate interactions. The
inability for programmers to effectively reason about parts of the
system, develop components independently, and even configure
customized systems at build time are well known ramifications of the
deep structural flaws in OS code. The goal of the a-kernel
project is to determine if aspect-oriented programming can be used to
improve OS modularity, and thereby reduce complexity and fragility
associated with system implementation.
The structural problems we are targeting involve the implementation of
concerns in the system that have never before been modularized. Parts
of the system are, for example, inevitably scattered throughout major
functions in virtual memory, pagers, and file systems, and tangled
with the primary functionality of page fault handling and file system
requests respectively. As a natural consequence of their lack of
modularity, these concerns are inextricable from the primary
functionality, are hard to comprehensively appreciate, and impede
independent development as they couple, and are coupled with, the
primary functionality they crosscut.
We began by proposing a small scale design of an aspect-oriented
approach to prefetching and assessed its impact on modularity. We are
currently evaluating the application of this original design to a
system-wide implementation of prefetching covering VM, 4 pagers, and 7
file systems, or 17 execution paths in total. To date we have not
introduced any new functionality, but instead refactored the original
implementation of prefetching as it is found along each of the
relevant execution paths into special crosscutting modular units, or
aspects. In order to do this in an existing UNIX kernel, it
was also necessary for us to develop AspectC - a simple subset of
AspectJ.
AspectC provides linguistic mechanisms that allow us to localize and
structure crosscutting implementation according to execution
paths. Specifically, we can name the execution paths involved, attach
functionality at specific points on these paths, and propagate dynamic
context along these paths. Using these mechanisms, we have been able
to modularize the implementation of prefetching, which before now
has never been
untangled from primary functionality and implemented in a separate,
localized, modular-unit.
So far, our work has demonstrated the ways in which explicit
structural support for crosscutting improves OS modularity. That is,
prefetching code is localized; interactions between prefetching and
primary functionality are explicit in the aspect's interface;
propagation of dynamic context related to prefetching is declarative;
and points of intersection between different aspects on the same
execution path are declarative. These qualities are some of the ways
in which we have been able to reduce the complexity and fragility
associated with prefetching's implementation. We are now in the
process of assessing how this approach scales in terms of system
complexity and performance.
These results have potential impact in several areas associated with
OS research, which we are continuing to explore. In keeping with
recent work in product-line architectures, aspects introduce new
possibilities in the design and implementation of reconfigurable,
extensible, and application-specific systems. Similarly, both the
comprehensive implementation of evolutionary changes, and the use of
contracts to ensure correctness, stand to benefit from structural
support for crosscutting concerns in OS code. To this end, we are
exploring evolutionary changes associated with the Network File
System, and the style of contracts identified by recent work on
meta-level compilation.
Position Summary: Aspect-Oriented System
Structure. Yvonne Coady, Gregor
Kiczales, Mike Feeley, Norm Hutchinson, Joon Suan Ong and Stephan
Gudmundson. In The 8th Workshop on Hot Topics in Operating Systems (HotOS), May 2001.
Structuring System Aspects.
Yvonne Coady, Gregor Kiczales, Mike
Feeley, Norm Hutchinson, and Joon Suan Ong. In International
Conference on Software Engineering, Aspect-Oriented Programming
Workshop, May 2001.
Can AOP Support Extensibility in
Client-Server Architectures? Yvonne
Coady, Alex Brodsky, Dima Brodsky, Jody Pomkoski, Stephan Gudmundson,
Joon Suan Ong and Gregor Kiczales. In European Conference on
Object-Oriented Programming (ECOOP), Aspect-Oriented Programming
Workshop, June 2001.
Structuring Operating System Aspects
Yvonne Coady, Gregor Kiczales, Mike
Feeley, Norm Hutchinson and Joon Suan Ong. In Communications of
the ACM (CACM) (to appear), October 2001.
Using AspectC to Improve the Modularity
of Path-Specific Customization in Operating System Code
Yvonne Coady, Gregor Kiczales, Mike
Feeley and Greg Smolyn. To appear in Proceedings of the
Joint European Software Engineering Conference (ESEC)
and 9th ACM SIGSOFT International Symposium on
the Foundations of Software Engineering (FSE-9), 2001.
Aspect-Oriented Incremental Customization
of Middleware Services
Alex Brodsky, Dima Brodsky, Ida Chan,
Yvonne Coady, Jody Pomkoski, and Gregor Kiczales. Submitted.
Exploring an Aspect-Oriented Approach to
OS Code
Yvonne Coady, Gregor Kiczales,
Michael Feeley, Norman Hutchinson, Joon Suan Ong and Stephan Gudmundson.
4th ECOOP Workshop on Object-Orientation and Operating Systems,
June 2001.
Coping with Evolution: Aspects vs Aspirin?
Alex Brodsky, Dima Brodsky, Ida Chan,
Yvonne Coady, Stephan Gudmundson,Jody Pomkoski,
Joon Suan Ong.
OOPSLA Workshop on Advanced Separation of Concerns,
2001.
Thesis proposal:
Crosscutting the Great Divide: Exploring an
Aspect-Oriented Approach to OS Code. Yvonne Coady, January 2001.
Talks:
Structuring System Aspects: How do small aspects compose?
International
Conference on Software Engineering, Aspect-Oriented Programming
Workshop (slides), May 2001.
Using AspectC to Improve the Modularity
of Path-Specific Customization in Operating System Code
Yvonne Coady, Gregor Kiczales, Mike
Feeley and Greg Smolyn.
Joint European Software Engineering Conference (ESEC)
and 9th ACM SIGSOFT International Symposium on
the Foundations of Software Engineering (FSE-9)
(slides), Sept 2001.