How's the Parallel Computing Revolution Going?
Towards Parallel Scalable Virtual Machine Services

By Kathryn McKinley

To answer this question, we first overview trends in scalability and power
efficiency of modern application executing on a cross section of current and
past processors.  These results show that many managed applications are not yet
scalable and consume disproportionate amounts of power.  Achieving performance
scaling of parallel applications written in managed languages must start with
scalable virtual machine (VM) services.  Because the VM schedules, monitors,
profiles, compiles, optimizes, garbage collects, and executes along with the
application, it must be scalable and has a unique opportunity to enhance
application scalability.  This talk uses results from my research group on
concurrent dynamic analysis to illustrate the potential and challenges of
obtaining scalability on modern chip multiprocessor hardware.

We conclude with an overview of challenges for scalable virtual machines and
suggested future directions.


Biography: Professor Kathryn S McKinley received her Ph.D. from Rice University
working with Ken Kennedy. Her research interests include compilers, memory
management, runtime systems, programming languages, debugging, and
architecture.  She and her collaborators have produced tools that are in wide
research and industrial use, including the DaCapo Java Benchmarks, the TRIPS
Compiler, the Hoard memory manager, the MMTk garbage collector toolkit, and the
Immix mark-region garbage collector.  Her service includes program chair for
ASPLOS, PACT, and PLDI; co-Editor-in-Chief of ACM Transactions on Programming
Language Systems (TOPLAS); and mentoring in CRA-W programs for under
represented students. She is currently supervising five and has graduated
fourteen PhD students. McKinley is an ACM Fellow.