Michael Franz (Univ, of California at Irvine)

Distributing and Managing Mobile Code
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Using mobile code is fraught with risks. If an adversary deceives us
into executing a malicious program, this may have catastrophic
consequences and may lead to a loss of confidentiality, loss of
information integrity, loss of the information itself, or a
combination of these outcomes. Unfortunately, additional provisions
for security frequently cause a loss of efficiency, often to the
extent of making an otherwise virtuous security scheme unusable for
all but "toy" programs. Moreover, current mobile-code distribution
models are surprisingly primitive; for example, they assume that all
constituent parts that make up a mobile program are downloaded to a
single location, then verified, linked, possibly dynamically compiled,
and finally executed at that same location.

Our research in the past 6 years has focused on making mobile code
*practical*, to the extent that it can completely displace native code
for the whole spectrum of applications from PDAs to desktop
applications to scientific computing on supercomputers.

Our contributions include work on dynamic compilation, which raises
the performance of such programs, often to the extent of outperforming
statically compiled code.

A second major contribution is a class of representations for
target-machine independent mobile programs that can provably encode
only legal programs. Hence, there is no way an adversary can
substitute a malicious program that can corrupt its host computer
system: Every well-formed mobile program that is expressible in our
encoding is guaranteed to map back to a source program that is deemed
legal in the original source context, and mobile programs that are not
well-formed can be rejected trivially. Further, our encoding not only
guarantees referential integrity and type-safety within a single
distribution module, but it also enforces these properties across
compilation-unit boundaries. As a side-effect, our encoding is
exceptionally dense: it outperforms the next best compression scheme
for Java by almost 50%.

A third focus of research is on managing the mobile-code pipeline from
code producer to code consumer: capturing all meaningful modes of
mobile-code deployment in a model, and then creating a code management
architecture that makes this model enforceable by mechanical means.

---

Prof. Michael Franz leads a research group of 9 Ph.D. students and one
Post-Doctoral fellow at the University of California, Irvine. His
research is grouped into the three main threads of dynamic
optimization ("accelerate a program while it is already running"),
mobile program representations ("alternatives to the Java Virtual
Machine"), and component-oriented languages and software systems
("paradigms beyond object-oriented programming"). Franz received a
Dr. sc. techn. degree in computer science and a
Dipl. Informatik-Ing. degree, both from the Swiss Federal Institute of
Technology, ETH Zurich.