About Us - Future
Application-Driven Computing Growth
The IMPACT Group continues to build and strengthen our strategic research
partnerships with industry. These relationships enable us to approach
research problems with more expertise what most academic groups can achieve.
- Our long-term relationships with Intel and HP
are strong as ever, and they are working closely with us on their future microprocessors.
- Through GSRC, the IMPACT Group is also partnering
with Xilinx, developing Linux capabilities for the Virtex Pro on
advanced pre-release parts. Virtex Pro FPGA chips and baseline tools, supplied to
us by Xilinx, have greatly facilitated our work on customized compute engines in
the high-level parallelism work.
- The IMPACT Group recently established a research partnership with
Microsoft to mature some of our analysis technologies in their
Phoenix framework. The IMPACT scalable pointer analysis framework
has the potential to advance the capability of software visualization and management
tools.
- In 2003, Wen-mei was invested as the first Jerry Sanders-AMD Endowed Chair in ECE,
which establishes for us a new line of collaboration with Advanced Micro Devices.
- IMPACT team members regularly participate in research internships
with IBM , working in compiler, architecture, and system performance
analysis groups.
If the past is an accurate predictor of the future (as we computer architects like
to assume), next-generation architectures and software will be dictated by today's
emerging applications. The need for codebreaking and ordnance delivery tables drove
innovation in the early computers of the 1940s. A decade later, John Backus and
his team at IBM designed FORTRAN to render the scientific algorithms of the day
in a form both natural for a programmer to express and easy for a compiler to optimize.
More recently, in the RISC era, HP's PA-RISC architects intensively studied workstation
application needs in designing new machines. Today the desire for realistic audiovisual
experiences in gaming and digital content creation is driving the design of heterogeneous,
highly parallel architectures such as the IBM/Sony/Toshiba Cell. Situations and
implementation constraints may vary, but what remains consistent from the era of
tubes to today's billion-transistor-plus chips is the fact that necessity - not
raw possibility - is the mother of invention.
So what applications will actually drive the next wave of computing technology development?
We see two broad "driver applications" dominating in the next 10 to 20 years:
1. First is digital content creation and audiovisual experience delivery. These
applications share both necessity (rapidly growing markets) and immediate possibility
(inherent parallelism facilitating short-order solutions). The IMPACT
Group is already using diverse coding, decoding, and manipulation algorithms to
experiment with ultra-efficient alternatives to classical computer architectures.
Today's industry faces a conundrum: single-purpose silicon solutions are too expensive
and too transitory for most of these applications, but classical programmable architectures
can't deliver the desired performance-per-watt. IMPACT members
consider this an opportunity to use powerful software tools to rebalance the hardware
/ software equation. In particular, we are attacking the memory interface: typical
software assumes a uniform interface to memory -- which is an increasingly challenging
abstraction to maintain in highly parallel and distributed systems. New architectures,
like IBM Cell and those proposed by our group, distribute the memory interface to
reduce complexity and increase available access to parallelism. The downside is
that this strategy complicates the programming model. IMPACT's memory access analysis
tools will continue to evolve to address this important problem, healing the growing
breach between productive programming models and efficient execution architectures.
2. Second is biochemical / biomedical simulation and modeling. The human genome
has been mapped and the biochemical industry is beginning to be able to identify
and measure a broad array of proteins with sophisticated sensing devices. But the
promised dividends in disease prevention, diagnosis, and treatment have not yet
materialized. Just as computing revolutionized codebreaking, it will also help glean
useful information from this ever-increasing muddle of biological and chemical data.
The IMPACT Group is beginning to investigate these areas in collaboration
with leading researchers in these fields. But finding effective solutions to these
big-iron problems will take more than just connecting huge clusters of COTS workstations.
And at the same time, the scalability of those solutions is becoming limited by
slower per-processor performance and tighter and tighter limits on acceptable power
consumption and dissipation. As our research in this area develops, we will apply
our experience with ultra-efficient computing solutions to this exciting new realm
of grand challenge problems, where we plan to have real societal IMPACT. We are
excited about our work and what the future promises.
The IMPACT Research Group is ideally positioned to push our vision
of next-generation computing and find the best solutions. Our location in the University
of Illinois Coordinated Science Laboratory puts us in the epicenter of one of the
most densely populated computing and scientific communities in the USA, including
biochemical simulation and stochastic modeling. Our solid connections with industry
and academic researchers give us additional insight into both up-and-coming applications
and state-of-the art implementation techniques. Join us in shaping the future.