|
|
|
|
|
|
PASSAT GROUP @ UIUC |
|
|||
|
|
|
||||||||
|
|
Home |
|
|
|
Projects in the Passat group are in the following three broad areas: Architectural
and Programming Models for Many-core Computing: To
address the issue of performance scalability and programmability for
architectures with tens or possibly hundreds of cores on the die, we are looking at programming models and
architectures that encourage
programmers to think concurrently and allow
better discovery and exploitation of program parallelism. The models that we
are looking at also enable dynamic
load balancing, dynamic adaptation to runtime variation in data level
parallelism, migrations of code to data,
and seamless hardware-software co-design. Our research on
computational models for many-core computing falls within the general theme
of Amoebic Computing that we are investigating in the Passat Group. Amoebic
computing is characterized by adaptability through replication, migration,
and diversity, and gets reflected in the models that we are developing. Responding
to Hardware/Software Technology Challenges: As transistors get smaller, faster, and less reliable, power and reliability become first order hardware
design constraints. Similarly, as consolidation (in terms of datacenters),
distributed computing, (computation over network), and connectivity become widely accepted as
software execution environments/paradigms, security and availability become first order
software/system design constraints. We are looking at system/hardware designs
that guarantee lower peak power
while providing a guaranteed higher availability, reliability, and security. Domain-specific
Processing: We are developing frameworks and tools that
can be used customize an architecture to a set of applications or an
application domain. Frameworks being investigated include reconfigurable
architectures as well processors with a software layer running on top of them
(that performs appropriate mapping and translation). Tools include design
exploration tools that allow efficient navigation through the processor design
search space. The domains that we are investigating include STLs, gaming libraries, functional programming models,
scientific computations libraries, etc. Note that our research
on
domain-specific customization also falls within the general theme of Amoebic
Computing. |
|
|||
|
|
|||||||||
|
|
|
|
|
|
|
|
|||
|
|
|
|
|
|
|
|
|||
|
|
|
|
|
|
|
|
|
||
