Research - Karthikeyan Sankaralingam
- CV, References
-
Research statement
-
Teaching statement
-
List of Publications
- Select papers:
Distributed Microarchitectural Protocols in the TRIPS Prototype Processor,
MICRO-39, 2006
pdf
-
Exploiting ILP, TLP, and DLP with the Polymorphous TRIPS Architecture,
ISCA-30, 2003
pdf
-
Universal Mechanisms for Data-Parallel Architectures,
MICRO-36, 2003
pdf
-
A Design Space Evaluation of Grid Processor Architectures,
MICRO-34, 2001
pdf
-
Research Summary
- The Grid Processor Architecture
[ Paper Abstract ]
- A scalable distributed microarchitecture which provides ultra-wide issue computation, large instruction windows, scales to future wire-delay dominated technologies, and can replace conventional out-of-order superscalar processors. The three main features are:
- Encode dataflow dependences in the ISA to enable direct
instruction-instruction communication and reduce the overheads of
detecting and managing dependencies that conventional out-of-order
processors must pay.
- Partition the program into well-defined
blocks to limit the scope of the dependencies so that the number of
dependence arcs does not become too many to encode in the instruction
space.
- To manage design
complexity and address wire delay scaling and reliability, the
computation core is completely distributed using microarchitecture
control and data networks with only nearest-neighbor links used for
communication.
- Polymorphous architectures - TRIPS
[ Paper Abstract ]
- My dissertation research introduced the concept of a polymorphous processor - one that can adapt to the workload by altering the behavior of coarse microarchitecture units to suit the application. TRIPS prototype chip
ISA, microarchitecture, processor verification and physical design
lead.
- Mechanisms for data level parallelism
[ Paper Abstract ]
- In my dissertation research, I identified fundamental
attributes for classifying DLP programs based on memory behavior,
control structures, and amount of available parallelism. I proposed a
set of universal mechanisms, which can provide architectural
capability for run-time processor customization for any application
sub-domain within DLP.
- Ranking metrics for P2P Systems
[ Paper Abstract ]
- I developed a
distributed formulation of the Google pagerank algorithm that can be
executed on a Peer-to-Peer network. This work provides the first
solution for ranking documents on P2P networks. Extending this idea to
web servers I showed how to build a completely decentralized search
engine.
-
TRIPS Prototype Processor
[
Paper Abstract ]
-
This paper describes
the control protocols in the TRIPS processor, a distributed, tiled
microarchitecture that supports dynamic execution. It details each of
the five types of reused tiles that compose the processor, the control
and data networks that connect them, and the distributed
microarchitectural protocols that implement instruction fetch,
execution, flush, and commit. We also describe the physical design
issues that arose when implementing the microarchitecture in a 170M
transistor, 130nm ASIC prototype chip.
-
List of Publications
-
Patents