Emulation: Expanding Evaluation Context via Outward Replications
of Experimental Design Impacts
Tuesday October 12, 2004
Hamerschlag Hall D-210
Carnegie Mellon University
Recently, real-time hardware-based cache emulators have been
proposed as an alternative approach to software-based simulators.
Real-time emulation provides several advantages over simulation,
such as faster run-time and smaller storage requirements. However,
existing emulators are passive. They obtain traces from the
system under test (host) in real-time and process them for
emulation in the hardware. This approach limits the context
in which an experimental design is evaluated as performance
measurements can be gathered only within the boundary of the
emulator. This is because the impacts on the experimental design
on the real-system hosting the emulator are not replicated.
In order to extract performance measurements from the host
system itself, the emulator has to replicate the impacts of
the experimental design from the host system perspective.
We propose the concept of active emulation in which the hardware
actually replicates the impact of an experimental design under
study on the host system (i.e. outward replication). This allows
for expanding the boundary of the evaluation of the experimental
design from the emulator context to system-wide context. This
concept is implemented in the Active Cache Emulator (ACE),
an FPGA-based cache emulator that has the capability to induce
delay to the front-side bus of its host system for active emulation
of cache delays. This talk will present the preliminary work
with ACE. The notion of active emulation, along with the design,
implementation, and evaluation of ACE will be presented. Lastly,
current status, challenges and future work of this research
will be discussed.
Eriko Nurvitadhi is a graduate student in the Electrical
and Computer Engineering Department at Carnegie Mellon. He
received his BSs, BA, MS, and MBA degrees from Oregon State
University. His current research is in the emulation framework
for the TRUSS project. His advisor is Prof. James Hoe.