Objective:
The purpose of this
research
is to investigate the use of dynamically optimized cache
sharing as a means of increasing performance of CMP
architectures across various multi-process and multi-threaded
workloads.
_________________________________________________________________________________________________________
- Methodology
- In order to adequately
characterize and evaluate the performance of our proposed cache
system, we implemented various versions
of our DOCS structure within the confines of an execution-driven
multiprocessor simulator called
Multifacet GEMS
- The GEMS Ruby
module in particular was used as a detailed memory system simulator.
It can simulate a wide variety of
memory hierarchies and systems ranging from broadcast-based SMP to a
hierarchical directory-based Multiple-CMP system.
- Design Considerations
- Design and Implementation
- Simulations Results
- Performance Evaluations
- Conclusions
|
Note:
Since this research is still in progress, I am withholding final
drafts until they are published. For more information,
please
contact
me.
|
|