|
|
|
Computer Architectures
Simulations: |
| there is an increasing interest on scalable parallel architectures with distributed shared memory. We are investigating
CC-NUMA and COMA multiprocessor architectures based on software and hardware cache/page coherency techniques. We are also investigating techniques for hiding the long communication latency in distributed memory architectures, for instance, data pre-fetching, multithreading and relaxed memory consistency models. In order to aid the development of new architectures, basic software system and parallel algorithms, we are working with program-driven simulation programs and visualization tools. Currently most of our work are based on the Rochester MINT simulator, but we are also studying other ones, such as the UIUC
Augmint. Some research topics are: |
 |
Scalable shared memory multiprocessors with cache coherence; |
|
CC-NUMA and COMA architectures; |
|
Software and hardware techniques for scalable cache coherence; |
|
Execution and program driven simulation of innovative architectures; |
|
Interconnection networks for large scalable parallel processing; |
|
|
Prototype
Construction |
| we are implementing several prototypes of parallel machines. These prototypes can be classified in two
categories: |
|
Message passing cluster |
| This kind of system has been subject of strong investigation during the last years. Beowulf class cluster systems are very popular nowadays in universities and R&D institutions worldwide. We implemented several cluster based on low cost COTS processing nodes and networks. We also are implementing custom interconnection systems based on communication adapters plugged on high speed computer buses, high speed links and high speed low-latency switches. In hardware, the activities are focused in new interconnection networks design, and in software, in cluster management systems and communication protocols, as well as in new applications. This kind of system is intended for
high performance computing and high availability computing system |
|
Shared memory
(cc-numa/coma) cluster |
| Scalable shared memory systems are probably the next generation of computer systems. Currently we have commercial systems such as SGI Origin and SCI based system
manufactured by Data General and Sequent. The next generation of Alpha processors - 21364, demonstrate that in the near future we are going to have scalable low cost high
performance systems with hardware shared memory. In hardware, the activities are
focused in investigation of new architectures (processing nodes and very high speed interconnection
networks) and implementation of concept of proof prototypes. In software, the activities are focused in
CC-NUMA Operating Systems, Scalable I/O
architecture, and programming tools. |
|
|
|
|
