Potpourri
Modern high-end computing design calls for large numbers of processors
and/or nodes to be combined into an image of a single machine. Such
computing environments are prone to periodic hardware failures and must
be able to overcome these failures and still perform useful work. In
order to minimize the amount of down time experienced, we are developing
new techniques and algorithms to provide fault tolerant and highly
efficient task scheduling in multi-processor computing systems.
One often neglected measure of robustness is the timeliness of
system operation. In addition to fault-tolerance and scheduling
work, we have developed some useful approaches for designing and
reasoning about systems with timeliness requirements, i.e.,
real-time systems.
As an area where robustness is a factor, we are also looking into
the building of clusters which minimize the "total cost of ownership
/ performance" ratio. Where as the traditional "price / performance"
ratio addresses the question "What is the highest performance
supercomputer I can afford to buy?" it does not address the more
important question "What is the highest performance supercomputer
that I can afford to own?" In addition to the cost of acquisition,
owning a supercomputer requires one to consider the facility and
utility costs (including space, power and cooling). It also requires
consideration of the maintenance and system administration costs.
The latter costs are directly impacted by robustness (or the lack
thereof). We call this project Supercomputing in Small Spaces.
