|
|
Computing has become an indispensable tool for scientific discovery.
Our research is focused on enhancing the scientists' parallel computing
experience. In particular, the PALM group studies several aspects
of data-intensive parallel computing:
-
Parallel I/O and storage systems:
I/O has becoming the weakest link in the
scientific computing cycle for many applications, due to the widening
performance gap between the I/O subsystem and other system
components. We work closely with parallel applications to provide
novel technology for improving the application perceived I/O
performance. This is mainly through intelligent buffering, caching,
and prefetching techniques, as well as the effective coordination
between the optimization efforts across multiple high-end computing
I/O stack layers.
-
Desktop parallel computing: Typically, the massive amount of data
generated by supercomputers or other instruments eventually have
to be viewed and analyzed by human, often on their desktop workstations.
Such activities are more and more limited by the single-machine
capability, while pervasive personal computing has generated many
idle resources (both for computing and for storage)
in the desktop computing environment. We study
how to utilize such unused resources to improve the productivity
of scientists' day-to-day data processing. Our goal is to achieve
better application performance without losing the convenience
and interactiveness of sequential processing.
|
