Report ID
2008-17
Report Date
Abstract
The next generation of large-scale machines for scientific computing
will consist of a large number of nodes, each supporting many-core and
multi-core processor con?gurations. On one hand, to extract performance
from such machines, the operating system will have to be streamlined. On
the other hand, the heterogeneity of these machines i.e., the variety
of communication channels and computing elements will demand a larger
set of OS services to achieve scalability. We believe that
high-performance OS paravirtualization techniques, where different
light-weight OS instances cooperate to provide the different OS
services, can resolve this con?ict. Current research in
paravirtualization aims to isolate OS instances for performance
guarantees and security; in a high-performance computing setting,
however, thread synchronization and inter-core communication within an
application require that the memory is shared and coordinated to meet
the application needs. In this paper, we present an approach to enabling
efficient memory sharing and synchronization across OS instances running
on a multi-core machine. Our performance evaluation shows that this
approach allows native execution speeds in a paravirtualized setting,
along with communication latencies that are lower than under any
existing inter-OS communication mechanism.
Document
2008-17.pdf109.65 KB