SC2000 Technical programme - Grids a conference main themeUtrecht 30 November 2000 Grids were one of the main themes of the SC2000 conference, apart from addressing them in panels, also a technical session was devoted to the subject. William Johnson from NASA Ames described NASA's Information Power Grid (IPG) in which they want to combine a large amount of services ranging from available HPC cycles to storage for massive datasets.
Johnson's talk was a summing up of the software technology
used and the problems that still need attention such as CORBA ORBs aware
of the HPC resources to be offered and the notion of co-scheduling,
i.e., scheduling of different programs that depend on each other in
concord in (possibly) different systems in the grid. At this moment none
of the existing batch systems have the facilities to do this. The IPG
uses the San Diego Supercomputing Center Storage Broker to manage large
storage requests and tools from the Globus toolkit to glue the parts
together. In the case of NASA's IPG it comes in handy that most sites
are linked with Gigabit/s speeds.
In addition to the session on grids proper, there was also a session on
grid middleware. Here various tools were presented that should help in
a better use of the grid resources. Two of these are the AppleS
Parameter Sweep Template from the University of Tennessee and a package
to enforce resource sharing in grids. The latter tool tries to identify
resources that can be shared and to assign them accordingly by
optimisation techniques. Tao Zhao reported about a case study from New
York University with this package modeling resource sharing between
ISP-level web proxies. The AppleS tool from Tennessee provides a
template that facilitates the optimal re-use of resources in a
situation in which a parameter sweep is done over the parameter space
in a computational model. This is an obvious target for grid computing:
many program instances can be submitted to the grid to cover a range of
the parameter space to be investigated. The AppleS template tool
attempts to optimise this kind of procedure with respect to the
available resources.
Aad van der Steen |
 |