Molecular Simulation and Systems Biology - Systems Biology: Networks, Models, and Applications - page 91

Exploiting Advances in Computer Architecture
As has been the case since the invention of the integrated circuit,
the landscape of computational capability is changing at an amazing
rate. Computers have become faster and less expensive, but so has the
ability to build clusters of personal computers and workstations with
high-speed interconnects. Special on-chip hardware that supports a
high degree of pipelining, vector operations, caching, and multiple
64-bit floating point arithmetic units has become almost standard.
Most production computer systems consist of nodes with multiple
processors having these capabilities. Advances in and standardization
of software interfaces for the development of parallel software have
enabled the ability to exploit these kinds of computer configurations.
Many of the standard modeling and simulation software packages
have been modified to exploit these parallel configurations.
The Internet has produced a new computing paradigm that we
are still learning how to exploit. With grid computing one will be able
to transparently distribute tasks and data to computational capability
across the Internet. One form of this exists in the highly successful
Folding@Home project at Stanford, where 100,000 people have
contributed otherwise idle computer time to protein folding simula-
tions performed by a screen saver [67]. A similar project operated by
Oxford University has been established with over 2.6 million con-
tributed computers that are performing a search for anticancer drug
candidates through the use of ligand-protein docking [69].
In addition to parallel clusters and grid computing approaches,
there is always interest in supercomputing efforts as well. The most
notable current example of this is the BlueGene project at IBM [70].
This system consists of cabinets of 1024 nodes of two processors per
node. An important aspect of this system is the existence of multiple
high-performance interconnection networks between the processors
that allow scaling to over 64,000 nodes. These include both a nearest
neighbor type of topology and a tree-type topology.
Advances in Theoretical Methods
Among the most important theoretical advances for the future will be
the development of better interaction models. These will take several
forms. It may be the case that further improvements can be made
using the functional forms in the current generation of fixed charge
force fields. With the computational capacity now available, effort
could be spent performing a thorough assessment of these force fields
for their ability to predict a variety of biologically important proper-
ties. Second, after quantifying inadequacies exhibited by these force
fields, one needs to consider what additional physical phenomena need
to be included, such as electronic polarization. Considerable effort