This ITR proposal is a request for funding to address the two critical scaling problems for large scale grid computing. Experience with the current generation of deployed grid systems makes it clear that, while these systems are practical for tens of sites, with tens of full-time managers, they do not scale to the hundreds or thousands of sites required by large scientific collaborations or to larger global scales.
The two essential non-scaling effects are: (a) Software Environment Computing (SEC): a lack of well-defined mechanisms for constructing and manipulating software environments on remote sites reliably, safely and without human intervention is a major source of non-scaling behavior. The solution proposed is to define and implement an infrastructure where software environment creation and manipulation are basic elements of grid computing. They propose to use the Pacman package as a first implementation of SEC; (b) Effective Resource Allocation (RA): a lack of grid-level mechanisms for controlling allocation of resources will cause grids to become difficult to manage and inefficient for typical users as the grids scale up in size. The solution proposed is to introduce a common grid currency to bootstrap an economy and promote policy decisions. Both SEC and RA require new approaches in order to achieve a qualitative leap forward in grid computing and in IT. Definitive solution of these problems would be a major intellectual advance in IT providing both theoretical understanding and practical tools that would be the basis of further research.
Broad Impact: By identifying and unblocking the essential problems in contemporary grid computing, this work will enable grids to grow in size towards limits that we are not currently able to calculate. A dramatic growth in grids will result in near SETI-at-home scale computing for broad ranges of scientific and other computing projects and would constitute a true change in the national computing landscape.
