This project will design algorithms for the advanced scheduling of communications and computation resources. A growing number of scientific computing applications require reliable services involving large amounts of data. These services may range from the simple transfer of a large data set to the complex execution of a collection of interdependent tasks that have varying requirements with respect to processing, storage, and communication. The support of such services requires a robust computing and network infrastructure that is capable of providing dynamic on-demand provisioning of bandwidth and scientific computing resources. The basis of this infrastructure is a collection of scientific computing resources that are interconnected by optical networks. Many on-demand provisioning techniques for optical networks focus on immediate reservations, in which resources are reserved and used immediately following a request; however, immediate reservations may be unable to guarantee the availability of resources with high probability, particularly if a significant amount of resources are required for an application. Advance reservation mechanisms enable dedicated resources to be reserved by an application in advance with the task not starting execution until it is possible to guarantee the availability of the needed resources. Advance reservation also allows more efficient use of resources and may be applied to both optical network resources and scientific computing resources in a coordinated environment.
The project will create a holistic multi-layer coordinated framework to support the coordinated advance provisioning of Grid and optical network resources. Specific goals of the project include 1) the development of theoretical bounds and algorithms to support generic advance reservation of optical network bandwidth resources by considering the time, space, bandwidth, and destination domains, 2) the design of a novel delayed allocation algorithm for supporting advance reservation requests, 3) the design and evaluation of coordinated Grid and network resource scheduling algorithms that incorporates different domains of flexibility, such as lightpath switching and supports enhanced services, such as anycasting, and 4) the design and implementation of the proposed advance reservation and co-scheduling algorithms on the nation-wide Energy Sciences network using the OSCARS framework.
Broader Impact: The integrated outreach and group mentoring program will promote female students' recruitment and retention in computer science and engineering disciplines. The implementation of the CARGONET algorithms on the OSCARS framework will assist in conducting efficient resource allocation of optical and Grid resources.
