Warren Smith  Senior Computer Scientist  Computer Sciences Corporation wwsmith@nas.nasa.gov  phone: (650)604-0521  fax: (650)966-8669  M/S 258-6  NASA Ames Research Center  Moffett Field, CA 94035

I am a contractor in the NASA Advanced Supercomputing Division of NASA Ames Research Center. Before that, I was a Ph.D. student at Northwestern University and supported by Argonne National Laboratory.

Research Interests

My research interests lie in the area of computational grids and, until our recent division reorganization, I was working in three areas:

1. Monitoring and management in distributed environments. I'm developing a framework for observation and control of resources, services, and applications in distributed environments such as computational grids. As part of this work, I led several working groups and co-authored two informational documents as part of the Grid Forum. This software has been released open source and is available here.
2. Distributed information services. In this work, I investigate and evaluate approaches for providing information in distributed environments. This includes investigating information models, algorithms, and implementations to determine their performance and overall suitability for both NASA's computational grid, the Information Power Grid (IPG), and for earth- and space-based participants in NASA missions.
3. Scheduling in computational grids. I'm investigating both system-level scheduling and user-level scheduling.
• I am working with Rupak Biswas of NAS and with Leonid Oliker and Hongzhang Shan of Lawrence Berkeley National Laboratory to develop and evaluate system-level grid scheduling algorithms. These algorithms consists of user-independent schedulers that exchange jobs to optimize turn-around time. These algorithms consider factors such as computation power of systems, network bandwidth between systems, the amount of computation to be performed by each application, and the amount of input and output data accessed by each application.
• I also investigate user-level scheduling techniques. In this type of scheduling, the scheduling systems for parallel computers or clusters are not modified. Instead, users have their own schedulers that will examine the state of various systems and schedulers and will select which system to submit an application to. In this work, I investigate techniques for predicting the execution time of applications on different systems and how long applications will wait in scheduling queues before they begin to execute.

For my thesis I looked at several problems related to resource selection and scheduling in computational grids:

• Predicting application run times using historical information
• Predicting queue wait times
• Using run-time predictions
• Using historical information about scheduler state
• Improving the performance of scheduling algorithms using run-time predictions
• Efficiently combining queuing scheduling with reservations

Projects

Until recently, I was involved in NASA's Information Power Grid project that is building an infrastructure for distributed, high-performance scientific computing. As part of this overall project, I was working on projects relqted to my research interests described above.

Before that, I was in the Globus group in the Mathematics and Computer Science division of Argonne National Laboratory. The Globus project has very similar goals to the IPG. In fact, the IPG makes use of the Globus toolkit.

Previously, I worked on the I-WAY project which was a testbed for wide-area ATM networking and metacomputing software environments that was demonstrated at Supercomputing '95. As you can tell, I just can't get enough of those computational grids!

Recent Papers

Feel free to learn more about my work by reading some of my recent papers.