|
Computer Science Department
Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213-3891 |
|
Office: (412) 268-1425
Mobile: (412) 606-7892 Fax: (412) 268-5576 bumba@cs.cmu.edu URL: http://www.cs.cmu.edu/~bumba/research/ Citizenship: Indian, Visa status: F1 student |
Operating systems; networking; mobile and wearable computing
Carnegie Mellon University, Pittsburgh, PA
Ph.D. in Computer Science, July 2002 (expected)
Thesis:``Operating System Support for Mobile Interactive Applications''
Advisor: M. Satyanarayanan
Indian Institute of Technology, Madras, India
B.Tech. in Computer Science and Engineering, May 1995
1995-96 AT&T Asia-Pacific Scholarship (one of 36 students from 12
Asia-Pacific countries)
1989-95 National Talent Search Scholarship (one of 750 students
nationwide)
Mobile interactive applications such as augmented reality rendering
and speech recognition must provide bounded interactive response times
for a good user experience, and limit their energy consumption to
prolong battery lifetimes. To do this, they must dynamically adapt
resource demand to resource supply, which varies continually due to to
changes in load, processor frequency, wireless connectivity, server
availability, and battery charge levels. Previous research has shown
that application adaptation can reduce the bandwidth and energy usage
of data access. In my thesis research, I introduce multi-fidelity
computation: a general model for adaptation that covers a wide range
of adaptive behaviors. I also present a resource model that can
uniformly represent system resources -- CPU, network, memory, disk,
file cache state, and remote server resources -- in terms of supply
and demand.
I have designed an API and built system support for multi-fidelity computation in Linux. The multi-fidelity API is small and designed to minimize the modification required to port legacy code. The system support provides a novel service: automatic learning and prediction of application resource demand. This shifts much of the burden of adaptation from the application programmer to the system: the programmer describes the application at a high level, and does not need to know the exact relationship between application behavior, resource demand, and performance. The API and system support have been tested with a number of interactive applications, including 3-D rendering, speech recognition, web browsing, and language translation. In my thesis research I show that applications are able to satisfy latency constraints through adaptation. E.g., we reduce the mean response time for a virtual walkthrough application by 3x, and the variance by 42x. Other researchers have used the same infrastructure can to achieve users' battery lifetime goals through adaptation.
Studying and interpreting the behavior of a mobile computing system under conditions of motion is difficult. We implemented a wireless obstacle course to facilitate such evaluations. The infrastructure consists of a physically compact, limited-access region over which the quality of wireless network performance can be controlled. As a test system moves through the obstacle course, it perceives wireless network quality that varies spatially according to a pattern set up by the experimenter. Our prototype implementation uses Active Badges to sense the location of the mobile computer; a network trace modulation layer then emulates the wireless network characteristics for that location.
As part of the Storage Systems group, I implemented a migration planner for Forum. Forum automatically computes a configuration for a large-scale storage system based on workload properties and performance requirements. It must periodically recompute the configuration to track workload or system changes; the migration planner then generates a plan to move data objects efficiently to the new configuration.
Odyssey is a system support layer for adapting data fidelity to network bandwidth availability. As part of the Odyssey group, I implemented various system components and an adaptive video player.
Professor M. Satyanarayanan
School of Computer Science
Carnegie Mellon University
5000 Forbes Ave
Pittsburgh, PA 15213
(412) 268-3743
satya@cs.cmu.edu
Professor Christos Faloutsos
School of Computer Science
Carnegie Mellon University
5000 Forbes Ave
Pittsburgh, PA 15213
(412) 268-1457
christos@cs.cmu.edu
John Wilkes
Hewlett-Packard Laboratories
Mailstop 1U-13
1501 Page Mill Road
Palo Alto, CA 94304
(650) 857-3568
wilkes@hpl.hp.com