pete

 CV | PapersPeter J. Keleher 
Email:keleher @ cs.umd.eduPGP Key
Position:Associate Professor, CS and UMIACS,
Univ. of Maryland at College Park.
Teaching:412
818
Other:book club (mail server)
Office:4157 A.V. Williams Bldg.;
Phone: (301) 405-0345;
Fax: (301) 405-6707
Address:Dept. of Computer Science,
University of Maryland,
College Park, MD 20742-3255

Research Interests

My research is mostly in the field of distributed systems. Loosely defined, this encompasses work in operating systems, runtime systems, and distributed object and database systems. In all cases, our approach is to identify problems, propose solutions in the form of specific policies and mechanisms, and then empirically evaluate them. We are systems builders: very few projects end in analysis or simulation. The systems that we study often operate in fluid, dynamic environments. Simply put, my group tries to make systems operate better in such circumstances. Central to our approach is the notion of building systems that transparently adapt applications to changing conditions.

UbiOS will provide versioned, ubiquitous object support for a variety of systems, including interactive paper-based systems, anytime/anywhere annotation and blogging, and wide-area file systems.

MoteFS provides fast, flexible, and secure access to remote files over untrusted wide-area networks. As part of this effort, we are developing fine-grained namespace mechanisms, decomposable credentials, cross-server versioning abstractions, client-side differencing techniques, and wide-area prefetch algorithms. The end result of this work should be a file system that performs significantly better than previous work in wide-area settings.

TerraDir is a distributed peer-to-peer directory protocol that can be used as the basis for implementing customized directories for Internet applications. TerraDir is based upon the twin tenets of scalability and flexibility, and will scale across multiple dimensions including the number of directory entries, the number of nodes the directory is distributed across, and available network, processing, and storage resources.

Older work: deno, harmony, and cvm.