Logistical Networking in Distributed High Performance Computing

Colloq: Speaker: 
Scott Atchley
Colloq: Speaker Institution: 
University of Tennessee, Knoxville, Department of Computer Science
Colloq: Date and Time: 
Thu, 2005-02-10 10:00
Colloq: Location: 
ORNL, Bldg. 5700, Room L204
Colloq: Host: 
Jeffrey S. Vetter
Colloq: Host Email: 
Colloq: Abstract: 
Today's leading edge simulation scientists have access to ever more powerful supercomputers and large-scale clusters to push their research forward. These simulations can create terabytes and soon petabytes of output. The science teams may span multiple organizations and/or institutions. Members of the team may have additional local resources for analysis and visualization which requires moving some or all of the data between the various locations. Even though multiple high-capacity networks exist (e.g. ESnet, Abilene, DREN) between supercomputing sites, researchers still encounter difficulties taking advantage of these networks. Most researchers still use basic data management tools (e.g. ftp and scp) which do not give high-performance. Even if they can move the data, the researcher may not have enough local storage resources available to hold his data in order to analyze or visualize the data on his local cluster. This talk will introduce the concepts and basic mechanisms of Logistical Networking, as well as how it is being used by researchers at ORNL, NERSC, and PPPL to assist their Distributed High Performance Computing.
Colloq: Speaker Bio: