Jai Dayal

School of Computer Science

College of Computing

Georgia Institute of Technology

Date: Friday, October 28th, 2016

Time: 11:00am

Location: KACB 3402

Committee:

---------

Dr. Matthew Wolf (Advisor and Committee Chair, Georgia Tech and Oak Ridge National Laboratory)

Dr. Karsten Schwan (Advisor, School of Computer Science, Georgia Tech)

Dr. Ling Liu (School of Computer Science, Georgia Tech)

Dr. Santosh Pande (School of Computer Science, Georgia Tech)

Dr. Ada Gavrilovska (School of Computer Science, Georgia Tech)

Dr. Jay Lofstead (Computer Science Research Institute, Sandia National Laboratories)

Abstract:

--------

The trend to exascale is causing researchers to rethink the entire computational science stack, as future generation

machines will contain both diverse hardware environments and run times that manage them. Additionally, the

science applications themselves are stepping away from the traditional bulk-synchronous model and are moving towards a more

dynamic and decoupled environment where analysis routines are run in situ alongside the large scale simulations. 

This thesis presents CoApps, a middleware that allows in situ analytics applications to operate in a

location-flexible manner. CoApps also explores methods to extract information from, and issue management operations

to, lower level run times that are managing the diverse hardware expected to be operating on next generation exascale machines.

This work leverages experience with several extremely scalable applications in materials and fusion, and has been evaluated on

machines ranging from local Linux clusters to the supercomputer Titan.