I am coordinating Parallel ARCOS, a group of researchers from ARCOS investigating on High Performance Computing and Parallel I/O for large-scale distributed memory supercomputing architectures. The aim is to provide a collaborative environment that propitiates cross-fertilization of ideas among researchers working in affine topics. The main research themes are: parallel I/O scheduling, auto-tuning of the HPC I/O stack, malleable HPC computing, energy efficiency, massively parallel applications, communication optimizations.
Decaf: High-Performance Decoupling of Tightly Coupled Flows
Decaf is a collaborative project (Argonne and Sandia National Laboratories) investigating novel techniques of decoupling scientific workflows based on a three layer stack (a transport layer, a dataflow model, and a data description layer) and cross-layer resilience. I am involved in supporting the dataflow model through multiple-layer coordinated data staging techniques.
Locality-aware scheduling in Swift workflows
Swift is a system for the rapid and reliable specification, execution, and management of large-scale science and engineering workflows. In this project we are investigating novel techniques of improving the performance of workflows by trade-offs between data locality and load balance (in collaboration with Argonne National Laboratory).
Joint Laboratory for Extreme-Scale Computing.
I am actively participating in the Joint Laboratory for Extreme-Scale Computing (University of Illinois at Urbana-Champaign, INRIA, the French national computer science institute, Argonne National Laboratory, Barcelona Supercomputing Center, Jülich Supercomputing Centre and the Riken Advanced Institute for Computational Science.)
Design and implementation of a hierarchical caching architecture for large-scale HPC infrastructures
This project investigates a multi-level hierarchy for improving the performance of asynchronous I/O on large-scale HPC architectures (in collaboration with Argonne National Laboratory)
AHPIOS: An Ad-Hoc Parallel Input/Output System
AHPIOS develops an on-demand elastic parallel I/O system for improving the storage I/O performance of parallel HPC applications (in collaboration with Northwestern University).
Clusterfile: A parallel file sytem for clusters.
Clusterfile parallel file system that leverages parallelizing compilers techniques for offering flexible high-performance access for common scientific patterns such as multi-dimentional arrays. Clusterfile is the first system to integrate cooperative caching, collective I/O, and flexible logical and physical data partitioning in novel high performance parallel I/O access techniques.