Skip to main content
more options
intel cluster

One of the Intel Equipment donations at work.

Why "Fusion"? (Think cuisine, not physics. We meld cultures and skillsets.) We work on software and hardware (hence the yin/yang symbol), focusing on the interactions between them and the fuzzy lines that divide them. Projects are inter-related (back to the fusion theme), and include: designing memory hierarchies for many different computing models and platforms; applying machine learning to microarchitectural components; optimizing applications, middleware, and operating systems for improved memory performance on new or existing system designs; and various aspects of performance monitoring and analysis.

Much of this work is collaborative with colleagues at Cornell ECE and CS, the Lawrence Livermore National Lab, the Georgia Institute of Technology, the University of Oregon, Florida State University, and the University of Michigan.

Our work is sponsored by Cornell University (through ELI and PCCW fellowships), the National Science Foundation, the Department of Energy (through a Krell Institute Fellowship and LLNL internships), and multiple, generous gifts from Intel Corporation.

System-Wide Hardware/Software Performance Monitoring and Adaptation


We are building a flexible and unifying framework for non-intrusive hardware monitoring of virtually any system component. Whereas current systems provide only limited access to specific events (prohibiting system-wide event correlation and global state evaluation, and usually limiting the user to event sampling), we strive to provide:

  • System-wide, unified introspection as a building block for autonomic systems;
  • Data preprocessing on the actual monitoring probes (by leveraging reconfigurability);
  • Standardized access to performance information using high-level queries;
  • Integration of hardware and system software probes in a common framework;
  • Correlation of performance information from several sources to assemble global system state for requested metrics;
  • Autonomic optimization of system components, including the Operating System (OS), runtime, and compiler; and
  • Ability to optimize for performance, power savings, heat dissipation, security, and reliability, all within a single framework.

This is sponsored by an NSF ITR/NGS award, and is joint work with Hsien-Hsin Lee at the Georgia Institute of Technology. The simulation tools for this project run on clusters donated in large part by Intel Corporation

.

Scalable, Interoperable Tools to Support Autonomic Optimization of High-End Applications


We strive to enable practitioners to more easily build efficient, scalable applications, to solve very large and complex problems, and to do so more quickly than is currently feasible. Significant increases in the productivity of applications writers will not only enhance the development of scientific applications important to our national infrastructure, but will also open High-End Computing (HEC) to important economic and societal applications where computing is advancing science and technology.

This is work sponsored by an NSF ST-HEC award, and is joint with the University of Oregon and Florida State University. Tools developed for and used by this project run on clusters donated in large part by Intel Corporation.

Memory Systems Pot Pourri

Look at recent publications, or send email. Here are the teasers:

We do memory systems for symmetric multiprocessor systems (the building blocks of most High-End systems) and Chip Multiprocessors, from protocol verification to coherence controller and memory controller design and evaluation.

We're working on some cache-conscious data placement, a little funky feedback-directed prefetching, and some region cache work in collaboration with Gary Tyson, at Florida State University, and Mike Geiger, at the University of Michigan. Most of this is in conjunction with one or more of the other projects.

We're working with LLNL on source- and compiler-level optimizations to very large, irregular codes. This research is sponsored in part by LLNL themselves, and in part by a DOE Krell Institute High-Performance Computer Science graduate fellowship.

Simulation Tools

Work with the Keshav Pingali's ISS group (in which Paul Stodghill is a major contributor, and by whom the C3BASE software releases are maintained), has produced an interesting simulation technology. We call it SimSnap, and it leverages ISS's Application Level Checkpointing (developed for software fault tolerance) to fast-forward applications to a simulation point (or wherever the user wants to put the application "under the microscope"). Taking checkpoints while running natively allows the application and its entire state to be loaded into a simulator, where execution may continue. Likewise, taking a checkpoint from within the simulated code allows another turn of native execution (which, in turn, makes it possible to run even a large application to completion to check that correct results are produced). Portable checkpointing technology will allow this simulation technology to allow the native execution and simulated machine model to differ (some restrictions apply, of course).