submenu »

project home | overview | downloads | publications | related links

The high-performance networking (HPN) group has a long and successful history on research with high-performance computer networks (e.g. system-area networks, local-area networks, etc.) for high-performance computing and high-performance embedded computing. For the past decade, the group has been working with a variety of cutting-edge HPN technologies such as 10 Gigabit and Gigabit Ethernet, InfiniBand, RapidIO, Scalable Coherent Interface (SCI), Myrinet, Fibre Channel, ATM, SuperHIPPI, Giganet cLAN, Synfinity, etc. A broad range of testbed experiments, coupled with the development of a number of simulative and analytical models for HPNs, have led to new and better insight about the inherent performance characteristics and tradeoffs of HPN protocols and technologies for application in general-purpose HPC systems as well as embedded and real-time systems. Below is a summary of projects currently active in the group.

Increasing the Energy Efficiency of the Internet with a Focus on Edge Devices

This new joint project at UF and USF, funded by NSF, addresses the increasingly critical need to improve the energy efficiency of the Internet by focusing on the primary and often neglected energy consumer, edge devices. Studies at LBNL show that about 74 TWh/yr of electricity (costing ~$6B) is consumed by the Internet in the USA alone, of which about a third could be saved with full use of power management on desktop computers, the most common of edge devices on the Internet. Unfortunately, due to limits of existing protocols and architectures, networked desktop computers typically remain powered-up during frequent and often lengthy periods of idleness. As network devices, they are prevented from operating in an energy-efficient manner due to their need to respond to network transactions of various types without warning. Our approach to addressing this challenge is to investigate and exploit a synergistic set of novel research concepts for protocol and subsystem infrastructure, and algorithms for effectively controlling them based on traffic and system constraints, so that edge devices can be put to sleep during periods of relative idleness while network connectivity is maintained by a low-power hardware proxy integrated into the system. Our approach also promises to provide additional increases in energy efficiency by reducing consumption of network-related resources during active periods where graceful degradation of performance is acceptable, in effect trading off speed for energy. In addition to desktop computers, this approach may also be amenable to a wide variety of emerging wired and wireless edge devices such as television set-top boxes, network appliances, remote cameras, etc.

WDM Fiber-Optic Network Architecture Analysis, Modeling, Optimization and Demonstration for Aerospace Platforms

This new project funded by the U.S. Navy is undertaking the research, design, and development of key concepts, tools, and technologies for local-area optical networking, based on wave-division multiplexing, specifically targeted towards existing and emerging requirements for communication networks in advanced aerospace platforms. In the first phase of the project, the emphasis will be on the identification and integration of disparate requirements, the development and analysis of computer-based simulation models for candidate topologies and control methodologies, the determination of key metrics of performance, scalability, dependability, power, etc., and optionally the modeling and mapping of one or more legacy and emerging network protocols into the simulation environment. The results from this phase will lay the foundation for follow-on phases where candidate system and subsystem architectures will be developed and evaluated via advanced techniques in rapid virtual prototyping, network node requirements will be determined, and node and small-scale network prototypes will be constructed to demonstrate efficacy and properties of the approach in meeting the needs of advanced Navy aerospace systems.

In addition, the HPN team is currently and has for many years been active with Beta testing of new HPN techologies in collaboration with several leading HPN vendors, such as Ammasso, Nortel Networks, Cisco Systems, Matisse Networks, etc.

A number of collaborative developments are also underway in the HPN group, such as support for LBNL/UC-Berkeley in their testing of UPC on AlphaServer ES80 (Marvel) and Opteron cluster platforms in our testbed, as well as continuing work on an extension of UCB's GASNet communication layer for Berkeley UPC with a conduit for SCI, based on products from Dolphin. Additionally, the HPN group has been active in assisting with UPC activities at other institutions, such as cooperative UPC testing on Marvel with UPC groups at Ohio State University and Michigan Tech.

OTHER HPCC Group PROJECTS

Reconfigurable Computing (RC) Hardware Empowered Grid Computing

Parallel and Distributed Computing for Fault-tolerant Sonar Arrays

Computational Framework for Simulating Joint Mechanics

GEMS Project