Background Science Applications Physics Climate Groundwater Fusion Energy Life Sciences Materials & Chemistry SciDAC Institutes Collaboratories Enabling Technologies Applied Mathematics Computer Science Visualization & Data Mgt. SciDAC Outreach Participating Orgs Grant Solicitations FY2007 FY2006 FY2005 FY2004 FY2001 Collateral Materials SciDAC Review magazine '06 Progress Report (pdf) Publications 2001-5 (pdf) Logos Search Collateral (from SciDAC 1) Browse by Affiliation Browse by Author Browse by Title DOE Office of Advanced Scientific Computing Research (ASCR) Program Managers only

INCITE—Edge-Based Traffic Processing and Inference for High-Performance Networks

Richard Baraniuk, Edward Knightly, Robert Nowak, Rolf Riedi, Rice University
Wu-chun Feng, LANL
Les Cottrell, SLAC

Summary

The explosive growth of high-speed computer networks, combined with rapid and unpredictable developments in applications and workloads, has rendered network modeling, control, and performance prediction increasingly demanding tasks. High-end applications critical to the DOE mission, including distributed computation, remote visualization, and high-capacity data transfers, routinely fail to meet end-to-end performance expectations when deployed on high-speed networks. The INCITE (InterNet Control and Inference Tools at the Edge) Project aims to transform modern high-speed inter-networks into manageable and predictable systems to enable these critical applications. Our interdisciplinary team is developing new theory and methods for network monitoring, probing, and analysis based solely on edge-based measurement at hosts and/or edge routers.

Distributed applications running on clusters and computational grids are complex and difficult to analyze. Moreover, optimizing their performance requires that end-systems have knowledge of the internal network traffic conditions and services. Without special-purpose network support (at every router), the only alternative is to indirectly infer dynamic network characteristics from edge-based network measurements. There is a great need for theories and methods to understand the complexities of distributed applications and network environments with the ability to choose the level of detail to fit the task, be it debugging, tuning, monitoring, or control.

The INCITE Project is developing on-line tools to characterize and map host and network performance as a function of space, time, application, protocol, and service. In addition to their utility for trouble shooting problems, these tools will enable a new breed of applications and operating systems that are network aware and resource aware.

Monitoring tools
INCITE’s monitoring tools include MAGNET (Monitoring Apparatus for General kerNel-Event Tracing), MUSE (MAGNET User-Space Environment), and TICKET (Traffic Information-Collecting Kernel with Exact Timing). Together they act as a kind of “network oscilloscope” that can measure (capture packets) at different points in a host, cluster, or network, from the application to data link layer.

MAGNET and MUSE permit applications and developers to obtain detailed information about the environment on a host and enable new resource aware applications that adapt to changes in their environment (load balancing when needed, sensing when a node’s resources are scarce or are bottle-necked, and so on). MUSE monitors without requiring modification or re-linking of applications. TICKET serves as a high-speed “tcpdump” replacement.

Edge-based probing tools
INCITE’s probing tools include PathChirp, NetTomo, and NetTopo. Much as x-rays probe our bodies to find cracks and breaks in bones, these tools inject probe packets into a network to determine its conditions and characteristics. PathChirp accurately estimates the available bandwidth and delay distribution along an end-to-end path using an efficient exponentially spaced train of probe packets. And much as x-ray tomography reconstructs a 3-d internal view of a person, NetTomo localizes delays and losses on individual network links by injecting probes along multiple network paths. NetTopo allows a user to discover the internal topology of a network through edge-based probing. These algorithms are being incorporated into the PingER measurement toolkit; data will be distributed by the Globus MDS or similar mechanisms.

Simple example of a topology map from
SLAC to various North American sites, with color indicating service provider.

Edge-based probing enables applications to become network aware. We are working to optimize applications such as GridFTP or bbcp bulk file copy/transfer for the Particle Physics Data Grid (PPDG). We are also developing sophisticated new multiscale traffic models and analysis software based on multifractals and wavelets.

Current INCITE users include: Particle Physics Data Grid Collaboratory Pilot, Scientific Workspaces of the Future, SciDAC Center for Supernova Research, TeraGrid, Transpac at Indiana U., San Diego Supercomputing Center, ns-2 project, Telecordia, and CAIDA. We hope to develop a 64-node experimental cluster to test our theories and tools in both a cluster and grid environment (using a high-performance WAN infrastructure – OC-192 or multiple OC-192s). For more information, see the INCITE website at http://www.ece.rice.edu/INCITE/.

For further information on this subject contact:
Dr. Thomas Ndousse, Program Manager
High-performance Networks Research
Phone: 301-903-9960
Tndousse@er.doe.gov

back to project page