IQ-ECho: Interactive Quality of Service Across Heterogeneous Hardware

Karsten Schwan, Georgia Tech
Greg Eisenhauer, Georgia Tech
Matthew Wolf, Georgia Tech

Summary

This project addresses the efficient transfer of large data across wide-area networks, focusing on important scientific applications such as remote visualization and real-time collaboration. The IQ-ECho middleware mechanisms can involve both the application and the network level in adapting to changes in network and platform conditions, with a goal of enabling high-fidelity, end-to-end scientific collaboration, targeting distributed computing environments that range from Internet-based to high end wide area network infrastructures.

1. Introduction

To attain high performance in the real-time exchange of data across collaborating machines and end users, this project is developing and evaluating middleware methods and techniques for coordinating application-level with network transport-level adaptations of data communication. This basic approach complements previous work on TCP-friendly communication and on adaptive transport protocols, attempting to meet application needs without violating fairness in network resource usage.

2. IQ-ECho

The approach is embodied in the IQ-ECho middleware, which implements the distribution of scientific data to remote collaborators. Using IQ-ECho, application-level adaptations like selective data down-sampling are triggered by transport-level information provided by the instrumented IQ-RUDP protocol underlying IQ-ECho's communications. The application- to network-layer exchange of information necessary for such coordinated adaptations is implemented with quality attributes, which provide a lightweight way for an application to provide quality of service information and to describe its adaptation to the transport layer, and for IQ-RUDP to share network status information with an

In addition to triggering application-level adaptations and reacting to certain changes in network state, IQ-RUDP can also re-adapt its own communication behavior after an application adaptation has been performed, in part to remain fair to other network flows. Such transport-level reactions can be performed at higher rates and with smaller overheads than possible at application level.

3. SmartPointer and IQ-ECho

IQ-ECho is a key piece of technology behind SmartPointer , which serves as a driving application and has communication demands similar to wide-area collaboratories and tera-scale supercomputing ventures.

The SmartPointer application requires a communications infrastructure that can be flexible, adaptive, and yet support high performance. Traditional HPC-style communications systems like MPI offer the required high performance , but rely on the assumption that communicating parties have a priori agreements on membership lists and on the basic contents of the messages being exchanged. For the applications we target, however, both data types and subscription lists of communicating parties must be flexible. This need for flexibility has led some designers to adopt techniques like Java's RMI or meta-data representations such as XML+SOAP. These methods have high costs that interfere with total performance, because data marshalling becomes a key issue.

The IQ-ECho middleware addresses these concerns in several different ways. In particular, it provides the following:

•  Information flows are represented as event streams , using the publish/subscribe model.
•  There is no centralization; channels are represented by distributed data structures.
•  Connections are managed Subscribers preserve transparency of local versus remote receivers, with implementations for underlying peer-to-peer communications over TCP, UDP, Wireless, multicast, and others.
•  Dynamic extension of existing formats and discovery/operation on format contents is enabled through run-time dynamic code generation of subscriber-specified filters.

3. Milestones Achieved

The IQ-RUDP implementation has been integrated with several different network measurement packages, including those from Nagi Rao and Constantinos Dovrolis, both funded under the SciDAC project. This enables end users to choose their preferred bandwidth-measuring approach.

Further, a key point of integration this year was to enable resource-aware filtering inside of the SmartPointer application based on these network measurements. The visualization infrastructure has been modified so that it can automatically adapt to real-time changes in available bandwidth by modifying quality, color depth, or frame rate. The results of this work are being submitted as a paper to HPDC this year.

4. Future Work

Our future work with the evaluation of IQ-ECho services for remote collaboration will (1) expand upon our service-level adaptations that utilize different network-level techniques for assessing current network bandwidth, (2) use overlay networks to combine the lightweight data filtering and downsampling methods used in this paper with heavier-weight methods for data transformation and summarization executed by additional machines interposed into the path between data providers and consumers, and (3) integrate with 3 rd party collaboration infrastructures such as the Access Grid, also a SciDAC project. Such work will focus on high end links using the large data volumes produced by applications like the DOE Supernova Initiative. The intent is to use actual TSI data across the 10GB link and other links connecting GT with ORNL and other DOE sites.

Figure 1: Current implementation of the SmartPointer infrastructure, employed in a molecular dynamics visualization.

For further information on this subject contact:
Dr. Thomas D. Ndousse , Program Manager
High-Performance Networks Program
Mathematical, Information, and Computational
Sciences Division
Office of Advanced Scientific Computing Research
tndousse@er.doe.gov
Phone: (301)-903-9960

back to project page