Interconnection Network Group
Department of Computer Engineering and Computer Science
University of Missouri - Columbia
jurczykm@cs.missouri.edu
Interconnection networks are used both in parallel processing and in communication systems
to exchange information.
While the network topologies used in both systems are similar, the performance
characteristics highly depend on the data traffic scenarios that are very different
in those systems. In parallel processing systems, a major concern is the low communication latency,
while in communication systems, additional aspects such as quality-of-service (QoS) characteristics (including
delay jitter and cell loss) come into play.
This group is concerned with modeling, performance assessment, and performance
enhancement of interconnection networks.
In the following, research projects are listed that this group is currently interested in.
Distributed Heterogeneous Networking Environments
Adaptive Information Control Techniques for the Agile Information Control Environment
One of the major aims of the Agile Information Control Environment (AICE)
program currently developed by DARPA/ITO is the dynamic control of information
flows stemming from users' requests, across heterogeneous military and commercial networks, in support of
military operations. These information flows could be preemtive or in response
to real-time requests by users. Each request might have an associated priority, deadline, and
quality of service (QoS) attributes. When more requests are posted than the system can handle,
requests get dropped and/or modified (e.g., through QoS degradation).
This research examines satisfying users' requests that have priorities and deadlines,
in a fashion that allows the system to achieve a near maximum global quality of performance (QoP).
In the AICE system, the network interface is handled by the MetaNet. The Adaptive Information Control (AIC)
layer of the AICE is responsible for presenting channel requests to the MetaNet
in order to allocate network resources for those channels. The MetaNet attempts to
create the channels requested. It might preempt existing channels to free enough resources to
accompany the new channel. In order to maximize the QoP, the AIC layer has to decide which
channel requests to present to the MetaNet. Only the MetaNet has complete knowledge about the
underlying network topologies. This knowledge is not present in the AIC layer and has to be learned
by the AIC layer over time to make intelligent channel choices.
Current research is focused on (1) learning the network topology over time in the AIC layer, (2) aggregate
the underlying network topologies into a simplistic resource picture that is presented to the AIC layer ,
and (3) develop incremental mechanisms to temporarily map new requests onto existing channels
during a military situation change when there is not sufficient time available to set up
new channels.
Fault-tolerant data staging
The DARPA Battlefield Awareness and Data Dissemination (BADD) program
includes designing an information system forwarding (staging) data to
proxy servers prior to their usage by a local application, using
satellite and other communication links.
Data staging is an important data management problem for a distributed heterogeneous networking
environment, where each data storage location intermediate node may have
specific data available, storage limitations, and communication links.
This research focuses on the fault-tolerance of the data staging problem.
In this case, fault-tolerance refers to the ability of re-acquiring data
in a timely manner in the event of partial or total data loss at the receiving station.
Interconnection Networks for Parallel Processing Systems
Parallel simulation of interconnection networks
Simulation is inevitable for dimensioning interconnection networks
for parallel processing and communication systems.
Even if the network topology is simple and highly regular, its size
and the needed simulation accuracy result in high performance
requirements of the simulator, creating a need for parallel simulation
as a means to reduce turnaround times.
So far, this research has resulted in the implementation of a
parallel time-driven packet-switching and wormhole-routing multistage
interconnection network simulator on the MasPar MP-1 SIMD machine
and a parallel event-driven wormhole-routing direct network simulator
running on the Intel Paragon MIMD machine.
Nonuniform traffic patterns in packet-switching and wormhole-routing interconnection networks
Nonuniform traffic patterns such as hot-spot traffic or certain permutations
can degrade the overall performance of the interconnection network in a parallel
processing system that might result in a performance loss of the overall system.
This research focuses on the performance enhancement of packet-switching and wormhole-routing
interconnection networks under those traffic scenarios. Several enhancement
mechanisms have been proposed already.
ATM Networks
Mechanisms to enhance the quality of MPEG video over ATM networks
The quality of video streams delivered over asynchronous transfer mode (ATM) networks highly
depends on the cell loss probability and delay jitter introduced by the ATM network.
Protocol and hardware mechanisms are needed that guarantee small cell loss and
delay jitter.
We are in the middle of enhancing an ATM simulator to incorporate
MPEG video streams, leaky bucket mechanism, etc. This simulator
will serve as a tool to assess the performance and influence on
video quality of priority mechanisms on the protocol and ATM switch
levels.
This page has been accessed at least
times since the counter was last reset.