Wireless Subnetworks across the Gulf of Mexico for the Next Generation Internet

Harold H. Szu\**, Nian-Feng Tzeng, and Gui-Liang Feng

Center for Advanced Computer Studies
University of Southwestern Louisiana
Lafayette, LA 70504-4330
Tel. (318)482-6304; E-mail: tzeng@cacs.usl.edu

** on leave to Naval Surface Warfare Center, Dahlgren Division, Virginia.


1.  Introduction

Wireless connections will soon play an important role in connecting sites,
which are otherwise too expensive and inconvenient to get access to NGI,
as the Federal Communications Commission (FCC) commissioner Susan Ness
has just announced that FCC has made it easier for schools, hospitals,
libraries and other non-profit organizations to access the internet by providing
additional 300M Hz bandwidth (between 5.15 to 5.35 GHz, and 5.725 to 5.825 GHz)
to transmit data, voices and images over the distances of up to several miles.
Such a distance is comparable to the up-link/down-link distance of a satellite
to reach the off-shore platforms in the Gulf of Mexico, through a unique USL
hub for allowing platform scientists/engineers to access NGI.

In this white paper, we describe a wireless testbed across the Gulf of Mexico,
formed by connecting some 6,000 off-shore oil/gas production platforms
through wireless intranet links to the NASA/NOAA satellites, to which
USL/CACS's newly established NASA/Regional Data Validation Center (RVC)
can offer real-time, direct read-out.
These wireless subnetworks across the Gulf will be hubbed at USL/CACS,
which serves as a gate to NGI for access to the supercomputing centers.
Network computing fits particularly well to this testbed,
as computing resources needed by oil/gas platforms can be shared and managed
through the USL/CACS's hub to lower the cost and maintenance/upgrade efforts.
This is made possible by taking advantage of NGI's high bandwidth.
All-connected Gulf platforms, as the proposed wireless subnetworks,
can make better business decisions in the oil/gas production, which has
already brought to the Federal OCS tax revenue of about $2.2B per annum.
We estimate that the cost of NGI R/D ($100M) will be returned to the IRS
in just one summer by this proposed Gulf of Mexico application alone,
as a result of additional revenues generated by more oil/gas production
during the summer season when hurricanes have consistently posed threats
to Gulf platform operation.

This white paper focuses on the issues specifically related to our proposed
wireless subnetwork testbed across the Gulf.
Solutions to these issues may also benefit the NGI in general.
The following issues are described in sequence:
(a) wireless subnetworks, (b) network computing,
and (c) scalability, reliability, and effective multicast.


2.  Wireless Subnetworks

The National Weather Services (NWS) gives a Meso-scale Forecast
(size about 16 km and beyond), whose pixel footprints are large enough
to cover several oil/gas platforms in Gulf of Mexico.
A larger safety margin is thus observed
in the issue of Gulf hurricane watch and warning.
There are low-tech weather systems on all the platforms, but no cooperation
linkage exists among the platforms owned by different companies.
NGI makes it possible to establish a high-tech dedicated, fine-resolution,
real-time hurricane tracking system, with all-connected linkage present
among platforms through wireless connections.
We propose to design a field experiment based on the NGI capability,
which can provide a finer resolution in the tracking and forecasting
of hurricanes to avoid premature or false evacuation of
people from oil/gas production platforms in the Gulf.
Being a geographically closest university to the off-shore production,
and having Ph.D. programs in Computer Science and Computer Engineering
over a decade, we shall involve in the design and establishment of
NGI wireless subnetworks, by which better recommendation can be made to
the some 6,000 off-shore oil/gas production platforms,
ensuring efficient business operation.
We propose to offer wireless intranet connections to all platforms in the Gulf
via the NASA/NOAA satellites direct read-out in real-time to USL/CACS,
where a NASA/RVC has been newly created.
The wireless subnetworks hubbed at USL/CACS are then connected via NGI to
the supercomputing centers on which a better resolution numerical hydrocode is
executed for hurricane tracking and prediction in a fine resolution
Micro-scale Nowcast (1 km).

Such a fine mesh numerical computation has been attempted by Navy
with some success in the context of numerical fleet weather forecast.
We wish to build upon the unique feature offered by NGI wireless connections
over the well established commercial infrastructure in the Gulf of Mexico.
This NGI opportunity gives us the ambition to investigate whether or not
we can better understand the Gulf hurricanes
and better serve thousands of special clients in the Gulf,
by means of the hydrocode on a supercomputer with the real-time inputs of
fluid dynamics variables obtained from the on-site measurements of vertical wind
and temperature profiles on thousands of platforms with accurate GPS.
The archival data of the past hurricane history in the Gulf for the last two
decades will also provide us an important lesson from the statistical
perspective.

Adaptive wavelet transforms (AWT) [LiSz96, SzTG96] will be applied for
feature-preserved image compression and error-resilient transmission to
achieve improved bandwidth over wireless subnetworks.
The network user's profile is easier to be compiled by employing the
Artificial Neural Networks (ANN) approach [Szu87] pioneered by Szu in 1987.
This profile offers needed information to make better network management.
It facilitates guaranteed QoS (Quality of Service) over the network,
by reserving resources for anticipated requests without over-commitment.
The approach will lead to improved resource utilization in the network
and is being pursued by us.


3.  Network Computing

Our proposed wireless subnetworks make it possible for the Gulf off-shore
oil/gas industrial users to take advantage of the NGI capability.
Network computing is particularly suitable for oil/gas platform users,
because (a) they work with a limited set of programs and rarely need
new applications, (b) they are in the Gulf and difficult to support on-site,
and (c) they enjoy higher security by keeping data and software in the hub,
so that the loss of a computer system at the platform
would not be potentially catastrophic.
Software upgrades can be carried out easier this way,
lowering the support and maintenance cost.
Storage and computation power can be shared, better utilizing resources.

As the hub of wireless subnetworks, USL/CACS offers some computation power
through parallel processing on resident computers/workstations,
and also may provide expertise for improving or developing software.
Those computation-intensive applications, such as the hydrocode for
whether prediction and the codes for oil/gas exploration analysis,
are to be executed on high-performance systems at supercomputer centers
connected by the NGI.


4.  Scalability, Reliability, and Effective Multicast

Network switches for the NGI must be self-routing, without the central
controllers for establishing communication paths, and be scalable and
highly reliable in order to accommodate growth and
support mission-critical applications.
We are investigating a self-routing, scalable switching architecture,
which is on the basis of "cascaded" indirect cube structures [Peas77],
capable of achieving any specified message dropped rate.
This switching architecture requires fewer stages than the open-loop
shuffleout structure [BaDe94], and its building block is far simpler,
without the logics for calculating the routing tag of a message
at every stage as the shuffleout design.
This distributed routing switch is better scalable (up to thousands of ports
needed for connecting platforms in the Gulf) and can tolerate failures [TzYZ88],
presenting a more cost-effective design suitable for the NGI implementation.

Reliable information transmission
is typically achieved using certain error control codes.
Since ATM is expected to be widely deployed in the NGI,
we address the issue of reliable transmission over ATM networks.
On entering the ATM network, a frame of variable length (up to 64K bytes)
is divided into fixed length segments (of 48 bytes) at the transmitter
by the ATM Adaptation Layer (AAL).
For each segment, AAL 3/4 adopts the CRC of at least 9 bits
for detecting any corruption in the 48-byte payload of a cell.
At the frame level, there is a 32-bit CRC to detect errors in each frame.
Simmons and Gallager proposed to add 34-bit CRC to each frame [SiGa94]
for error detection.
For the fast NGI, and in particular, over the wireless connections
(like the subnetworks formed across Gulf platforms), however,
error detection is inadequate, since a single bit error is non-negligible
and will destroy the whole frame.
A new error control scheme is therefore highly desirable for the frame data.
As a CRC with length at least 9 bits is present in each segment,
a cell data error (whether random or burst one) can be detected,
rendering the error at the frame level an erasure error,
which can be easily corrected.
We propose to add redundant bytes in each frame to store
the erasure-only correcting code, which will be
based on the Improved Algebraic-Geometric (IAG) codes [FeRa94, FeRa95],
since such a code offers high flexibility in the choice of code length,
essential in this application because the frame length is not fixed.

The multicast operation is basic in support of
efficient multi-party communication services and is indispensable
for our proposed wireless subnetworks across Gulf platforms,
which belong to different companies and those platforms of the same company
may require to receive a new set of data or application codes at the same time.
A multicast group is often set up dynamically, allowing the multicast
participant(s) to be removed or added duing a communication session.
Multicast participants can be specified by a Steiner tree,
and the problem of updating the multicast tree after each addition or deletion
in the network is known as the dynamic Steiner problem.
Recently, a rearrangement inexpensive edge-based on-line Steiner algorithm
has been proposed [BaVa96] for the problem, requiring low computation with
as few modifications to the existing multicast tree as possible on
each update and offering a provision to rearrange the Steiner multicast tree
should its performance fall below a specified bound, through the reconstruction
of a so-called "modified region" (whose accumulated damage is excessive).
We are extending the "modified region" into a "reconstructing zone"
which is specified by a parameter, with a larger parameter value signifying
a bigger zone, and thus a better reconstructed multicast tree
involving a higher computation cost.
The parameter allows one to make the trade-off between
performance and cost in different situations.
All edges connecting nodes within the reconstructing zone and
nodes outside the zone are removed, and then a shortest path algorithm
is applied to reconnect the node pairs with shortest paths.
The shorest path algorithm computes distances between "selected" nodes
inside the reconstructing zone and ones outside it;
the number of nodes selected dictates the reconstruction cost involved and
also the quality of the reconstructed result.
It is important to evaluate different rules for rearranging the multicast tree
after repeated updates, under various parameter values and selected nodes sizes.
Our research work has arrived at promising and interesting outcomes
on effective multicast tree rearrangement.


References

[BaDe94]
S. Bassi, M. Decina et al.,
"Multistage Shuffle Networks with Shortest Path and Deflection Routing for
High Performance ATM Switching: The Open-Loop Shuffleout,"
IEEE Trans. on Communications, vol. 42, pp. 2881-2889, Oct. 1994.

[BaVa96]
F. Bauer and A. Varma,
"ARIES: A Rearrangeable Inexpensive Edge-Based On-Line Steiner Algorithm,"
Proc. IEEE INFOCOM '96, vol. 1, pp. 361-368, Mar. 1996.

[FeRa94]
G. L. Feng and T. R. N. Rao,
"A Simple Approach for Construction of Algebraic-Geometric Codes from
Affine Plane Curves"
IEEE Trans. On Information Theory, vol 40, pp.1003-1012, Apr. 1994.

[FeRa95]
G. L. Feng and T. R. N. Rao,
"Improved Geometric Goppa Code - Part I: Basic theory,"
IEEE Trans. On Information Theory, vol 41, pp.1678-1693, June 1995.

[LiSz96]
Y. Li, H. H. Szu et al.,
"Wavelet Processing and Optics,"
Proceedings of the IEEE, vol. 84, pp. 720-732, May 1996.

[Peas77]
M. C. Pease, III,
"The Indirect Binary n-Cube Microprocessor Array,"
IEEE Trans. on Computers, vol. C-26, pp. 250-265, May 1977.

[SiGa94]
J. M. Simmons and R. G. Gallager,
"Design of Error Detection Scheme for Class C Service in ATM,"
IEEE/ACM  Trans. On Networking, vol. 2, pp.80-88, Jan. 1994.

[SzTG96]
H. H. Szu, B. Telfer, J. Garcia,
"Wavelet Transforms and Neural Networks for Compression and Recognition,"
Neural Networks, vol. 9, no. 4, pp. 695-708, 1996.

[Szu87]
H. H. Szu,
"Three Layers of Vector Outer Product Neural Networks,"
SPIE's Advanced Optical Technologies, vol. 634, pp. 312-330, 1987.

[TzYZ88]
N.-F. Tzeng, P. Yew and Q. Zhu,
"Realizing Fault-Tolerant Interconnection Networks via Chaining,"
IEEE Trans. on Computers, vol. 37, pp. 458-462, Apr. 1988.

---------------------------------------------------------- end of white paper