Sie sind auf Seite 1von 5

2009 World Congress on Computer Science and Information Engineering

A Three-Layer Network Management System


Jiang Guo1, Heru Xue2, Raj Pamula1
Department of Computer Science, California State University Los Angeles, USA
jguo@calstatela.edu
2
School of Computer Sci. and Information Engi., Inner Mongolia Agri. University, P.R. China
xuehr@imau.edu.cn
1

network will improve the performance of the entire


systems.
SOA Applications: These applications use
service-oriented architecture. The communications
between software components rely on the
performance of the internet. Poor network
connection will make these systems fail. So, high
QoS of the network is the one of the keys of SOA
applications.
VOIP and Large Data Applications: Yahoo IM,
MSN IM, Skype IM all support voice and video
transmitting in real time. Some website, such as
MSN, LA Times also support streaming news.
These applications need high performance global
networks. Another kind of application is download
movies from commercial companies, such as Time
Warner Cable. They provide movie on-demand
service. This also need high speed internet. All
these applications generate large UDP data
transmission.
Current network management systems do not have
enough power and flexibility for the network
management. The major problem is that the control
information (use to control network) and data
information (the information needs to be transmitted)
are not distinguished. All are transmitted in the same
network and same level of the architecture.
Therefore, these systems cant choose right path
before the message is caught in a "traffic jam". This
is because they use flat architecture to deliver timecritical network control information. And the flat
architecture will cant support real-time path reconfiguration due to no global information available
for the dynamically changing networks.
Some researchers propose to use reservation
protocol (RSVP) [1] to increase the QoS for some
time critical applications. The reservation protocols
are designed on top of the network. When a package
needs to be transmitted from node to node, the
reservation protocol can reserve a path through the
network by setting up routing tables to guarantee the
requested QoS. However, if somewhere is traffic
jammed in the network, the reserved path also will be
blocked.

Abstract
As VOIP, stream video, broad casting, View-onDemand, online shopping and banking are more and
more widely used; global network traffic is growing
more and more. It is difficult to keep pace with
increasing demands by continually adding capacity of
network and global network performance is not
satisfied. This paper discusses the design of
J3Manager our three-layer Network Management
System, which is a QoS mechanism that presents a
novel approach for using aircrafts and satellite in
network management systems.

1. Introduction
As the internet becomes more and more popular,
online banking, online shopping, and online gaming
go into most peoples life. So, how to improve the
network performance and reduce the computer
network traffic is a big challenge that we have to
face. Although some local optimization approaches
can be used, the global network capability still has
not maximized.
We propose a new three-layer network QoS
mechanism to help network administrators to collect
global network information so that they can
maximize the utilization of global network resources.
Our mechanism will improve the quality of service to
mission critical applications and users, and also
reduce the costs of providing these services based on
our concept of information classification.
Below describes the benefits for three application
areas that can be expected from our approach:
High QoS Required Applications: These
applications often need to provide high
performance services across internet. For example,
GET system at CSULA is a PeopleSoft based
system. These applications usually need quick
response to avoid session timeouts. So, the
network performance will affect the transaction
successful rate directly. Therefore, high QoS of the

978-0-7695-3507-4/08 $25.00 2008 IEEE


DOI 10.1109/CSIE.2009.444

230

Our architecture basically uses three different


layers networks to transmit different information. So,
when a problem found in lower layer, the upper layer
network management can re-establish a new path for

lower layer. In this way, RSVP problems can be


avoided and QoS can be guaranteed.

Servers Running
on Satellites
GPS and Network
Information

Servers Running on
Aircraft Network Nodes

Server

Global Network
Information
Agents Running on
Wired Network Nodes
(Information Collection)

Local Network
Information

Agent

Server

Server

Server

Agent

Figure 1. A Three-Layer Network


All agents at lowest layer communicate with a
parent server that maintains performance information
for paths crossing multiple regions [5]. The routers
and servers use real-time protocols that guarantee low
transfer delays for network control information and
agent communications. Most management and
control tasks will be performed by the servers in an
automated and timely fashion.
The server running at top layer (satellites) is used
to help create aircraft servers network. Servers
running on aircrafts are used to manage network
resources in regions based on the network
information collected from the agents. The servers at
this layer are also connected and form an aircraft
network. In this way, all the aircraft servers can
exchange network control information across regions,
and all the servers will have global network
information. Based on the global network
information, a global optimized routing path can be
provided to satisfy high QoS requirements.
J3Manager agents dynamically, automatically and
efficiently collect network control information and
transmit it to regional aircraft servers. The servers
will exchange the regional information with each
other effectively and in a timely manner, then they
will use the updated information to detect, diagnose
and repair any problems that are found in the
networks.
In our approach, a software agent is running on a
router, which is configured to collect network control
information. The agent collects relevant information
from a variety of network sources. It is responsible
for collecting performance statistics such as the
delays of packets passing through the link and the
loss rate of each service. A regional server running
on an aircraft receives the network control

2. Our Three-Layer System Architecture


Our three-layer system architecture for the internet
is similar to road traffic monitoring and control
during commute hours in a large city such as Los
Angeles, where TV and radio stations deliver the
traffic information timely. In the same way, we use
aircrafts and satellites to deliver the network control
information. The advantages of this approach are: (1)
Long path can be monitored. (2) Global information
of the network can be collected and analyzed. (3)
Early detection and prediction can be made. Some
researchers propose to build airborne network [2] [3].
They focus on providing broadband service for
passengers [4] while our approach focuses on
network management.
Our architecture achieves timely network
management by employing a set of management
servers ("aircraft" and satellite) that we will refer as
J3Manager servers and associated agents to
periodically collect control information about the
network. J3Manager servers are physical servers
running on aircrafts and satellites (Figure 1).
In our architecture, a server will oversee the traffic
of a region and communicate with other servers
through server network and satellites. These servers
manage the network traffic information and route
tables. The servers will be organized in a hierarchy.
We use a two-level server hierarchy. The lower layer
servers reside on aircrafts and upper layer servers
reside on satellite as see in Figure 1. A region is
represented by a circle. At the lowest layer, each
agent will collect and maintain up-to-date path
information for a region. All the nodes at this layer
are connected by the wired network (optical fiber
network).
231

address allocation approach. In this approach, each


region has a backbone aircraft. It is in charge of the
registration and communication of the aircrafts in that
area. If an aircraft enters a region and cant find a
backbone aircraft to register with, then it can
communicate with upper level satellites to find
other backbone aircrafts nearby and in the mean time
promote itself as backbone aircraft and collect related
network information. In this way, our approach does
not have a lot of broadcasts through the whole
aircraft network and therefore reduces network
overhead and improves performance.
When an aircraft enters a region, it always can to
register with a backbone aircraft in that area specified
by the GPS. Then, it can get the network information
about the gateway from this backbone aircraft. Then,
it can get the network information and specify
possible routes to other aircrafts.

information from different agents running on


different routers. A regional server maintains a
regional path database and can provide feedback
optimized network information to the agents in its
region.
The agents running at lowest layer are developed
and maintained separately. They are global network
information sources for network management. The
servers periodically update the path database using
the data collected by individual agents running on the
routers, and take control actions when it discovers
any significant QoS degradation for a path within its
region. Servers in charge of different regions are
required to communicate with each other so that they
can maintain and update network performance
information for paths crossing multiple regions,
which is what we called global optimization.

3. Aircraft-based Server Network

4. Logical Servers and Their Cooperation

At our aircraft server level, since one antenna can


easily cover an area with radius of 600 miles, thus the
whole sky above US can be divided into several
overlapped areas [2] [3] [4]. So, we can use a new
group formation approach based on network resource
discovery and geographic network address allocation
to establish the communications between the
aircrafts.
The biggest challenge for the construction of the
aircraft network is that senders need to know
who/where receivers are and aircrafts can join or
leave the network dynamically. So, we propose to use
the approach to combine the broadcast [6] approach
with core approach [7].
A broadcast approach means that if an aircraft
pass a region, it has to broadcast network control
information throughout the network. Then, other
aircraft will be aware of its existence. These
broadcasts can be used to capture network dynamics.
We can use this approach establish the
communication between the aircrafts in one region
and in other regions. To reduce broadcast overhead,
we use a core register approach. In this approach, if
an aircraft flies into a region, it has to register itself at
the backbone aircraft (core node). In this way, the
newly registered aircraft can send and receive data
information to/from other aircrafts in this region.
By using this approach, we can avoid the whole
network crashing because of one single backbone
aircrafts problem, because we can use multiple
backbone aircrafts and a dynamic backbone aircraft
promotion approach that is based on the idea of
domain controller and backup domain controller in
domain administration. Since each aircraft has GPS
ability, aircraft can easily identify the backbone
aircrafts address based on our geographical network

Based on the regional information provided by


GPS and aircrafts mobility, an individual aircraft can
choose several relatively stable paths to communicate
with other aircrafts. Here, we can store the route
information on the backbone aircraft.
As each aircraft has its own flight path, we can
estimate its mobility and obtain region network
information before it enters a region. This approach
will greatly reduce the effect of absolute mobility and
increase the network regions relative stability.
Another challenge for the aircraft network is the
network address allocation [8]. This is essential for us
to find the aircraft group address and network routing
information. Since we can use satellites as a
centralized mechanism to manage this kind of
information for our aircraft network, thus we can
utilize the hierarchy of the Internet to manage the
network routes and map region prefixes into domains
through the network address allocation [9] based on
network routing hierarchy in our aircraft network.
We also can use a geographic location based
approach to allocate the network address for aircraft
network. In this approach, we can divide the network
address space into different prefixes with each
geographic region having an address prefix. All the
aircraft in this region have local aircraft information
and all the local aircrafts have the same address
prefix. Since all the aircrafts have GPS ability, it is
easy for each aircraft to obtain its geographic
location, and based on this information, easy for each
aircraft to obtain its network address. Once this is
achieved, it can communicate with other aircrafts.
In addition to sending queries to the agents in its
region, a region server also needs the capability to
send back global network information to other
232

from upper level satellite GPS and network


information makes the system flexible enough to
scale to large number of information sources.

servers in response agents queries. So, J3Manager


needs to deal with transmitting queries, returning the
appropriate information, and building the appropriate
internal structures. For example, the Server Jupiter
needs to combine network data from the Server
Uranus, Saturn and Neptune. The information that
has been obtained, in turn, is from the servers of
different regions. The expression can be:

5. The Design of Logical Server


After the network information has been collected,
some actions can be taken, such as re-configuring the
physical network that is adding physical routers
and/or servers, or redirecting the dataflow traffic. Of
course, administrators are always able to use the raw
data collected at the lower level of agents and higher
level servers. However, network management tools
are definitely very helpful. Additionally, we need to
analyze the raw data to determine when we need to
re-configure the router table or re-optimize the
network connection.
Each server running on an aircraft is specialized to
a single region and provides access to the available
information sources within that region. The region
model is intended to be a description of the network
region from the point of view of users or other
aircraft servers that may need to obtain information
about other regions.
The region model of a server defines its area for
communicating with it. That is, it provides links to
the agents in its region. This model consists of
descriptions of the information in the region and
locations of these agents. These agents and their
locations are mapping directly to the objects
described in its servers information database. This
model provides a semantic description of the region,
which is used extensively for processing queries.
Our server organization is static and preconfigured in the sky according to the region
division. In this way, the routing tables and paths
between the aircraft servers are relative stable and the
system will be much simpler and more efficient. We
use a 1:1 mapping between the physical servers and
the logical servers. However, physical servers are
dynamic and depend on the aircraft position in the
sky identified by GPS. Since the logical servers are
relative stable mapped to the regions in the sky, we
can pre-configure the servers and do not need to reconfigure the router regions often.
To make the system more efficient, we need to
determine the optimal content of regions and provide
mechanisms and policies for changing the physical
organization because the mapping of logical routers
to regions is relative stable and physical servers must
change in response to their mobility and position
changing. Built into our system architecture are
regions that are monitored by logical servers, whose
union covers the entire network space.
The network information about each of the
regions is equally replicated on the physical servers

<retrieve
<server Uranus>
<network>
<bandwidth> ? </bandwidth>
<latency> ? </latency>
<transmission time> ? </transmission time>
<throughput> ? </throughput>
</network>
</server>
<server Saturn>
<network>
<bandwidth> ? </bandwidth>
<latency> ? </latency>
<transmission time> ? </transmission time>
<throughput> ? </throughput>
</network>
</server>
<server Neptune>
<network>
<bandwidth> ? </bandwidth>
<latency> ? </latency>
<transmission time> ? </transmission time>
<throughput> ? </throughput>
</network>
</server>
</retrieve>
A region server also needs to response other
region servers queries, which means we need to
transmit the messages among the region servers,
which provide information for each individual region.
So, we need to exchange information and based on a
specific message format and a message-handling
protocol to support run-time knowledge sharing
among agents. We use XML and SOAP as a
language and protocol for one server to interact with
another server or for two or more servers to share
network information in support of cooperative
problem solving. There are several issues related to
the autonomy of the aircraft servers. First, each
server can acquire information from other severs as
needed to accomplish the tasks that it is designed for.
Second, we may need to create new servers if we
cannot find an existing one that cover a specific
region or play as a bridge between different region.
In summary, each individual aircraft server should be
able independently built and maintained with the help

233

charge that region will start to run. (3) the physical


servers are dynamic and mobile. They fly in the sky
and communicate with satellite to get their position
and network control information. (4) the multi logical
servers on the aircrafts form a wireless based network
management system.
The J3Manager system is an on-going research
project for the development and refinement of its
agent-based distributed network management tools.

requires knowledge of the region to deal with its


dynamic and mobility within a region. After dividing
the sky into regions, logical servers are mapped to
physical servers when aircrafts enter their pre-defined
region. Logical servers will monitor a sub-net's
performance based on the information provided by
the agents running on the ground bottom level.
Mapping a single logical server that spans multiple
physical servers may be made in order to provide a
particular logical server with additional resources, but
this may increase extra communication overhead.
As the J3Manager aircraft server manages a static
network on the ground, the number of all possible
paths under its management can be pre-simulated. So,
we only need to update network information for all
the paths when we need to re-schedule the network
traffic in case of traffic jam happens and reported by
the agents in that region.

References
[1] L. Zhang, S. Deering, D. Estrin, S. Shenker, and
D. Zappala, RSVP: a new resource reservation
protocol, IEEE Network Magazine, pp. 8-18,
September 1993.
[2] R. Trafton, S. Pizzi, The Joint Airborne Network
Services
Suite,
Military
Communications
Conference, MILCOM 2006, Washington DC,
October 23-25, 2006.
[3] A. Ayyagari, J. Harrang, S. Ray, Airborne
Broadband Communication Network, US Patent
Number: 6018659, 2000.
[4] J. Lai, Broadband Wireless Communication
Systems, US Patent Number: 6285878, 2001.
[5] J. Guo, Y. Liao, B. Parviz, and R. Hawk,
Integrating Active Agent Technology with Network
Management, Proceedings of the 20th International
Conference on Computers and Their Applications,
New Orleans, Louisiana, USA, March 16-18, 2005.
[6] J. Wu and F. Dai, Broadcasting in Ad Hoc
Networks Based on Self-Pruning, IEEE INFOCOM,
2003.
[7] B. Kaliaperumal, A. Jeyakumar, Scalable
Multicasting Adaptive Core Based Approach,
Academic Open Internet Journal, Vol 16, 2005.
[8] M. Thoppian and R. Prakash, A Distributed
Protocol for Dynamic Address Assignment In Mobile
Ad Hoc Networks, IEEE Transactions On Mobile
Computing, Vol. 5, No. 1, January 2006
[9] S. Kumar, P. Radoslavov, D. Thaler, C.
Alaettinoglu, D. Estrin, and M. Handley, The
MASC/BGMP Architecture for Inter-domain
Multicast Routing, Proceedings of ACM
SIGCOMM, August, 1998.
[10] J. Bennett, and H. Zhang, Hierarchical packet
fair queuing algorithms, Proceedings of ACM
SIGCOMM '96, pages 143-156, CA, August 1996.

6. Conclusion
High quality broadband is very important for our
business, entertainment, and daily life. We designed a
three-layer network to combine the information
delivery ability of aircrafts and satellite to improve
the global network management. As using satellite to
delivery information is very expensive, so we divided
the information flowing in the global network as
control information and data information. Only
network control information are sent and received
from satellite. Network servers running on aircrafts
manage different network regions based on their
positions in the sky identified by the GPS from
satellite. This approach will avoid RSVP protocol
problem, in which network control information and
data information are mixed and share a transmission
link. This design also avoid the issues that come from
bandwidth approach, in which each control
information is assigned a weight to reflect the share
of bandwidth, and a packet schedule policy us used at
the link to ensure that the control information receive
their respective shares over time [10]. This design is
to improve the global network performance through a
three-layer architecture. The first layer is on ground
with wired connection, the second layer is on
aircrafts with wireless connection, and the third layer
is on satellite.
J3Manager is based on the client/server model,
where the servers are centralized managers running
on the aircrafts and the clients are distributed agents
running on the ground. The biggest difference
between our system and traditional models is: (1)
there are more logical servers in our system, each
server manage a specific region. (2) the logical
servers are static. When each physical server
(aircraft) enters a specific region, the logical server in
234

Das könnte Ihnen auch gefallen