General

Abstracts
Group Objectives
PhD/MPhil
Apply for a PhD

Agents

Agent-based, distributed security system
Mobile Agents in Network Applications
Agent-based Fault Diagnostics

Network routers filter the network traffic so that the only traffic that flows into and out of a network is required to go into or out of the network. In many cases there are several possible routes that can be taken between two nodes on different networks. To stop traffic taking a long route or even one that does not exist, each router must maintain a routing table so that it knows where the data must be sent when it receives data destined for a remote node [1].

For routers to find the best way they must communicate with their neighbours to find the best way through the network. This measure can be defined in a number of ways:

• Router hops. This defines the number of router hops to the remote node, where each router communicates with its neighbours and uses this information to build-up a routing table. Unfortunately, this is not a good measure of the delay of data transmission as a route with a fewest number of hops may be congested or may be connected to a slow link.

• Delay. This is an improved method of routing and involves transmitted the expected delay between the host and the router. Obviously, this is a difficult parameter to determine as traffic levels can vary over long and short intervals.

• Reliability. This technique determines the best route by analysing the reported errors from a given route. The route with the lowest error probability will then be the most reliable route. Unfortunately, this is a difficult parameter to measure as routes vary, over time, in their reliability.

• Reachability. Many applications, especially WWW-based applications, are more interested in actually reaching the remote node and than for other parameters, such as delay and reliability.

The proposed research will try and overcome these problems by integrating the most common routing methods with some from of intelligence. The three main methods for routing are:

• RIP [2]. In this method each router transmits RIP packets with their entire routing table every 30-60 seconds. This information contains the number of router hops and the number of ticks (in 1/18 second) that it will take for packet to reach a given destination. Unfortunately, this method tends to swamp networks with routing information.

• OSPF [3-5]. The OSPF is an open, non-proprietary standard which was created by the IEFF (Internet Engineering Task Force), a task force of the IAB (Internet Advisory Board). It is a link-state routing protocol and is able to maintain a complete and more current view of the total internetwork than distance-vector routing protocols (such as RIP). Link-state routing protocols have the features:

• They use link state packets (LSPs) which are special datagrams that determine the names of and the cost or distance to any neighbouring routers and associated networks.

• Any information learned about the network is then passed, using LSPs, to all known routers, and not just neighbouring routers. Thus all routers have a fuller knowledge of the entire internetwork than the view of only the immediate neighbours (as with distance vector routing).

• Additional hierarchy. OSPF allows the global network to be split into areas. Thus a router in a domain does not necessarily have to know how to reach all the networks within a domain, it simply has to send to the right area.

• Authentication of routing messages using an 8-byte password. This length is not long enough to stop unauthorised users from causing damage. Its main purpose is to reduce the traffic from misconfigured routers. Typically a misconfigured router will inform the network that it can reach all nodes with no overhead.

• Load balancing. OSPF allows multiple routes to the same place to be assigned the same cost and will cause traffic to be distributed evenly over those routes.

• EGP/BGP [3-11]. The two main interdomain routing protocols in recent history are EGP and BGP. EGP has the disadvantage that it treats the Internet as a tree-like structure. This structure, as illustrated in Figure 1(a), is normally made up of parents and children, with a single backbone. A more typical topology for the Internet is illustrated in Figure 1(b). BGP overcomes this problem, but it is more complex than EGP (but not as complex as OSPF).

• Unfeasible routes length.

• Withdrawn routes.

• Total path attribute length.

• Path attributes.

• Network layer reachability information.

A major object of the research is to integrate intelligent methods into the standard routing protocols. This will involve training a Neural Network to route packets, given a required optimisation (such as least delay or reachability). Fuzzy logic may also be applied to define the how the optimisations are implemented..

(a) (b)

Figure 1 (a) Tree-like topology and (b) a network with multiple backbones.

Figure 2 Autonomously attached networks

REFERENCES