FUZZY BASED MULTI-METRIC PATH RANKING SCHEME FOR MULTIPATH ROUTING

: The paper focuses on the design of multipath ranking scheme on the basis of multi-metric approach for computer networks. The scheme is based on fuzzy logic with the underlying assumption that multiple paths are known in advance with the link utilization of each link. The proposed scheme is using two metrics: hop count and link utilization. The link utilization value is based on the load and capacity of all the underlying links. The proposed scheme finds the ranking of the paths to select a quantum of paths for distributing loads over them. The proposed scheme finds the fuzzy values of each path using fuzzy controller. The path ranking and fuzzy value of each path provides a insight into the quality of the paths.


INTRODUCTION
OSPF Protocol in the Internet architecture tends to find a single best path between a source and destination pair. Efforts in the direction of exploring multiple paths have gained substantial prominence in literature. Inclination towards fault tolerance and congestion control makes multipath routing an alluring field of research. The first major issue in multipath routing is to find multiple paths from a source to destination. After finding the multiple paths [1], the major bottleneck is to find better paths so that instead of using all possible paths, only a limited number of good path are used to keep overhead at a minimum. This paper is an attempt to find such a mechanism using fuzzy logic.In this paper, the paths have been ranked using fuzzy logic and better paths can be selected for data transmission on the basis of path ranks. The problem is to rank the input paths on the basis of link utilization value of each link in each path and the hop count of the paths.

FUZZY LOGIC
Fuzzy Logic [2] (FL) is a method of reasoning that resembles human reasoning. Fuzzy logic helps computers to take decisions related to problems having all possibilities of solutions lying in between "completely true" and "completely false" just like human reasoning. Though Fuzzy Logic has been used at various places in many real world applications successfully, yet the use of Fuzzy Logic in computer networks is still very limited. Today's computer networks are complex in nature and a fair number of uncertainties exist in the traffic conditions with unpredictable loads, available bandwidths, multiple links on a path with varying conditions of traffic on each link, etc. The reason for uncertainty is due to the fact that multiple metrics influence each other. Due to this fact these systems might behave randomly on different metrics that may be analyzed statistically but may not be predicted precisely. Thus fuzzy logic may fit in such situations. However, for a fuzzy system to work smoothly, fuzzy controllers and perceptive of fuzzy mathematics plays a significant role. It has been analyzed in literature by [3], [4], [5], [6], and [7] that the results are better in terms of link delay, link utilization, path length, packet loss, throughput, bandwidth, communication cost, etc., when fuzzy mixed metric is used. A Fuzzy Controller [8] consists of four parts (a) fuzzification module, (b) rule base, (c) inference engine and (d) defuzzification module as displayed in figure 1.

PROBLEM FORMULATION
The problem to be considered in the paper is to design fuzzy controllers for ranking the paths identified in a computer network, so that data can be transmitted faster and at the same time overhead of the network should not increase. In this research, the paths have been ranked using fuzzy logic and better paths can be selected for data transmission on the basis of path ranks. The input for this problem is the link utilization value of each link of all paths as obtained in [10]. The input paths are identified using ITMR algorithm described [10]. The problem is to rank the paths on the basis of link utilization value of each link in each path and the hop count of the paths.

FUZZY BASED MULTI-METRIC PATH RANKING SCHEME
The multiple paths are found using the ITMR algorithm described in [10] and is presented below:

Algorithm: ITMR (Intelligence Triggered Multipath Routing).
(i) Input Source, Destination, Load, and Capacity for the input network. (ix) At the destination node a fuzzy controller FITMR algorithm works to find the path ranking of all the paths which were found using ITMR algorithm and then information is sent back to source node. (x) End. Figure 2 and 3 shows the proposed Fuzzy Controllers PLU (Path Link Utilization) and PRFC (Path Ranking Fuzzy Controller) that are used to rank the paths.  µ link utilization ={ µ low, µ medium, µ high } µ pathlink utilization ={ µ VL, µ L, µ M, µ H, µ VH } On similar lines input and output fuzzy sets with three parameters have been defined for PRFC µ pathlink utilization ={ µ low, µ medium, µ high } µ hopcount ={ µ low, µ medium, µ high } µ PRFC ={ µ VL, µ L, µ M, µ H, µ VH } The output membership functions for PLU and PRFC has been shown in figures 7 and 8 respectively.  Paths are decided according to the analysis of rules of rulebase and fired rules. Minimum operation is applied to find out the minimum or least value of parameters. Output response PLU value and PRFC value are obtained by applying logical product (AND) on fired rules. These fired rules are combined to make an optimal decision. Composition merges the possessions of all applicable rules and gives the best-weighted influence of fired rules as shown in table below. The rule structure formulation is based on the consideration that minimum value for PRFC in the fuzzy scheme is assumed to be the best path. c) Defuzzification: Centriod method has been used for both the fuzzy controllers for defuzzification. The lower crisp values of the paths are considered as better paths. µ CA = *x ci / This is an approximation of center of Area (CoA) defuzzifier method. X ci denotes the center points of output linguistic terms xi, µ CA is membership function of output linguistic term and µ xi is membership function value of input linguistic terms and linguistic centers of control systems, PLU are obtained as shown in table IV below.

EXPERIMENTAL SETUP AND RESULTS
The proposed algorithm is implemented for the network shown in figure 9 using MATLAB. The network considered is having six nodes. In this network, the capacity is considered to be fixed at 10 and the load is displayed for each link. The ranking of paths using path link utilization is depicted in table VII.

INTERPRETATION OF THE RESULTS
It is observed from the results that by adding hop-count as a metric for finding the paths ranking, the paths having more number of hops are ranked lower than the paths having lesser number of hops. When the two tables 7 and 8 are compared, it can be easily observed that the paths 1-2-4-3-5-6 and 1-3-4-2-5-6 were having better ranks when only link utilization was taken into account. As soon as hop-count was added, the ranking of these paths get lowered. However, the ranking of the paths having similar number of hops remains unaffected.

CONCLUSION
A fuzzy controller PRFC has been proposed in this paper. The controller uses two metrics link utilization and hopcount for calculating the ranking of the paths. The multiple paths are identified using ITMR and their rankings are found initially by using PLU and then PRFC. It may be concluded that when both link utilization and hop-count were considered, the paths having lesser link utilization and less number of hops were found to be having lesser fuzzy values and better path ranking. These results were different from the PLU based ranking only when the hop-count was different i.e., the lesser hop-count yields a better ranking.