A Quality of Service Routing Scheme for Packet Switched Networks based on Ant Colony Behavior

Liliana Carrillo, J.L Marzo, Pere Vilà, Lluís Fàbrega, Carles Guadall Institut d’Informàtica i Aplicacions (IIiA), Universitat de Girona (UdG) Campus Montilivi, Av. Lluís Santalò s/n, Edifici P-IV, 17071 Girona, Spain {lilianac, marzo, perev, fabrega, cguadall }@eia.udg.es http://bcds.udg.es

AntNet Routing which determines routes in the network inspired by the trail-following behavior of real ant colonies. AntNet has been adopted as a basis for this proposal because it offers good performance compared with other routing algorithms [2, 3] and has positive scalability [4] and essential part of any specific QoSR proposal [1]. In this paper, the proposed work focuses upon the provision of an innovative scheme to provide QoS routing, where the use of multiple metrics are used to better characterize a network and to support a wide range of QoS requirements. The basic theory associated with Ant-based algorithms was summarized in the original thesis proposal [5] and is considered in detail in [2, 3, 6, 7]. We present some details in the subsequent section and comment on some recent work related to the application of Ant Colony Optimization (ACO) to QoS. Next, the AntNet-QoS algorithm is presented. Finally, we present conclusions, future work.

Abstract. Quality of Service (QoS) guarantees must be supported in a network that intends to carry real-time multimedia traffic effectively. A key problem in providing QoS guarantees is routing which consists of finding a path in a network that satisfies several constraints such as bandwidth, delay, delay jitter, etc. This paper proposes a novel QoS routing scheme called AntNet-QoS which provides QoS by adaptively learning routing tables in packet switched networks. AntNet-QoS is based on the socalled AntNet routing, a swarm intelligence router developed by G. Di Caro and M. Dorigo, that only considers delay when calculating the probabilities used in its routing tables. The innovative AntNet-QoS proposal is based on the utilization of different types of ants (routing packets) for the different classes of services (e.g. premium, best effort and so on) with different QoS requirements. AntNet-QoS considers parameters such as guaranteed bandwidth and maximum delay, enabling the distribution of different packet-based services using a model derived from standard AntNet.

2. ROUTING BASED ON ANT COLONY BEHAVIOR Well known Swarm Intelligence (SI) has been the basis of different approaches to solving distributed system optimization problems. In the area of network routing, algorithms have taken inspiration from the notion of stigmergy that describes the indirect communication taking place among individuals through modifications induced in their environment. Most of the algorithms, in this area, have taken their inspiration from natural ant colonies because ants are able to find shortest paths using as only the pheromone trail deposited by other ants. In a packet network context, such pheromones are simply probabilities represented at each node by a routing table and are derived from the number of visitations associated with route-finding packets or ants. Our proposal is based on the so-called AntNet routing algorithm, an ant-based algorithm developed by G. Di Caro and M. Dorigo, where mobile agents [2, 3] search for short routes in packet switching networks. In AntNet, the information processed by ants is represented in every node by a routing table and an

Keywords. Quality of Service, Routing, Ant Colony Optimization, Ant-based algorithms, DiffServ. 1. INTRODUCTION To carry real-time multimedia traffic effectively, a network must provide quality of service (QoS) guarantees. QoS guarantees are based on such parameters such as loss rate, throughput, security, and delay. Furthermore, the support of QoS services is underpinned by QoS routing (QoSR). QoS-based routing is defined as “a routing mechanism under which paths for flows are determined based on some knowledge of resource availability in the network as well as the QoS requirement of flows” [1]. Thus, the task of QoSR is to find a path in the network subject to constraints on metrics such as bandwidth, delay, delay jitter and cost. Some recent QoS routing algorithms consider only a single metric, such as hop-count, delay, cost, etc. to characterize routes in a network. One such scheme is

SPECTS '04

637

ISBN: 1-56555-284-9

routing table is updated by incrementing the goodness Pj5, j=k+1, of the last node k+1 the ant B5ĺ1 came from, for the case where node i=k+1,…,5 is a destination node, and by decrementing the values of P for the other neighbors (not shown here). The increment is a function of the trip time experienced by the forward ant going from node k to destination node i. As for M, the routing table is always updated for the case where node 5 is a destination, while the other nodes i’=k+1,…,4 on the sub-paths are taken into consideration as destination nodes only if the trip time associated with the corresponding sub-path of the forward ant is statistically good.

additional data structure that contains statistics about local traffic – delay statistics (see Figure 1).

Forward Ant (1 o 5)

Figure 1. Representation of the information in every node by means of the traditional AntNet routing.

1

Figure 1 shows the node structures used by mobile agents in AntNet for the case of a node with L neighbors and a network with N nodes. The routing table is organized as in vector-distance algorithms, but the entries are probabilistic values. The Local Traffic Statistics (LTS) function as an adaptive model for the traffic toward each possible destination. In AntNet, two types of ants (routing packets) are used, forward ants and backward ants. Forward ants are explorative ants that have the goal of discovering new routes and constantly evaluate the state of existing routes, analyzing the delay. When a forward ant reach its goal, another ant is activated, a backward ant, which returns to its source node using the reverse path taken by the forward ant and updates the LTS structure and the routing table at each node along the return path. In the routing table, every value of Pij represents the probability or appropriateness of choosing j as next node when the destination node is i. These values are calculated according to the information (delay) gathered by forward ants and modified by taking into account historical aspects or statistics of local traffic. The concept is more clearly illustrated with an example (see Figure 2). The forward ant that finds the path from source node s (s=1) to destination node d (d=5), works as follows: The forward ant, F1ĺ5, moves along the path 1ĺ2ĺ3ĺ4ĺ5 saving onto a memory stack Ssĺd(k) the identifier of each visited node k and the time elapsed between its departure from the start node and its arrival at this k-th node, that following on it is inserted in a dictionary structure D1ĺ5 carried by the forward ant. Once the forward ant arrives at node 5, it launches the backward ant B5ĺ1 that will travel in the opposite direction. In each node k, k=4,…,1, the backward ant will use the stack contents S1ĺ5(k) to update the values for Mk(µ5,ı52,W5) – LTS structure, and, in case of good sub-paths, to update also the values for Mk(µi,ıi2,Wi), i=k+1,…,4. At the same time, the

SPECTS '04

2

3

4

5

(1 m 5) Backward

Figure 2. Example of AntNet behavior The above methodology has been the basis for different kinds of routing strategies. There have been some successful applications of ant behavior to network control, the most prominent being AntNet [2, 3], and Ant-based Control (ABC) [8]. However, there also exist some attempts to apply ant behavior to QoS. In [9], Zhang Subing and Liu Zemin present a formulation of the application of ant algorithm to solve the QoS routing problem with multiple metrics, such as bandwidth, loss rate, delay and delay jitter. In [10], Chao-Hsien Chu and others present a heuristic ant algorithm for solving QoS multicast routing problem. Their algorithm considers multiple QoS metrics, such as bandwidth, delay, delay jitter, and packet loss rate, to find the multicast tree that minimizes the total cost. In [11], Kazumasa Oida and Masatoshi Sekido describe an agent-based routing system for a datagram network, called ARS, which supports QoS routing, resource reservation and admission control functions by using antlike agents. Their algorithm supports two QoS constraints: bandwidth and hop-count. These algorithms have innumerable advantages, however lack a number of important features such as scalability, routing oscillations, routing loops and full utilization of the network capacity. Then, here we attempt to profit from their advantages and avoid their weaknesses and so propose in order to suggest a potentially move powerful QoS algorithm called AntNet-QoS. 3. A NOVEL ANTNET-QOS APPROACH: ANT COLONY BEHAVIOR TO PROVIDE QUALITY OF SERVICE AntNet-QoS is a quality of services routing algorithm based on ant colony behavior that enables the distribution of

638

ISBN: 1-56555-284-9

different packet-based services using a model derived from traditional AntNet routing. Such services are likely to follow the differentiated service (DiffServ) model [12], and the scheme can, in summary, operate as follows: 1. Each flow arrives at a node with certain parameters of requested quality of service, defined by its differentiated services codepoint (DSCP) field [12]. These parameters take into account both delay values bandwidth, which must be guaranteed. 2. To find an achievable and low-cost route and to determine the network load, L-class ants which represent L-traffic classes or forwarding classes according to type of services (ToS) [12] (ToS such as premium, best effort, etc.), and it is assumed that the properties of pheromones deposited by the L-class ants are independent from each other. 3. Two types of ants are used in AntNet-QoS, in a manner similar to the traditional AntNet (forward ants and backward ants). ƒ Forward ants are exploration ants that have the goal of discovering new routes and constantly evaluate the state of existing routes, analyzing the delay and the available or residual bandwidth. ƒ At each node along the return path, backward ants update two structures in every node of the QoS data model and the routing table. 4. The probabilities in the routing tables represent the probability of that forwarding class of data achieving the required QoS.







In this way, AntNet-QoS modifies the AntNet node model (shown in figure 1), as illustrated in Figure 3.







Figure 3. Node model associated with AntNet-QoS The innovative AntNet-QoS scheme has the following key aspects:  To find an achievable and low-cost route and to investigate the network load, we use L-class ants to represent L-traffic classes (premium, best effort, etc.), and assume that the properties of pheromones deposited by the L-class ants are independent from each other. An

SPECTS '04



639

ant selects a route solely based on the pheromone deposited by the same type of ant on the path. For each L-class ant, an independent forward ant, Fsod that experiments the network load as L-traffic class, is activated at regular time intervals 't from each node s towards a destination node d. Each forward ant shares the same queues with data packets appropriate to QoS requirements in order to experience queue delay appropriate to its L-class ant. The destination nodes are locally selected according to the probability Pd, which is related to the traffic patterns generated by local load related to each L-class ant. Pd represents the probability of activating an ant from the node s to the node d, and therefore it adapts the exploratory activity of forward ants according to the current traffic distribution for each traffic class. The bandwidth available and the time passed since the ant was activated until its arrival to the k-th node is introduced into a memory stack Ssod(k). The forward ant Fsod transports a structure Dsod that contains information about all visited nodes in their memory stack. Every forward ant Fsod that is in a transit node k, selects the next hop node n among all neighboring nodes, omitting those visited previously by the same ant. If all neighboring nodes have been visited, one is selected at random. The neighbor node n is selected according to probability Pn, which is related to the probabilities in the routing table of each traffic class (one table for best effort traffic, another for premium traffic and so on) and the current link state. As in traditional AntNet, if a cycle is detected, the nodes in this cycle are removed from the memory stack. If the cycle is longer than the lifetime of the ant, the ant is destroyed to disable ants that are carrying old and invalid values about the network state. When a forward ant reaches the destination node d, another ant lives because of its death (the forward ant is destroyed). The new ant, named backward ant Bdos, receives from the forward ant all its memory (information about times and bandwidth), and carries now the information contained in the structure Dsod previously converged by the forward ant Fsod. The backward ant Bdos, one independent for each Ltraffic class, returns to node s using the reserve path taken by the forward ant Fsod. At every node k, along the path, the backward ant uses the information about the last position of its memory stack, to determine the next hop node. Backward ants do not need to share the same queues as the data packets because their task is to propagate the information collected in the memory stack to the routing tables as quickly as possible. Such ants are given the highest possible priority on their return journey. When each backward ant arrives at a node k from a neighboring node f, the backward ant Bdos updates two main data structures of the node (see figure 3):

ISBN: 1-56555-284-9

1.

2.

The Local QoS Statistics that contains two structures with information about the cost of the route from node k to each node i of the network. a. The first structure contains the delay averages and their respective degree of trust (e.g. trust intervals). Using these values, the probability PT of obtaining the requested delay conditions for the different service classes is calculated (each backward ant can reflect its network status experience only about its own class). b. The second structure contains the minimum values of remaining detected bandwidth for each traffic-class (hop by hop), its respective degree of trust and a threshold value for the degree of trust which is used to decide if a better remaining minimum exists. With these values, the probability PBw of obtaining the requested bandwidth conditions for the different service classes is calculated. A routing table similar to the used in the vector– distance algorithm is constructed. In this table, a probability value PQoS-ij is associated with the selection of node j as the next hop, for the requested QoS, when the destination node is i. The probability weights are subject to the following constraint:

¦P

QoS _ ij

The idea of Ant Algorithm to solve the QoS routing problem with multiple metrics, such a bandwidth, delay, etc. is presented. The aim of AntNet-QoS is to support a distribution of types of services described by DiffServ using the QoS parameters carried by the AntNet-QoS ants. In defining the new AntNet-QoS routing mechanism, the proposals from B. Barán and S. Sosa to improve the behavior of traditional AntNet [13] will be considered along with alternatives proposals for AntNet working with QoS [9-11]. As a simulation model, we plan to develop a realistic simulation in order to evaluate our QoS scheme performance. We are interested in measures of performance such as scalability, delay, overhead, resource utilization and so on. The knowledge about the characteristics of the incoming streams is fundamental in networks able to provide QoS. AntNet-QoS routing could act in different ways depending on the traffic class. We plan to consider an AntNet-QoS scenario with four classes of ants related to four traffic classes. An Expedited Forwarding (EF) class is defined to transport real-time traffic, two Assured Forwarding (AF1 and AF2) classes are used by traffic with two different flavors for losses and, as usual, a Best Effort (BE) class for traffic with no QoS requirements (see table 1 for more details).

1

jN k

Table 1. Details about the AntNet-QoS scheme working with four traffic classes. Traffic Class (DS field) Services Ants EF class – requires low loss, low latency, low jitter and Premium EF class assured bandwidth AF1 class – high level of forwarding assurance, with 3 Assured AF1 class sub-classes of drop Gold precedence: AF11, AF12, AF13 AF2 class – lower level of forwarding assurance, with 3 Assured AF2 class sub-classes of drop Silver precedence: AF21, AF22, AF23 BE class – the simplest traffic is best effort, with no QoS Best Effort BE class requirements

where i  [1,N] and Nk = { neighbor node k} Probability PQoS-ij is a function that depends on the probabilities PT-ij and PBw-ij ( PQoS-ij = F(PT-ij, PBw-ij) ). The probability PQoS is incremented by a value proportional to received reinforcement r (positive or negative according to accomplishment of certain conditions of reinforcement and inspired by the original reinforcement used in AntNet) and to the previous value of the probability in its corresponding routing table. The routing table is changed by either increasing the probability PQoS_df associated with the node f when destination is node d or decreasing probability PQoS_dj associated with the other nodes j in the neighborhood for the same destination d. 4. CONCLUSIONS AND FUTURE WORK QoS-based routing has been recognized as a key element in the evolution of QoS-based service offerings in the Internet. This contains a proposal for research examining QoS-based routing in the Internet based on Swarm Intelligence. This framework is based on extending the traditional AntNet routing model for supporting QoS. Most recent QoS routing algorithms consider only a single metric, such as hop-count, delay, cost, etc. to characterize routes in a network, but the use of multiple metrics is needed to better characterize a network and to support a wide range of QoS requirements. This is the prime motivation in this work.

SPECTS '04

Furthermore, we plan to make simulations considering two network topologies, the NSFnet (14 nodes, 21 bidirectional links) and NTTnet (57 nodes, 162 bi-directional links) to test the performance of AntNet-QoS. In order to compare the performance of AntNet-QoS with the Ant algorithm described in [9], we plan to make simulations using the same environment and network topology. We will also consider overhead and scalability problems related to AntNet-QoS scheme. Currently, we are improving the system and implementing the AntNet-QoS algorithm.

640

ISBN: 1-56555-284-9

13. B. Barán and R. Sosa. “AntNet: Routing Algorithm for Data Networks based on Mobile Agents”. Argentine Symposium on Artificial Intelligence ASAI-2000, Tandil - Argentina.

5. ACKNOWLEDGEMENTS The authors would like to thank the Spanish Education Ministry for their partial financial support by means of the grant MCyT [TIC2003-05567]. Liliana Carrillo would like to thank the ‘Universitat de Girona’ for the financial support (UdG grant) awarded to assist in the completion of her Ph.D. dissertation. References 1.

Crawley, R. Nair, B. Rajagopalan and H. Sandick. “A Framework for QoS-based Routing”. RFC 2386. August 1998. 2. G. Di Caro and M. Dorigo. “AntNet: a mobile agents approach to adaptive routing”. Tech. Rep. IRIDIA/9712, Université Libre de Bruxelles, Belgium. 3. G. Di Caro and M. Dorigo. “AntNet: Distributed Stigmergetic Control for Communications Networks”. Journal of Artificial Intelligence Research 9. Pp. 317365. 1998. 4. L. Carrillo, J.L. Marzo, D. Harle and P. Vilà. “A Review of Scalability and its Application in the Evaluation of the Scalability Measure of AntNet Routing”. In proceedings of the IASTED-CSN 2003 Conference. September, 2003. 5. L. Carrillo and J.L. Marzo. “Quality of Service Routing based on Ant Colony Behaviour”. Research Report IIiA 03-13-RR. Universitat de Girona. July, 2003. 6. M. Dorigo, V. Maniezzo, and A. Colorni. “The Ant Systems: An Auto catalytic Optimizing Process”. Technical Report 91-116. Polytechnic of Milan. 1996. 7. M. Dorigo and L.M. Gambardella. “Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem”. IEEE Transactions on Evolutionary Computation. April 1997. Vol. 1, No.1, pp. 53-66. 8. R. Schoonderwoerd, O. Holland, J. Bruten and L. Rothkrantz. “Ant-based Load Balancing in Telecommunication Networks”. Adaptive Behavior, vol. 5, pp. 169-207, 1996. 9. Z. Subing and L. Zemin. “A QoS routing algorithm based on ant algorithm”. ICC 2001 - IEEE International Conference on Communications, Vol. 5, No. 1, June 2001, Pp. 1587-1591. 10. C-H. Chu, J-H. Gu, X.D. Hou and Q. Gu. “A heuristic ant algorithm for solving QoS multicast routing problem”. Evolutionary Computation, 2002. CEC'02. Vol. 2, 12-17 May 2002. Pp: 1630 -1635. 11. K. Oida and M. Sekido. “An agent-based routing system for QoS guaran-tees”. Proceedings of the IEEE International Conference on Systems, Man, and Cybernet-ics. Oct. 1999. Vol. 3, 12-15, pp. 833 -838. 12. M. Carlson, W. Weiss, S. Blake, Z. Wang, D. Black, and E. Davies. “An Architecture for Differentiated Services.” RFC 2475, December 1998.

SPECTS '04

641

ISBN: 1-56555-284-9

A Quality of Service Routing Scheme for Packet ...

Abstract. Quality of Service (QoS) guarantees must be supported in a network that intends to carry real-time multimedia traffic effectively. A key problem in providing. QoS guarantees is routing which consists of finding a path in a network that satisfies several constraints such as bandwidth, delay, delay jitter, etc. This paper ...

408KB Sizes 0 Downloads 97 Views

Recommend Documents

Quality-of-Service Routing Using Maximally Disjoint Paths
destination so that any link with low resource availability is highly unlikely to belong .... node of the current domain or peer group). Crankback in. Phase I is ..... Name. Number Diameter Core. C2C PROB MAX of Links (hops). Nodes. DEGREE.

An Overview of Quality-of-Service Routing for the Next Generation ...
In the current Internet, data packets of a session may follow different paths ... upper bound. A feasible path (tree) is one that has sufficient residual. (unused) resources to satisfy the QoS constraints of a connection. The basic function of QoS ro

A Fast and High Quality Multilevel Scheme for ...
Mar 27, 1998 - University of Minnesota, Department of Computer Science ... The multiple minimum degree ordering used almost exclusively in serial direct.

Adaptive Quality of Service for a Mobile Ad Hoc Network
Adaptive Quality of Service for a Mobile Ad Hoc Network. Antonis Dimakis ... routing system that can provide different classes of service in ... typical applications.

Improved Secure Routing Scheme in WSN - International Journal of ...
evaluate the performance of Ad hoc On Demand Distance Vector (AODV) routing protocol for monitoring of critical conditions with the help of important metrics like delay, throughput and network load with different techniques in different scenarios for

improving the quality of service (qos) in wsn routing ...
In this module Cluster-based WSNs allows for the great savings of energy .... Conference on Information Processing in Sensor Networks (IPSN'04), Apr. 2004.

Quality of Service Architectures for Wireless Networks ...
School of Electrical Engineering and Computer Science. Washington State ..... less networks currently being low bandwidth systems, this is a point of concern.

Handover based Backlogged Packet Discard Scheme ...
1. the statistics of the call arrival process ... A BS (base station) is assumed to be located at the center of ... From busy cycle analysis for an M/M/1/1 queue.