Many applications do not fit well with the traditional best effort packet delivery policy of the Internet. These include applications such as Internet telephony and video conferencing which require voice and bulky graphical images transfer. Therefore, the policies of assigning traffic to various service classes and providing service as per the service level agreement of the user with the network provider came into existence. Multi-protocol Label Switching is the backbone of fast switching technology that helps the network service providers to implement these policies. It provides Quality of service oriented reserved paths from the source to the destination for the user’s traffic. Selection of these paths is a cumbersome task, especially when the traffic forecast is totally unknown. Furthermore, nodes and link failures in the Internet worsen the situation. This paper addresses the issue of selecting Label Switched Paths (LSPs) for various traffic demands in the network so that the resultant network has the characteristics like high failure resistance, low LSP demand blocking probability, low impact from the node or link failure, load balancing and low over-all resource utilization. By extensive simulations, the proposed cost function has been compared with the various cost functions mentioned in the literature and it was found to score over them in major aspects.
The drastic growth of Internet and the use of computer networks have encouraged service providers to offer high priority Internet applications. These applications require continuous bandwidth and high availability of the network resources. Since the resources like bandwidth are limited and it is not always feasible to enhance them, it is necessary that they are used efficiently. Multiprotocollabel switching (MPLS) was essentially proposed for fast forwarding the packets over the Internet [
Failure information signal (FIS) has to travel to the source node to initiate the switching of traffic to the recovery path which leads to the packet loss. This is because the source keeps on transmitting the packets in the mean time. Rerouting has the disadvantage of high network restoration time since the new routes are established only after the node or link fails.
Failure in the network cannot be fully avoided but it can be reduced if some consideration is paid to the failure history of the link during its selection [
The rest of the paper is organized as follows. Section 2 discusses related work on the efficient path selection. Section 3 describes the model formulation. Section 4 provides the details of various cost functions used. Section 5 describes simulation results and performance analysis. Finally, the conclusion and the scope for future work are provided in Section 6.
Pertaining to the issues discussed in Section 1, many authors have proposed various solutions for the efficient selection of the LSP in MPLS networks. This section discusses the proposals related to the work put forward in this paper.
Paper [
1) How to distribute the affected traffic to the failure free working LSPs?
Solution: The paper reflects the use of minimum cost flow solution for this problem by establishing a simple graph.
2) How to redirect the affected traffic to the failurefree working LSPs?
Solution: Changing the routing tables of the IP Access Network before MPLS networks for redirecting the traffic to new LSPs.
3) How to forward the affected traffic along the route of a failure-free working LSP?
Solution: Using IP tunneling mechanism.
4) How to solve packet loss and disorder?
Solution: Transferring the sequence number of the unsent packet to the source and thereafter all the packets starting from that number are transmitted by working LSPs.
There are certain issues that have not been addressed in this paper:
1) The paper does not mention how to select the failure free LSPs from that particular source to destination? If the backup LSP is selected simultaneously with the selection of active LSP then following problems can arise:
• Convergence will take considerable time since failure signals will have to travel to source router.
• We cannot predict whether the LSPs will be free when needed since they are allowed to carry other traffic also.
On the other hand, if the LSPs are selected in real time then the specific load balancing algorithm having the same effect as the minimum cost flow approach to transmit the failed LSPs traffic to failure free LSPs should have been mentioned. In minimum cost flow, the LSP having the minimum number of routers will be selected to transmit maximum packets. An instance described in the paper is to transmit 10 Mbps by balancing the load between LSP1 and LSP2. LSP2 is having cost 2 and residual bandwidth 8 whereas LSP3 is having cost 3 and residual bandwidth 10. The algorithm proposed to transmits 8 Mbps by LSP2 and 2 Mbps by LSP3 which does not solve the purpose since the aim is to have the packets in order. The speed with which the packet reaches will be the speed of the slower LSP having 3 as the cost. And more over it is not a good idea to use all the residual bandwidth of a LSP since it will limit its further usage when required.
2) In the permission token approach, the proposal hands over the token to the egress routers of the failure free LSPs. The router which possesses the token will forward the packets. Until then it will keep the packets in its buffer. The buffer size of the router is limited and there will be packet loss if the buffer gets overflow awaiting the permission token in the absence of flow control method.
In paper [
In the above works authors except that of paper [
Model: G = (V, E) is a directed graph representing the network in which:
V is the set of LSR (Label switched routers) and E is the set of edges Action: Determine the optimal set of binary variables a(e) and b(e) that:
Equation (1) is the objective of the model which calculates the minimum summation of the cost of the LSP calculated by the model. Binary variables a(e) and b(e) have the values 1 and 0 depending on whether the edge e is included in the LSP or not. Equations (2) and (4) are the flow conservation constraints which impose the condition that the total flow entering the node should be equal to the total flow leaving the node for every node which is not source or destination. For the source (destination) the incoming (outgoing) flow should be zero. Equation (3) applies the constraint that the sum of used bandwidth of a link and the bandwidth demand of a LSP should not be greater than the capacity of the edge. Equation (5) which is the base of this model defines the cost calculating functions. In this paper various cost functions are calculated by MinHop, load balancing, Residual Bandwidth, Link Cost, MIRA, and Proposed Algorithm.
Shortest path in the network is calculated by the famous Dijkstra’s algorithm [
In the MinHop cost function, every link is given a unit weight. Shortest path algorithm selects the path which has minimum number of links. Therefore, same links are selected every time whenever there is a demand between the set of nodes. Consequently this causes rapid congestion of the links which leads to a scenario in which a part of network is heavily loaded while the remaining part is left underutilized.
Load balancing refers to the distribution of load so that the network under consideration is uniformly loaded. In order to do this the cost of every link is given by:
where D is the various queuing and the propagation delay experienced by the packets traversing the link and U is the load on the link due to the current passing by traffic.
Cost function for every link using this technique is calculated as:
where U and B in Equation (7) are load of the present traffic and bandwidth of the link respectively.
In the Link Cost function every link is assigned a cost as shown in Equation (8):
where, PD is the delay induced while propagation of packet through link and QD is average delay of the packets while waiting in the queue
In MIRA the critical links are calculated on the basis of MaxFlow between source and destination. Cost of the link is then assigned as:
Criticality of a link in Equation (9), denoted by Cr is the numerical value incremented whenever the MaxFlow crosses a link. Thus, cost is directly proportional to the criticality of the link.
This paper implements the major cost functions and compares them with the proposed novel cost function to calculate the cost of a link based on three factors namely Link capacity, Link survival probability and Link Distance from source as:
where C, S and D in Equation (10) are capacity, survival probability and distance of link from the source respectively. Distance of link from the source is calculated by all pair shortest path algorithm. Constants α, β and γ are used for assigning relative weightage to the three metrics.
Extensive simulations are performed on the proposed model in Section 3. This paper generates the topology by using BRITE [
Network Protection Degree (NPD) of a network is computed as:
In Equation (11) P is the survival probability of link e and C is the count variable which is denotes the total number of edges in all the LSPs. A plot of number of LSPs with Network Protection Degree in
Failure Impact Degree (FID) is the impact of the failure on the network. Impact of the failure is the amount of packet loss and packet disorder due to the link failure.
Most probably the link with low survival probability will fail and as suggested in this paper, this type of link has high cost function and therefore would not be considered in minimum cost path. But other proposals do not consider this metric. FID is calculated as:
In Equation (12), B is the bandwidth of LSP l and D is the distance of low survival probability link from the source. The links with low survival probability having distance more than one hop from the source are selected. The sum of such is divided with the total number of paths. Plot the FID is illustrated in
NLP is calculated as:
In Equation (13), B in the numerator denotes the bandwidth of edge of a LSP having probability more than or equal to 0.9 whereas in the denominator, B is the total bandwidth of all the edges.
Blocked request is the number of LSP requests blocked
by the model. When plotted with total number of LSPs requested in
Average load on the network is calculated as:
In Equation (14) Lu is the link usage. As shown in
This paper presents a model for path allocation for dynamic LSP request in a MPLS network. A novel cost function with three metrics is proposed. Proposed cost function has been simulated and was found to increase the survivability of network considerably when compared with five other algorithms mentioned in the literature. For the future, the present work can be extended and models for efficient backup path can be devised and compared with other traffic protection techniques proposed in this realm.
This work has been made possible by extensive support from Jaypee University of Engineering and Technology, Guna. Authors also acknowledge the software provided by www.ampl.com and IBM for academic research.