An Enhanced Secure Heuristic-stochastic Routing Arithmetic in Mpls Network

To improve routing security in MPLS network, base on the stochastic routing algorithm, we propose a proac-tive mechanism we call enhanced secure heuristic-stochastic routing (ESHSR), which brings to bear Bayes-ian principle, explores the existence of multiple routes and forces packets to take alternate paths probabilis-tically. In this paper, we investigate game theoretic techniques to develop routing policies which make interception and eavesdropping maximally difficult. Through simulations, we validate our theoretical results and show how the resulting routing algorithms perform in terms of the security/delay/drop-rate, and we contrast them with the mechanism, secure stochastic routing (SSR). We observed that our scheme makes routing more secure than traditional secure stochastic routing, as they make use of the information of detecting the other side's behavior.


Introduction
The purpose of traffic engineering (TE) [1][2][3][4][5][6][7][8] is to improve network performance through the optimization of network resources.The emerging Multi-Protocol Label Switching (MPLS) technology has introduced an attractive solution to TE in IP networks.MPLS can efficiently support the explicit routes setup through the use of Label Switched Paths (LSPs) between the ingress Label Switched Router (LSR) and the egress LSR.Hence it is possible to balance the traffic through the network, thus improving the network utilization and minimizing the congestion.However, one of the most obvious attacks to a communication network is packet interception which prevents data originating from one (or several) nodes to reach the destination.Eavesdropping can be thought as a "passive" form of interception, in which packets are "snooped" but not removed from the network.In "traditional" shortest-path routing protocols, the path over which a data packet travels is fairly predictable and easy to determine.Even if several paths with the same number of hops exist, routing algorithms typically select one of the possible options and utilize that same path for all packets.Indeed, a study by Zhang et al. [9] reveals that Internet routes are fairly persistent (e.g., often the same route between a source-destination pair persists for days; only 10% of the routes persist for a few hours or less).This makes IP networks vulnerable to packet interception and/or eavesdropping attacks.Notable exceptions to single-path routing schemes are Equal-Cost Multi-Path (ECMP) [10] and OSPF Optimized Multi-Path (OSPF-OMP) [11].However, these algorithms were developed to increase throughput and not to make routing robust to attacks.In practice, they do not introduce unpredictability and therefore packet interception is fairly easy to achieve.
In this paper, we describe enhanced secure heuristicstochastic routing, or ESHSR, whose main goal is to make packet interception maximally difficult.These algorithms explore the existence of multiple paths between two network nodes and route packets to minimize predictability.Routers compute all possible paths between a source-destination pair and, according to a given probability distribution, assign some probability to each nexthop.The net effect is that data packets traverse random paths on their way from the source to the destination.We should point out that, unlike the secure stochastic routing, SSR [12], we take a proactive and heuristic approach to making routing less vulnerable to attacks.In other words, according to partially detecting attacker's behavior, packets are always sent along multiple paths according to some probability.

Enhanced Heuristic-Stochastic Routing
We consider a MPLS network where multiple parallel LSPs exist between any given ingress LSR and egress LSR pair.The main objective is to distribute the traffic at each ingress LSR among the multiple LSPs so as to balance the load through the network and thus improving the network performance.Take the routing problem as a game between the network designer that specifies the routing algorithm and an adversary that attempt to intercept data in the network.We consider here a zero-sum game in which the designer wants to minimize the time it takes for a packet to be sent from node 1 to node n, and the adversary wants to maximize this time.To accomplish this, the adversary attempts to intercept the packet at particular links in the network.For short we say that the adversary scans link l when she attempts to intercept the packet at that link.

L 
We start by considering an on-line game in which the adversary selects a new link to be scanned every time the packet arrives at a new node and makes the selection knowing where the packet is, and the player determines a new path to forward data and makes the selection knowing the link to be scanned in the previous time.For generality, we take the probability of intercepting a packet to be link dependent and denote by l the probability of intercepting a packet traveling in link , given that link is being scanned by the adversary.P q q q q a qa b l , The state n is an absorbing state, i.e.,   , , t t t t P q q q n a l b l , The cost to be optimized is the average time it takes to send the package from node 1 to node n and can be written as: To optimize this cost, for each node the player that designs the routing chooses the distribution     The two-person zero-sum game just defined falls in the class of stochastic shortest path games considered in [12].In [12], it has been proved that the game exists a saddle solution point, however, In [13], the player just selects the stochastic next hop, it's too blind to do like this, and even if we do like this, it's still possible that the data can be Interception or eavesdropping by the adversary, and it's very possible to give birth to the routing loop.In our scheme, SHSR, we adjust every l based Bayesian principle termly, and then adjusts routing strategy to make the transmission more secure.People uses Bayesian principle to modify the prior probability, and get the new posterior probability constantly, here, we suppose the adversary has K types, here, the type means which link the adversary will attack, and has H possible actions, uses according to the probability formula: now, if we observe the adversary's action , we can forecast the new posterior probability that the adversary belongs to so the stationary Markov chain above can be re-written as: | p t and , the player will detect the link that the adversary has attacked continually, and adjusts his faith about the probability that the adversary attacks every link, sequentially changes his routing strategy.It's very possible that the adversary also will adopt similar attack strategy with the player, the process that the player and the adversary change their faith about each other and strategy makes up of the game between them.
 1 p t  

Simulation Results
To evaluate the routing algorithm proposed in Section 3, we simulated the network in Figure 2, data were transmitted from the blue point to the red point, using the ns-2 network simulator [13].In the simulations presented, all links have propagation delay of 25 ms and bandwidth of 2Mbps.Each queue implements drop-tail queuing discipline with maximum queue size set to 100 packets for the case of the CBR simulations.All packets are 400 bytes long.The simulation time for each trial was 20 seconds.Experiments data were performed using CBR according to TCP connecting.
Similar to SSR in [14], we were interested in determining the effect of ESHSR on security, drop-rate, and packet transmission delay.We assumed here that the attacker chooses the set of links that maximizes the percentage of packets seen, i.e., the worst-case scenario.Base on the on-line game, we evaluate the routing algorithm SSR and ESHSR (red line represents ESHSR, blue line represents SSR), and Figure 3 shows the simulation results of the percentage of packets seen of them respectively.As expected, ESHSR is most secure than SSR, since according as the adversary's behavior in history, packets will be transmitted in the more secure path, not   just be transmitted along stochastic path, Figure 4 shows the simulation results of average delay of SSR and ESHSR respectively, and Figure 5 shows the simulation results of drop-rate of SSR and ESHSR respectively, because under ESHSR, it's more difficult be seen than under SSR, the average delay and drop-rate under ESHSR are markedly smaller than in SSR.
to be scanned.
represents the prior probability that the adversary belongs to k  , and then, we can get: