Virtual Network with Security Guarantee Embedding Algorithms

Transcription

1 2782 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER 2013 irtual Network with Security Guarantee Embedding Algorithms Chiqiang Xing National Digital switching System Engineering & echnological R&D Center, Zhengzhou, China Julong Lan and Yuxiang Hu National Digital switching System Engineering & echnological R&D Center, Zhengzhou, China Abstract Network irtualization has been widely concerned as the new technology to remedy the current ossification Internet architecture. Preious irtual network (N) embedding algorithms focus on optimizing the use of resources with regard to performance with constraints on irtual nodes and links. here are few researches to analyze the security threat to the irtual network. In this paper, we first present two security threats of irtual network and inestigate them in depth. hen we propose a noel irtual network with security guarantee that will take the trust alue and security protection leel as the new security constraints during its embedding. he irtual network with security guarantee embedding algorithm is gien at last, and the simulation results show that the algorithm is effectie. Index erms irtual network embedding;security guarantee; trust alue; security protection leel. I. INRODUCION he packet switching used by the Internet Protocol (IP) has its own quality characteristics [1], and its simple structure that the network terminals are intelligent makes the Internet an important information infrastructure that supports economic deelopment, social progress and scientific innoation in modern society, while generates some new problems. First, the rigid Internet architecture, the single function of network layer and the excessie separation between the business and network cannot meet the dierse needs of the business; secondly, because the Internet cannot be changed, the implementation and promotion of new technologies, such as IP6, DiffSer and IntSer are ery difficult [2]. In recent years, network irtualization has been widely concerned by the industry and academia [3]. Network irtualization has been propounded as a fundamental diersifying attribute of the future internetworking paradigm that will allow multiple heterogeneous network architectures to coexist on a shared substrate [4]. It also enables researchers to design and ealuation new networking protocols on the heterogeneous experimental architectures [5] [6]. Constructing irtual backbone network in wireless sensor networks also has a wonderful performance in improing the performance of broadcast [7]. herefore, the network irtualization can support network technology innoations effectiely, make it possible to deploy new network architectures, protocols and applications without changing the existing network. It not only proides a feasible eolution way from the current network to the future, but also a key feature of the future Internet [8]. Preious irtual network embedding algorithms are aimed to archie optimization objects, such as the acceptance ratio, the aerage reenue/cost of the substrate network oer time and the load balancing, with constraints on irtual nodes and links. hese irtual network embedding (NE) algorithms can be diided into different kinds: uncoordinated and coordinated, static and dynamic, distributed and centralized, concise and redundant. Minlan Yu[9] designed a irtual link mapping algorithm and path migration algorithm based on the multi-commodity flow (MCF)under the condition that the substrate network support diision and migration, improed the link mapping phase of the baseline algorithm to get higher substrate network resource utilization. In literature [10], the main objectie of the irtual network embedding algorithm based on traffic constraint was to find a irtual network that can not only meet the traffic demand but also get efficient utilization of the substrate network resource. In literature [11] a irtual network mapping algorithm with QoS guarantees (QoSMap) was proposed, its basic idea is to map a irtual link onto a single link and use a hop relay routing node to proide high quality backup routing path for the irtual link, improing the quality and reliability of the irtual network. Literature [12] presents two algorithms in order to make the substrate network load balanced: irtual network mapping algorithm without reconfiguration and irtual network mapping algorithm with reconfiguration, the first algorithm s basic idea is to map all of the irtual nodes onto the substrate nodes with little load and nearer to the irtual node that hae already been mapped, then map the irtual link with the shortest path algorithm; the latter cyclically examine the load of the substrate nodes and links. And when the load exceeds a predefined threshold alue, the Ns mapped to the node or link will be re-mapped to eliminate resource bottlenecks. doi: /jcp

2 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER Nowadays, the issues of network security are becoming increasingly serious and attract more attention [13]. hese algorithms are proposed to increase the acceptance ratio that map as more as possible N requests onto specific nodes and paths in the substrate network with constraints on irtual links and nodes resources. Howeer, the security of the substrate hosts in the substrate network is not considered. Because all the irtual nodes of different N requests are mapped onto the substrate nodes, the N will be affected if the substrate nodes are under attack or cannot be trusted. In response to these problems, we will consider new security constraints that the substrate nodes need to meet the security requirements of the irtual networks during the mapping to guarantee the security of N and resist the possible attacks to the substrate hosts. he remainder of this paper is organized as follows. Section 2 formalizes the irtual Network with Security guarantee (SN) model and gies the description and ealuation of the embedding problems. Based on the Section 2, we proposed two algorithms: N embedding algorithm based on the greedy algorithm (NE-GA) without security constraints and N embedding algorithm with security constraints (SNE-GA), and then compared the both in Section 3. Section 4 presents simulations results that ealuate the proposed algorithms and Section 5 concludes the paper. II. IRUAL NEWORK MODEL AND EMBEDDING PROBLEM A. Model of the irtual Network with Security Guarantee he hierarchy architecture of Ns is shown in Figure 1. he substrate nodes that the irtual nodes are mapped onto are called host nodes. Figure 1. the hierarchy architecture of the irtual network he security of information is an issue related to information and communication technology and a management issue as well [14]. he risks that the N introduced can be categorized in two areas: A host node attacking one of its irtual nodes. he resources that the irtual node uses are ultimately the resources owned by host nodes. All the computations of the irtual node are carried out by the hardware and software of host nodes. he host nodes can monitor or change any act of the irtual nodes in theory. And irtual nodes hae no option to defend themseles. In this case we need to map the irtual nodes onto trusted host nodes. Other malicious nodes attacking the host nodes affecting the working of Ns. In this case, host nodes being under attack, the Ns cannot work properly, een causing information leakage. hen we need to consider the anti-attack ability, called security protection leel of the substrate nodes. And we can only map the irtual nodes onto the substrate nodes that hae a higher leel of the security protection. In the first case, we can use the trust relationship between nodes to get the trust alue as one of the security constraints on irtual nodes. Some researchers apply the trust relationship mechanism to the communication network to enable the network with intrusion tolerance and increase the security. Usually, the trust alue node A to node B is a decimal between 0 and 1. he higher alue indicates the more trust from A to B. Meanwhile, the trust relationship management is a complex process. he trust alue will decrease if B does some malicious actiities to other nodes that A recognizes. For simple, we will not consider the trust relationship between nodes and set the trust alue of each node a unified decimal between 0 and 1 in the whole network. For the second case, we can introduce the computer information system security protection leel of substrate nodes, which is defined in the GB Classified oriteria for security protection of computer information system, as another security constraint on irtual nodes. GB proides the computer information system security protection with fie leels: the first leel of user self-protection leel, the second leel of system audit protection leel, the third leel of security token protection leel, the fourth leel of structural protection leel and the fifth leel of access authentication protection leel. his proision applies to diide leel of the computer information system security protection. he security protection capabilities of the computer information system increase as the leel increases. he relationships between fie leels and security functions are shown in the able 1: ABLE 1 HE RELAIONSHIPS BEWEEN FIE LEELS AND SECURIY FUNCIONS Leel 1 Leel 2 Leel 3 Leel 4 Leel 5 Discretionary Access Control Y Y Y Y Y Mandatory Access Control - - Y Y Y Mark Y Y Y

3 2784 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER 2013 Authentication Y Y Y Y Y Object reuse - Y Y Y Y Audit - Y Y Y Y Data Integrity Y Y Y Y Y Coert channel analysis Y Y rusted path Y Y Credible analysis Y So, during the embedding process of SN, except for the remaining node CPU resources and link bandwidth need to meet the N request, the trust alue and security protection leel of substrate nodes should also meet the requirements of Ns. B. irtual Network Embedding Problem Substrate network. For the embedding problem of SN, the substrate network can be modeled as a weighted undirected graph and denote it by G s =( s, E s ), where s is the set of substrate nodes and E s is the set of substrate links. Each substrate node s s is associated with the CPU capacity weight alue C( s ), trust alue ( s ) and safety protection leel S( s ). Each substrate link e s (i, j) E s between two substrate nodes i and j is associated with the bandwidth capacity weight alue W(e s ) denoting the total amount of bandwidth. he C( s ), ( s ), S( s ) and W(E s ) are non-negatie, and ( s ) [0,1], S( s ) {1,2,3,4,5}. N request. Similar to the substrate network, we model N requests as weighted undirected graphs and denote a N request by G =(, E ). Likewise, Each irtual node is associated with the required CPU weight alue C( ), trust alue ( ) and safety protection alue S( ). Each irtual link e E is associated with the required bandwidth weight alue W(e s ) denoting the required bandwidth. he C( ), ( ), S( ) and W(E ) are non-negatie, and ( ) [0,1], S( ) {1,2,3,4,5}. Each N is also associated with the required time and lifetime. N Embedding. A irtual network embedding for a N request can be defined as a mapping: M : G ( s, P ), where s s is a subset of the substrate nodes and P P, the mapping result of E s,is a subset of the substrate links. he N embedding can be naturally decomposed into node and link mapping and both the s and P are required to meet certain constraints. For the node mapping M : s, mapping irtual nodes onto substrate nodes that all M ( ) S subject to RC( M ( )) C( ) M ( ( )) ( ) (1) SM ( ( )) S ( ) For the link mapping RW ( es) W ( e), es P( M ( u), M ( )) (2) RC( s ) and RW(e s ) denote the remaining CPU resources of node s and the remaining bandwidth of link e s, where u and are irtual nodes. Objecties. Our main interest is to propose an efficient embedding algorithm for the SN and compare with the N. Because of the additional security constraints, the security protection leel and the trust alue on irtual nodes during the embedding process of N, we redefine the reenue and cost of the substrate network as follows. Similar to the preious work, we introduce the notion of reenue that corresponds to the economic benefit of accepting the N request. We denote by R(G, t) the reenue of sering the N request at time t. Reenue gies an insight into how much an InP will gain by accepting a N request. RG (, t) = α W( e) + α C ( ) 1 2 e E + α ( k ( ) + k S( )) We also redefine the cost of accepting a N request as the sum of total substrate resources allocated to the N. CG (, t) = α We ( ) + α C ( ) 1 2 ' ' es P s s + α ( k ( ) + k S( )) ' s s And the following three ealuations can be gien: 1) Aerage Reenue Oer ime: 2) Acceptance Ratio: R/ = lim t= 0 lim 3) Aerage Reenue/Cost: lim RG (, t) S RG (, t) CG (, t) III. IRUAL NEWORK WIH SECURIY GUARANEE EMBEDDING ALGORIHM We employ a node mapping algorithm based on greedy, since it is computational too expensie to employ other strategies. In order to map irtual nodes onto substrate nodes based on the greedy algorithm, we gies the following definition to ealuate the remaining bandwidth of substrate nodes and the demand bandwidth of irtual nodes: From the substrate node s point of iew, we denote by PR(t, s ) the potential remaining bandwidth of the substrate node at time t as the sum of the remaining bandwidth of all the links that connect node s : (3) (4) (5) (6) (7)

4 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER PR(, t) = RWte (, ) (8) s es L( s) We also define by PR(t, ) the potential required bandwidth of the irtual node at time t as the sum of the required bandwidth of all the irtual links that connect to : PR(, t) = Wte (, ) (9) e L( ) Algorithm 1 N Embedding Algorithm based on Greedy Algorithm, NE-GA Step 1: Sort the irtual nodes according to the PR(t, ) and the substrate nodes according to the AR(t, s ) in descending order when the N request arries. Define N and N s as the number of irtual nodes and substrate nodes respectiely; Step 2: for i=1:n, j=1: N s, if RC( s ) j RC( ) i, map the irtual node i onto the substrate node sj. If all the irtual nodes are mapped successfully, go to Step 3; otherwise go to Step 5, the request fails; Step 3: e E, find the shortest path that meets the bandwidth requirement with the shortest path algorithm to complete the link mapping. If all the irtual links are mapped successfully, go to Step 4; otherwise go to Step 5, the request fails; Step 4: Update the substrate network with the node CPU resources and link bandwidth resources; Step 5: Waiting for the next N request arries and then go to Step 1. In the SN embedding algorithm, because of the additional security constraints we first need to remoe the substrate nodes that does not meet the needs of the SN request, and then perform the aboe NE-GA algorithm. Meanwhile, due to the "bucket effect" of the security, we let the security protection leel S and the trust alue of the irtual nodes to be same in the same SN. We can gie the SN embedding algorithm based on greedy. Algorithm 2 SN Embedding Algorithm based on Greedy Algorithm, SNE-GA Step 1: Remoe the substrate nodes that cannot meet the security constraints S, when the SN request arries, getting an aailable set of substrate nodes. Define N and N s as the number of irtual nodes and aailable substrate nodes respectiely. If N s >N, go to Step 2; otherwise go to Step 5, the request fails; Step 2: for i=1: N, j=1: N s, if RC( s ) j RC( ) i, then map the irtual node i onto substrate node sj. If all the irtual nodes are mapped successfully, go to Step 3; otherwise go to Step 5, the request fails; Step 3: e E find the shortest path that meets the bandwidth requirement with the shortest path algorithm to complete the link mapping. If all the irtual links are mapped successfully, go to Step 4; otherwise go to Step 5, the request fails; Step 4: Update the substrate network with the node CPU resources and link bandwidth resources; Step 5: Waiting for the next SN request arries and then go to Step 1. s Compared with the NE-GA, the SNE-GA algorithm will lead to lower acceptance ratio as the aailable substrate nodes decrease because of the security constraints on irtual nodes. he following simulation experiments and results also erified this situation. I. PERFORMANCE EALUAION In this section, we first describe the performance ealuation enironment, and then present our main ealuation results. Our ealuation focuses primarily on comparing the acceptance ratio, reenue and cost of substrate network in the aboe two algorithms. A. Ealuation Enironment We implement a N embedding simulator to ealuate our algorithms. Substrate network. We use the G-IM tool [15] to generate the substrate network topology. he substrate network is configured to hae 100 nodes and around 500 links, a scale that corresponds to a medium-sized ISP. he CPU resources at nodes and the link bandwidths at links follow a uniform distribution from 50 to 100 units. he trust alue and the security protection leel follow a uniform distribution from 0 to1 and 1to 5 (integer) separately. N request. he arrials of N requests are modeled by a Poisson process with mean fie requests per 100 time units. he duration of the requests follows an exponential distribution with 500 time units on aerage. In one N request, the number of N nodes is randomly determined by a uniform distribution between 2 and 20. Each pair of irtual nodes are randomly connected with probability 0.5. As well, the CPU requirements of the irtual nodes and the bandwidth requirements of the irtual links are uniformly distributed between 0 and 50. he required trust alue and the security protection leel follow a uniform distribution from 0 to1 and 1to 5 (integer) separately. In NE-GA and G-SP [11], because there are no security constraints, we set α 1 =α 2 =0.5, α 3 =0 in (3) (4). In SNE-GA, we set α 1 =α 2 =α 3 =1/3, and considering the alues of and S, set k 1 =25, k 2 =5.We run all of simulations for time units, which correspond to about 2500 requests on aerage in an instance of simulation. B. Ealuation Results We used seeral metrics in our ealuations to measure the performances of our algorithms against the G-SP and compare the N and SN. he metrics conclude the acceptance ratio, aerage reenue(r) and cost (C) oer time. We also ealuated and compared the acceptance ratio by changing the distributions that the trust alue and security protection leel of substrate nodes follow. In all cases we plot the performance metrics against time to show how each of these algorithms actually perform in the long run. We summarize our key obserations in the following. (1) NE-GA leads to higher acceptance ratio than G-SP and because of additional security constraints, SNE-GA leads to lower acceptance ratio than NE-GA. Fig.2 depict that without the security

5 2786 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER 2013 constraints, the acceptance ratio of NE-GA is 65%, achieing about 10% increases oer the G-SP. And the acceptance ratio of SNE-GA is about 53%, 12% decreasing oer the NE-GA. his is because that NE-SC takes the trust alue and security protection leel as the additional constraints on irtual nodes, the aailable substrate nodes decrease for each N request. can proide customized serices and security guarantee, making the reenue increase significantly and the cost increase slightly, leading significantly improement of the R/C. Fig.3 and Fig.4 gie the aerage reenue and cost oer time separately. We can see that because of the security constraints, the acceptance ratio decreased, as the cost and reenue NE-GA G-SP SNE-GA NE-GA SNE-GA Acceptance Ratio Reenue/Cost Figure 2. N request acceptance ratio. Figure 3. ime aerage of reenue/cost (2)NE-SC leads to higher reenue/cost than NE-GA. Figure 3 shows the result. Because of the additional components (trust alue and security protection leel in (3) (4))in the reue and cost, the substrate network Aerage Cost Oer ime NE-GA SNE-GA Aerage Reenue Oer ime NE-GA SNE-GA 10 6 Figure 4. Aerage cost oer time reenue oer time Figure 5. Aerage (3) he substrate network with higher trust alue and security protection leel will lead to higher acceptance ratio. From the results aboe we can see that the acceptance ratio decreased due to additional constraints of the security protection leel and trust alue on irtual nodes. herefore, the different distributions of the trust alue and security protection leel (called as the ability ) of the substrate network will lead to different acceptance ratio. he following simulation shows this situation too. he different distributions of the trust alue and security protection leel of the substrate network are shown in able 2. We can see that the ability of substrate nodes in Distribution 2 is stronger than Distribution 1. ABLE 2 DIFFEREN DISRIBUIONS OF HE RUS ALUE AND SECURIY PROECION LEEL Security protection leel rust alue Distribution1 Distribution2 Leel 1 [0,0.2) Leel 2 [0.2,0.4)

6 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER he result of acceptance ratio is shown in Fig.5. he acceptance of Distribution 2 is 57%, 4% (about 100 SN requests) increase oer Distribution 1. his is because the "ability" of substrate nodes in Distribution 2 is stronger than Distribution 1, leading the number of aailable substrate nodes increasing for each N request, and the higher acceptance ratio. Acceptance Ratio Leel 3 [0.4,0.6) Leel 4 [0.6,0.8) Leel 5 [0.8,1] Distribution1 Distribution2 Figure 6. SN request acceptance ratio. CONCLUSION Network irtualization has been propounded as a fundamental diersifying attribute of the future internetworking paradigm that will allow multiple heterogeneous network architectures to coexist on a shared substrate. Howeer, being the foundation critical infrastructure, N must strie to ensure a certain leel of security. In this paper we first analyzes two security risks that the network irtualization introduced and then proposes a new irtual network model with security guarantee that take the trust alue and security protection leel of substrate nodes as new constraints. Finally, the irtual network with security guarantee embedding algorithm based on greedy algorithm is gien and the simulation results show that the algorithm is effectie and alidate. We also proed that the substrate nodes with higher security ability can increase the acceptance ratio. Meanwhile, the irtual network model with security guarantee in this paper is a new consideration that needs further study in many ways: First, the dynamically management and update of the trust relationships between substrate nodes. Second, the improement of the SN embedding algorithm, especially considering how to use the intelligence optimization algorithm to increase the acceptance ratio and make the substrate network load balanced are the future works. ACKNOWLEDGMEN his paper is originated from a project numbered as 2012CB315901, which is supported by national 973 foundations, and 2011AA01A103, 2011AA01A101, which is supported by national 863 foundations. REFERENCES [1] CHENG Dong-nian, WANG Bin-qiang, WANG Bao-jin, ZHANG Jian-hui. Preliminary study on the connotation of flexibility in dynamically reconfigurable networks[j].journal of Communication, olume33, Number8, pages , 2012 [2] Secure irtual network embedding. Praxis der Information serarbeiting and Kommunikation, olume34, Number4, pages , [3] CHENG Xiang, ZHANG Zhong-bao, SU Sen, YANG Fang-chun. Surey of irtual network embedding problem[j]. Journal of Communication, olume32, Number10, pages ,2011. [4] Mosharaf Chowdhury, Muntasir Raihan Rahman, Raouf Boutaba. ineyard: irtual Network Embedding Algorithms With Coordinated Node and Link Mapping. IEEE/ACM RANSACIONS ON NEWORKING, OL. 20, NO. 1, FEBRUARY2012. [5] GENI:Global enironment for network innoations, [Online].Aailable: [6] A. Baier, N. Feamster, M. Huang, L. Peterson, and J. Rexford, In INI eritas: Realistic and controlled network experimentation, in Proc. ACM SIGCOMM, 2006, pp [7] Yong ang, Yu Xiang, et al. Leeraging 1-hop Neighborhood Knowledge for Connected Dominating Set in Wireless Sensor Networks. Journal of Computers, 2012, 7(1), [8] FEAMSlER N, GAO L, REXFORD J. How to lease the Interact in your spare time[j]. ACM SIGCOMM Computer Communication Reiew 2007, 37(1): [9] YU M, YI Y, REXFORD J, et a1. Rethinking irtual network embedding: substrate support for path splitting and migration[j], ACMSIGCOMM Computer Communication Reiew, 2008, 38(2): [10] LU J. URNER J, Efficient Mapping of irtual Networks onto a Shared Substrate[R]. Department of Computer Science and Engineering, Washington Uniersity, [11] SHAMSI J, BROCKMEYER M. QoSMap:QoS aware mapping of irtual networks for resiliency and efficiency[a]. Proceedings of the IEEE GLOBECOM Workshop [C] Washington. IX, USA, [12] ZHU Y, AMMAR M. Algorithms for assigning substrate network resources to irtual network components[a]. Proceedings of the IEEE INFOCOM[C]. Barcelona, Spain, 2006, pp [13] Ye Du Jiqiang, Liu Ruhui, et al. A Dynamic Security Mechanism for Web Serices Based on NDIS Intermediate Driers. JOURNAL OF COMPUERS. 2011, 6(10), [14] Kuo-Hsiung Liao, Hao-En Chueh. Medical Organization Information Security Management Based on ISO27001 Information Security Standard. Journal of Software, 2012,7(4), [15] E. W. Zegura, K. L. Calert, and S. Bhattacharjee. How to model an internetwork. In Proc. IEEE INFOCOM, 1996.

7 2788 JOURNAL OF COMPUERS, OL. 8, NO. 11, NOEMBER 2013 Chiqiang Xing is currently a Post-Graduate student at the National Digital switching System Engineering & echnological R&D Center, Zhengzhou, China, superised by Professor Julong Lan. His research interests include information network, network security and new network architecture. Julong Lan is currently a professor at the National Digital switching System Engineering & echnological R&D Center, Zhengzhou, China. His research interests include new network architecture and Network modeling. Yuxiang Hu is currently a lecturer at the National Digital switching System Engineering & echnological R&D Center, Zhengzhou, China. His research interests include new network architecture, routing and switching technologies.