Preventing “Bad” Content Dispersal in Named Data Networking

Yi Wang, Zhuyun Qi＊, Bin Liu

1 Southern University of Science and Technology, Shenzhen 518055, China

2 Peking University, Shenzhen 518055, China

3 Tsinghua University, Beijing 100084, China

I. INTRODUCTION

In the conventional IP network, security is a function of the connection between hosts. By contrast, Named Data Networking [1] (NDN)incorporates security into data itself by forcing the content provider to sign every content with its private key. The signature of a content, implying the data provenance, allows the security of this content to be decoupled from where and how the data is obtained.

In theory, content signatures provide an effective and simple means to detect content poisoning attacks, since “bad” content can be easily identified via signature verification.In other words, NDN should be immune to content poisoning attacks which includes two means: corrupted content, i.e., the content cannot conform its signature; and faked content,i.e., the private key of the signature is forged.However, this assertion might not hold in practice. Though a consumer can afford to verify all content signatures, NDN routers, caching contents to improve network performance,face two challenges: (1) signature verification overhead; and (2) public key management. An effective public key cryptography has been considered by Lixia Zhang et al. [2]. This paper focuses on reducing the overhead of signature verification.

In this paper, the authors propose two content verification schemes to improve the verification performance of NDN routers to prevent content poisoning.

1.1 Problem statement

When a router receives a Data packet, as described in the NDN proposal [1], the router should systematically verify the content signature to avoid content poisoning attacks before forwarding and caching the content.The signature binds the content with its name,and provides original authentication no matter how, when or from where the content is retrieved. However, for NDN routers, the computations of content verification are too expensive to be executed against all the incoming Data packets. Previous work [3] showed that an optimized software implementation of RSA1024 signature verification running on Intel Core 2 Duo 2.53 GHz CPU allows a router to verify about 150 Mbps of traffic, assuming 1,500 Bytes per content packet, or even worse with smaller-sized packets. NDN routers with multiple Gigabit-speed interfaces would need an unrealistic amount of computing power to verify signatures of Data packets at wire speed.

Participants:1) End users request content by generating Interests; 2) Content Providers produce content by generating data in response to Interests; 3) Network Devices forward Interest and Data packets.

Assumptions:Routers systematically verify the signatures of the contents that they receive. Routers cache verified contents that can be used to directly respond to incoming Interest packets. Routers in the network are trustworthy and can identify each other. Users are trustless. And some users are the attackers or adversaries.

Attack:Adversaries disguise themselves as content providers to produce corrupted and faked contents to routers and users.

Impact:By fully exploiting the expensive computations of content verification, an adversary can attack the router and lead it to be out-of-service ultimately [4]. As a result, NDN routers sample a small set of received Data packets to verify their signatures, and leave a loophole which can be employed by adversaries to fill the caches of NDN routers with corrupted or faked contents.

Countermeasures:New content verification schemes are proposed in this paper Routers to improve the performance of content verification.

1.2 Our contribution

To prevent content poisoning in NDN, in this paper, we propose two verification schemes to improve the content verification performance:

1) The user-assisted content verification scheme (described in Section III) no longer verifies the content itself, instead it verifies the content provider to guarantee the correctness.By bypassing the content verification in NDN routers, this scheme can achieve perfect performance. But this scheme is under the threat of a scenario that customers and attackers collude in requesting and providing poison contents.

2) To prevent the aforementioned collusion attack, we improve the user-assisted content verification scheme and propose the Router-Cooperation content verification scheme.The Router-Cooperation content verification scheme (described in Section IV) lets the edge routers verify the content provider independently without the assistance of users. By replacing the asymmetric encryption operation with symmetric encryption operation, this Router-Cooperation scheme can effectively reduce the consumption of computing resource for cryptographic operation and improve the content verification performance. The simulation results demonstrate that this Router-Cooperation scheme can speed up 18.85 times of the original content verification scheme used in NDN [1].

The rest of the paper is organized as follows. Section II introduces the packet categories and the original content verification scheme in NDN. The user-assisted content verification scheme and the Router-Cooperation content verification scheme are described in Section III and Section IV, respectively.Then we present the simulation results in Section V. After reviewing the related literatures on NDN security in Section VI, we conclude our work in Section VII.

II. BACKGROUND: CONTENT VERIFICATION MECHANISM IN NDN

2.1 Packet categories in NDN

NDN is a request-driven networking architecture. Different with the sole IP packet in the conventional IP network, packets in NDN have two categories: Interest packet and Data packet. Interest packets are used to request contents, and Data packets are used to carry contents back. As depicted in figure 1, an Interest packet consists of the required content name, a selector, and a nonce; a Data packet carries the required data with the name, the signature and the signed information. When a data is returned to the client, the client will verify the signature to confirm whether this content is the exact one.

2.2 Content verification mechanism

To prevent polluted data from being spread out in the network, NDN routers should verify every content sent to the networks by the content providers or other clients. Figure 2 illustrates the content verification process in NDN.

1) When the router receives the Data packet, it parses the Data packet and extracts the content D′, the signature Kpri(Hash(D)), and other information;

2) The router calculates the hash value Hash(D′) of the content D′ by applying the message-digest algorithm (e.g., MD5,CRC32);

3) The router uses the content provider’s public key to decrypt the signature of this content to obtain the original message-digest Kpub(Kpri(Hash(D)));

4) The router compares the message-digest Hash(D′) of the sent content with the original message-digest Kpub(Kpri(Hash(D))) of the required content, if these two message-digests are equal, this Data packet will be sent upstream; otherwise, it will be dropped.

The content verification mechanism in NDN can work well in the low speed networks, but it cannot work in the high speed networks due to its poor verification performance.

III. THE USER-ASSISTED CONTENT VERIFICATION SCHEME

As described in Section II-B, the computational complexity of the content verification is too high for an NDN router to handle wire speed traffic. Therefore, in this section, we propose the user-assisted content verification scheme to improve the performance by bypassing the content verification in NDN routers. Similar to the client verification in current IP network,the user-assisted content verification scheme no longer verifies the content itself, but it verifies the content provider to guarantee the correctness.

Fig. 1. Two kinds of packet in NDN.

Fig. 2. The content verification process in NDN.

The process of the user-assisted content verification, demonstrated in figure 3, has 4 steps.

1) The Interest packet, sent by the user A,carries 3 extra data to help routers verify the content provider:

a) Kpub(KA). KAis the temporary symmetric key for the content provider to encrypt the verification message; Kpubis the public key of the content provider; and Kpub(KA), the ciphertext of encrypting KAwith Kpub, can only be decrypted by the content provider with its private key.

b) SA, a token, is a 64-bit random integer.

c) KA(SA), the ciphertext of encrypting SAwith KA, is used to verify whether the content provider can obtain KAcorrectly by decrypting Kpub(KA).

2) When the edge router receives this Interest packet, it extracts KA(SA) from the Interest packet before forwarding it to the content provider. The rest two data Kpub(KA) and SAwill be still forwarded.

3) The content provider decrypts Kpub(KA)with its private key Kprito obtain K′A=Kpri(K-pub(KA)) when it receives the Interest packet.Then the content provider encrypts SAwith K′Aand returns the ciphertext K′A(SA) in Data packet to the edge router.

Fig. 3. The content provider verification process.

4) After receiving the Data packet, the edge router extracts K′A(SA) and compares it with KA(SA) to validate the content provider. If K′A(SA)=KA(SA), it means the content provider is the exact one that publishes the required content, and then the edge router will remove K′A(SA) from this Data packet and forwards it upstream on the corresponding reverse path;otherwise, the content provider is a fake,hence this Data packet will be dropped.

For the Interest packets with the same required content name, routers in NDN can merge these Interest packets to one, regardless of whether these Interest packets have different tokens and different temporary symmetric keys. For example, in figure 3, user A and user B send the Interest packets with (SA, KA)and (SB, KB), respectively. These two Interest packets are converged in the edge router, and only one of them is sent to the content provider.

Note that any in-path router can return the Data packet without K′A(SA) whenever it caches the required content, since routers trust each other and contents cached in routers have been verified.

Meanwhile, in the user-assisted scheme,routers must identify the roles of the upstream nodes. If the upstream node is a router, the token KA(SA) will be forwarded to it; otherwise,the token KA(SA) will be removed from the Interest packet.

The correct Data packet is sent back to all the users that required this content. Meanwhile, the edge router caches the content to fast reply the same request and reduce the traffic.

IV. THE CONTENT VERIFICATION SCHEME BASED ON THE COOPERATION OF ROUTERS

The computation overhead is reduced dramatically in the user-assisted content verification scheme. Routers verify contents by verifying contents’ providers instead of verifying the signatures. The hash calculation of the content and the decryption of its signature are never needed in the user-assisted content verification scheme.

The user-assisted content verification scheme, however, is vulnerable. The malicious user and the counterfeit content provider can collude on the temporary symmetric key KAand the token SAto deceive the edge router and publish polluted contents. First, the malicious client sends an Interest packet with Kpub(KA),KA(SA), and SAas usual. When the counterfeit content provider receives this Interest packet,it sends the Data packet carrying the polluted content and KA(SA) to the edge router directly without decrypting Kpub(KA). The edge router cannot detect this polluted content as KA(SA)extracted from the Interest packet is equal to K′A(SA) extracted from the Data packet. As a result, the counterfeit content provider can publish any polluted content in the network.

To prevent the aforementioned collusion attack, we improve the user-assisted content verification scheme and propose the Router-Cooperation content verification scheme,in which the edge routers verify the contents independently without the assistance of users and the core routers no longer verify the contents. In a single Autonomous System (AS),the routers are in one trust group, i.e., routers trust each other. The Router-Cooperation scheme works well in a single AS. For inter-ASes scenarios, each edge router of an AS should verify the contents independently. As illustrated in figure 4, the edge router produces the token SAand the temporary symmetric key KA. The verification process works as follows:

1) The client sends out the Interest packet to require the content.

2) When the edge router receives this Interest packet, the edge router generates the symmetric key KAand then adds the ciphertext Kpub(KA) to this Interest packet.

3) After receiving this Interest packet, the content provider first decrypts the ciphertext Kpub(KA) with its private key Kprito obtain K′A; then it hashes the content to get the message-digest Hash(D); finally, it encrypts Hash(D) with K′A, and appends K′A(Hash(D))to the Data packet.

4) When the edge router receives this Data packet, it hashes the content to get Hash(D′)and encrypts Hash(D′) with KAto get the verification message KA(Hash(D′)). Then the router compares KA(Hash(D′)) with K′A(Hash(D))to verify the content provider and check the content’s integrity. Only the content, passing this verification, will be cached and forwarded to upstream.

5) The core routers directly forward the Data packets without verifying the contents any more since the all contents have been verified by the edge routers.

Compared to the user-assisted content verification scheme, the Router-Cooperation content verification scheme consumes more resources. In the path of Interest packets, an edge router in the Router-Cooperation content verification scheme has to produce the temporary symmetric key KAfor every Interest packet that is forwarded by this edge router.Besides that, the edge router needs to encrypt KAwith the provider’s public key Kpubto set up Router-Cooperation communication. Here,KAand Kpub(KA) can be precalculated off line to improve the performance. Furthermore, the edge router can allocate one symmetric key KAfor each content provider which directly connects the edge router instead of allocating a temporary symmetric key KAfor every Interest packet.

Fig. 4. The process of Router-Cooperation content verification.

In the path of Data packets, an edge router in the Router-Cooperation content verification scheme needs to hash the content to get the message-digest Hash(D′) and encrypts this message-digest with the symmetric key KAto encrypt the verification message KA(Hash(D′)).The verification message KA(Hash(D′))should be computed online, since the above calculation process is correlation with the content. In summary, the computing overhead of the Router-Cooperation content verification scheme consists of one hash calculation and one symmetric encryption computation.Compared with the original content verification scheme in NDN, the Router-Cooperation content verification scheme can effectively improve the verification performance since it replaces the asymmetric decryption computation with the symmetric encryption computation. Meanwhile, the verification message is only transferred once, i.e., from the content provider to the edge router. The edge router can directly forward the content to upstream routers without verification information(KA(Hash(D′)))) as routers trust each other and contents in the edge router have been verified.In the other cases, a router can directly response the Interest packet and sends back the Data packet without KA(Hash(D′)) whenever the required content is cached in the router.Hence, compared to the original verification scheme in NDN, the extra transmission overhead of the Router-Cooperation content verification scheme is 80 Bytes.

In the Router-Cooperation scheme, the edge router can off line generate the temporary symmetric key Kpub(KA) for one content provider and reuse the key Kpub(KA) for all Interest packets that require the same provider’s contents. An adversary may collect enough segments of KA(Hash(D′)) to crack the symmetric key KAby listening the path from the edge router to the content provider. Therefore,we should regenerate the temporary symmetric key KBfor the content provider after using the key KAafter some time. Assuming a node,in-between the edge router and the content provider, caches the requested content with the key KA. If the edge router uses the KAas the symmetric key, this copy will be valid; otherwise, this copy will be invalid, and the content from the original provider will be sent back.Since one content provider is only assigned one symmetric key KAover a period of time,the situation that the copy’s key cannot match the original content’s key happens in the interval of replacing the old key KAwith the new key KB.

V. SIMULATION RESULT

In this section, we evaluate the performance of the aforementioned 3 content verification schemes via simulation. The verification speed and the transfer overhead are the major metrics to evaluate the content verification schemes. In brief, the original content verification scheme is named as NDNVerify; the user-assisted content verification scheme is named as User-Verify; and the Router-Cooperation content verification scheme is named as Router-Verify.

The basic performances [5] of the message-digest algorithms, the symmetric cryptographic algorithms, and the asymmetric cryptographic algorithms are listed in Table I and Table II. All algorithms are coded in C++,compiled with Microsoft Visual C++ 2005 SP1, and run on an Intel Core 2 1.83 GHz processor under Windows Vista in 32-bit mode.X86/MMX/SSE2 assembly language routines are used for integer arithmetic and SHA-256.OpenMP [6] is disabled so that only one core of the CPU is used for this benchmark.

In the message-digest algorithms, CRC32,MD5 and SHA-256 can achieve 253 MB/s,258 MB/s and 111 MB/s, respectively. Compared to SHA-256, MD5 is 2.32 times faster.On the other hand, CRC32 produces a 32-bit hash value, MD5 produces a 128-bit hash value, and SHA-256 produces a 256-bit hash value. After considering both the performance and the application situation, in practice we choose MD5 as the message-digest algorithm.

DES algorithms uniformly use the CTR mode. The encryption speed of DES, DESXEX3 and DES-EDE3 can achieve 32MB/s, 29MB/s and 13MB/s, respectively. The AES algorithms using the CTR mode with different key lengths of 128-bit, 192-bit,and 256-bit can achieve 139MB/s, 113MB/s, 96MB/s. Compared to DES-XEX3, AES/CTR algorithms can speed up the decryption speeds 4.79 times, 3.89 times, and 3.31 times,respectively. Through overall consideration,we choose DES-XEX3 as the symmetric cryptographic algorithm.

Asymmetric cryptographic algorithms, e.g.RSA shown in Table II, is 23 orders of magnitude slower than symmetric cryptographic algorithms. Meanwhile, the processes of encryption and decryption are different in an asymmetric cryptographic algorithm. Generally, the encryption is 1～2 orders slower than the decryption. For example, as illustrated in Table II, RSA1024 can encrypt 12,500 keys per second or decrypt 684.93 keys per second, the encryption is 18.25 times faster than the decryption. Furthermore, the speed gap between decryption and encryption of RSA increases as the length of ciphertext grows. In our simulation, we apply RSA1024 as our asymmetric cryptographic algorithm.

5.1 The speed of content verification

The content verification overhead consists of two parts: the time for calculating hash value and the time for encryption and decryption.The time for calculating hash value of a content will increase as the content itself grows,and the time for encryption and decryption varies as using different cryptographic algorithms. In this section, therefore, we evaluate the verification speeds of different schemes with different content sizes. Among these simulations, we use MD5 as the message-digest algorithm, RSA1024 as the asymmetric cryptographic algorithm, and DES-XEX3/CTR as the symmetric cryptographic algorithm.

Table III illustrates the speeds of differ-ent schemes to verify a 1000 Bytes content.NDN-Verify costs 0.0839 ms to verify this content, which consists of 0.0039 ms to calculate the hash value of the content, and 0.08 ms to decrypt the digital signature. Since User-Verify assigns all computing tasks to users and content providers, the computing cost of the content verification in the edge router is 0. In online work mode, Router-Verify should encrypt the symmetric key KA with the content provider’s public key, hash the receiving content, and encrypt the hash value to confirm the content’s correctness. Hence Router-Verify with online work mode costs 1.46445 ms to verify a 1000 Bytes content. In off line work mode, however, Router-Verify only needs to hash the content and encrypt the hash value to verify the content, it only costs 0.0045 ms. In summary, Router-Verify with off line work mode can achieve 18.85 times speedup of NDN-Verify by replacing the asymmetric cryptographic algorithm with the symmetric cryptographic algorithm.

Table I. The basic performances of different algorithms.

Table II. The basic performances of asymmetric cryptographic algorithms.

Table III. The speeds of different schemes to verify a 1000 Bytes content.

To verify a content with 1000 Bytes, Router-Verify (off line) costs 0.0039 ms to hash the content, and 0.00055 ms to encrypt the hash value. Therefore, hash calculation occupies 87.64% of the total time and becomes the bottleneck of the content verification. Figure 5 clearly demonstrates the trend of verification speeds of different schemes as the content size grows. From figure 5, we can conclude that the time of different schemes costing on content verification are linearly increasing as the content size grows. In NDN-Verify, with the content size varying from 64 Bytes to 1472 Bytes, the time of MD5 costs from 0.31% to 0.67% of the total time. Hence RSA1024 is the bottleneck of the system. After replacing RSA1024 with DES-XEX3, in Router-Verify,the time of MD5 costs 31.09% of the total time to verify a content with 64 Bytes, and 91.21% of the total time to verify a content with 1472 Bytes. Therefore, MD5 becomes the system’s bottleneck as the content size grows. However, the verification speed can be improved by applying the hardware module to accelerate the hash calculation in a hardware router.

Fig. 5. The trend of verification speeds of different schemes as the content size grows.

5.2 The transfer overhead

Different schemes append different data (token, signature, etc.) to implement content verification. The transfer overheads of different schemes are listed in Table IV, where M is the number of hops from the user to the content provider. NDN-Verify appends the signature of the content to the Data packet, and it costs 64M Bytes after transferring M hops. To verify the content provider, the Interest packet in User-Verify carries the token SA, the symmetric key KAand the ciphertext Kpub(KA),and the Data packet in User-Verify carries the identification message K′A(SA) in addition to the signature of the content. Since the identification message is removed by the edge router,the total transfer overhead of User-Verify is 8+144M which is 2.25of NDN-Verify’s transfer overhead. In addition to NDN-Verify,Router-Verify only sends a ciphertext Kpub(KA)and an identification message K′A(Hash(D)),hence the transfer overhead of Router-Verify is 80 + 64M. Compared with NDN-Verify,Router-Verify merely costs extra 80 Bytes to implement 18.85speedup whatever the number of hops between the user and the content provider.

VI. RELATED WORK

Nowaday, the research of NDN in industry and academia mainly focus on the wire speed name lookup [7], [8], [9], [10], [11],[12], effective cache strategy [13], [14], [15],fast forwarding mechanism [16], [17], NDN applications [18], [19]. These research has demonstrated the scalability, feasibility and practicality of NDN, and has promoted the development of NDN.

Security mechanism, as an important part of NDN, has attracted widespread attention.Custom private protection [20], [21] and detection of cache missing attack [22], [23] as the two major security mechanisms have been studied. Custom private leakiness is intro-duced by the cache mechanism in NDN, i.e.,the contents cached in routers can leak the privacy of users. Specifically, a malicious cracker can violate users privacy through the response time of some specific Interest packets, since the response time of a required content cached in the router is smaller than the fresh content’s response time. Gasti et al. [20], [21] propose a few mechanisms to protect custom’s private by increasing the response time. The purpose of cache missing attack is to reduce the cache hit ratio in an NDN router. An attacker sends a large amount of unpopular Interest packets to the router to increase the traffic and speed up the frequency of cache replacement. The protection mechanisms, proposed in paper[22], [23], detect the attackers by counting the number of Interest packets and measuring the popularity of Interest packets sent by the user. When a user is deemed as an attacker, the router will drop all Interest packets come from this attacker to prevent cache missing attack.Different from the custom private protection,our work focuses on preventing corrupted and faked contents be spread in the network by improving the content verification speed.

NDN applies content-based [24], rather than connection-based, security mechanism which allows users to authenticate the returned content regardless of where it comes from the original source, or the cache in a route. Decoupling the content from the content provider,i.e., decoupling “what” from “where” requires security and network primitives that can refer to, and authenticate content itself, rather than the host and file containers where it resides.Similar to prior works [25], [26], [27], NDN authenticate the linkage between a name and a content, rather than authenticating the content or its publisher. The signature of both the name and the content represents their linkage.Users or in-path routers can authenticate the name and the content by verifying their signature. The verification process is complex and time-consuming, and is much slower than the line rate [28]. Based on this observation,Giuseppe et al. [29] devise a simple analytical approach which permits to assess performanceof an LRU caching strategy storing a randomly sampled subset of requests. This work is orthogonal with our work.

Table IV. The transfer overheads of different schemes (M is the number of hops from the user and the content provider).

LIVE [30] allows content providers to control content access in NDN nodes by selectively distributing integrity verification tokens to authorized nodes. LIVE is effective and helpful for the authorized nodes to verify the specified contents. However, LIVE is helpless and opaque for the other contents that are free to be cached in routers. Our fast content verification mechanisms complement LIVE by improving the verification performance of all contents.

To our best knowledge, we first propose the Router-Cooperation content verification scheme to improve the performance of content verification in NDN routers. By replacing the asymmetric decryption with symmetric decryption, Router-Verify can speed up 18 times of NDN-Verify with negligible extra transfer overhead.

VII. CONCLUSION

In this paper, we propose two content verification schemes to improve the verification performance of NDN routers to prevent content poisoning. The user-assisted content verification scheme verifies content providers to guarantee the correctness of contents. By bypassing the content verification in NDN routers, the user-assisted scheme can achieve perfect performance. However, this scheme can only be applied in the environment that all consumers can be trusted.

The Router-Cooperation content verification scheme verifies the content provider independently without the assistance of users. By replacing the asymmetric encryption operation with symmetric encryption operation, the Router-Cooperation scheme can effectively reduce the computing resource for cryptographic operation and improve the content verification performance. The simulation results demonstrate that this Router-Cooperation scheme can achieve 18.85 times speedups of the original content verification scheme used in NDN.

To verify a content with 1000 Bytes, the Router-Cooperation scheme costs 0.0039 ms on hashing the content, and 0.00055 ms on encrypting the hash value. Therefore, hash calculation occupies 87.64% of the total time and becomes the bottleneck of the content verification. In our further work, we can implement the message-digest algorithm in hardware to improve the total performance of the Router-Cooperation content verification scheme.

ACKNOWLEDGEMENT

This work has been financially supported by Shenzhen Key Fundamental Research Projects(Grant No.: JCYJ20170306091556329).

[1] L. Zhang, D. Estrin, V. Jacobson, and B. Zhang,“Named data networking (ndn) project. http://www.named-data.net/,” 2010. [Online]. Available: http://www.named-data.net/

[2] L. Zhang, D. Estrin, J. Burkeand, V. Jacobson, J. D.Thornton, E. Uzun, and B. Zhang. “Named data networking (ndn) project 2011 - 2012 annual report”. [Online]. Available: http://named-data.net/wpcontent/uploads/2013/08/ndn-proj-pub.pdf

[3] P. Gasti, G. Tsudik, E. Uzun, and L. Zhang, “DoS and DDoS in Named-Data Networking,” ArXiv e-prints, Aug. 2012.

[4] T. Lauinger, “Security & scalability of content-centric networking,” September 2010. [Online]. Available: http://tuprints.ulb.tudarmstadt.de/2275/

[5] Crypto++ 5.6.0 benchmark. [Online]. Available:http://www.cryptopp.com/benchmarks.html

[6] The OpenMP API specification for parallel programming. [Online]. Available: http://openmp.org/wp/

[7] Y. Wang, K. He, H. Dai, W. Meng, J. Jiang, B. Liu,and Y. Chen, “Scalable name lookup in ndn using effective name component encoding,”Proc.IEEE 32nd International Conference on Distribut-ed Computing Systems (ICDCS),june 2012, pp.688–697.

[8] Y. Wang, H. Dai, J. Jiang, K. He, W. Meng, and B.Liu, “Parallel name lookup for named data networking,”Proc. IEEE Global Telecommunications Conference (GLOBECOM),dec. 2011, pp. 1 –5.

[9] Y. Wang, Y. Zu, T. Zhang, K. Peng, Q. Dong, B.Liu, W. Meng, H. Dai, X. Tian, Z. Xu, H. Wu,and D. Yang, “Wire speed name lookup: A gpu-based approach,”Proc. the 10th USENIX Conference on Networked Systems Design and Implementation,ser. nsdi’13. Berkeley, CA,USA: USENIX Association, 2013, pp. 199–212.[Online]. Available: http://dl.acm.org/citation.cfm?id=2482626.2482647

[10] Y. Wang, T. Pan, Z. Mi, H. Dai, X. Guo, T. Zhang,B. Liu, and Q. Dong, “NameFilter: Achieving fast name lookup with low memory cost via applying two-stage Bloom filters,”Proc. INFOCOM 2013 mini conference, IEEE, 2013.

[11] Y. Wang, D. Tai, T. Zhang, J. Lu, B. Xu, H. Dai,and B. Liu, “Greedy name lookup for named data networking,”Proc. the ACM SIGMETRICS/International Conference on Measurement and Modeling of Computer Systems,ser. SIGMETRICS ’13. New York, NY, USA: ACM, 2013, pp.359–360. [Online]. Available: http://doi.acm.org/10.1145/2465529.2465741

[12] Y. Wang, H. Dai, T. Zhang, W. Meng, J. Fan, and B. Liu, “Gpu-accelerated name lookup with component encoding,”Computer Networks, vol.57, no. 16, pp. 3165 – 3177, 2013.

[13] H. Wu, J. Li, Y. Wang, and B. Liu, “Emc: The effective multi-path caching scheme for named data networking,”Proc. Computer Communications and Networks (ICCCN), 2013 22nd International Conference on,2013, pp. 1–7.

[14] J. Li, H. Wu, B. Liu, J. Lu, Y. Wang, X. Wang, Y.Zhang, and L. Dong, “Popularity-driven coordinated caching in named data networking,”Proc.the Eighth ACM/IEEE Symposium on Architectures for Networking and Communications Systems,ser. ANCS ’12. New York, NY, USA: ACM,2012, pp. 15–26. [Online]. Available: http://doi.acm.org/10.1145/2396556.2396561

[15] E. Rosensweig, D. Menasche, and J. Kurose,“On the steady-state of cache networks,”Proc.INFOCOM, 2013 Proceedings IEEE, 2013, pp.863–871.

[16] H. Yuan, T. Song, and P. Crowley, “Scalable ndn forwarding: Concepts, issues and principles,”Proc. International Conference on Computer Communications and Networks (ICCCN),2012,pp. 1–9.

[17] C. Yi, A. Afanasyev, I. Moiseenko, L. Wang, B.Zhang, and L. Zhang, “A case for stateful forwarding plane,”Computer Communications, vol.36, no. 7, pp. 779 – 791, 2013.

[18] V. Jacobson, D. K. Smetters, N. H. Briggs, M.F. Plass, P. Stewart, J. D. Thornton, and R. L.Braynard, “VoCCN: voice-over content-centric networks,”Proc. the 2009 workshop on Re-architecting the internet,ser. ReArch ’09. New York,NY, USA: ACM, 2009, pp. 1–6.

[19] Z. Zhu, S. Wang, X. Yang, V. Jacobson, and L.Zhang, “Act: Audio conference tool over named data networking,”Proc. the ACM SIGCOMM Workshop on Information-centric Networking,ser. ICN ’11. New York, NY, USA: ACM, 2011, pp.68–73.

[20] G. Acs, M. Conti, P. Gasti, C. Ghali, and G. Tsudik,“Cache Privacy in Named-Data Networking,”Proc. Distributed Computing Systems (ICDCS),2013IEEE 33rd International Conference on,2013, pp. 41–51.

[21] S. DiBenedetto, P. Gasti, G. Tsudik, and E. Uzun,“ANDaNA: Anonymous Named Data Networking Application,”CoRR, vol. abs/1112.2205,2011.

[22] M. Xie, I. Widjaja, and H. Wang, “Enhancing cache robustness for content-centric networking,”Proc. INFOCOM, 2012 Proceedings IEEE,2012, pp. 2426–2434.

[23] M. Conti, P. Gasti, and M. Teoli, “A lightweight mechanism for detection of cache pollution attacks in named data networking,”Computer Networks, vol. 57, no. 16, pp. 3178 – 3191, 2013.

[24] D. Smetters and V. Jacobson. (2009) “Securing network content”. [Online]. Available: https://www.parc.com/content/attachments/securing-networkcontent-tr.pdf

[25] I. Clarke, O. Sandberg, B. Wiley, and T. W. Hong,“Freenet: A Distributed Anonymous Information Storage and Retrieval System”, 2000.

[26] J. Kubiatowicz, D. Bindel, Y. Chen, S. E. Czerwinski, P. R. Eaton, D. Geels, R. Gummadi, S. C. Rhea,H. Weatherspoon, W. Weimer, C. Wells, and B.Y. Zhao, “OceanStore: an architecture for global-scale persistent storage,” Sigplan Notices,vol. 35, pp. 190–201, 2000.

[27] B. C. Popescu, M. V. Steen, B. Crispo, A. S. Tanenbaum, J. Sacha, and I. Kuz, “Securely Replicated Web Documents,”Proc. International Parallel and Distributed Processing Symposium/International Parallel Processing Symposium,2005.

[28] A. Detti, A. Caponi, G. Tropea, G. Bianchi, and N.Blefari-Melazzi, “On the interplay among naming, content validity and caching in information centric networks,”Proc. Global Communications Conference (GLOBECOM), 2013 IEEE, Dec 2013,pp. 2108–2113.

[29] G. Bianchi, A. Detti, A. Caponi, and N. Blefari Melazzi, “Check before storing: What is the performance price of content integrity verification in lru caching?”SIGCOMM Comput. Commun.Rev., vol. 43, no. 3, pp. 59–67, Jul. 2013.

[30] Q. Li, X. Zhang, Q. Zheng, R. Sandhu, and X. Fu,“Live: Lightweight integrity verification and content access control for named data networking,”Information Forensics and Security, IEEE Transac-tions on,vol. 10, no. 2, pp. 308–320, Feb 2015.