Professor Hongbin Luo Beijing Jiaotong University

1. CoLoR: An Information-Centric Future Internet Architecture for Innovation Professor HongbinLuo Beijing JiaotongUniversity August 11, 2013

2. Roadmap Background 1 Design Goals 2 Design details 3 Benefits of CoLoR 4 Feasibility Analysis 5 Conclusions 6 Beijing Jiaotong University

3. 1. Background The current Internet has made great success in the past years. However, it also faces many serious issues, such as: Scalability Mobility Security The DFZ routing table size grows rapidly, the internet faces serious routing scalability issues. No inherit security mechanism. There are too many security threats. The Internet Cannot efficiently support mobility. There is a growing consensus that these drawbacks cannot be remedied by incremental changes, and a clean-slate design of the Internet architecture is desired. Beijing Jiaotong University

4. 1. Background Because of these drawbacks, many countries have founded many projects to investigate future Internet architecture in recent years. Beijing Jiaotong University

5. 1. Background In recent years, China has also founded many projects to investigate future Internet architecture, under its well-known 973 program. • The project lasted five years , from Jan. 2007 to Aug. 2011. In the final examination made by the Ministry of Science and Technology of China in Nov. 2011, the project got the top level score “excellent”. In Chinese: “优秀”； In English: “Excellent”. Beijing Jiaotong University

6. 1. Background China has also founded many projects to investigate future Internet architecture, under its well-known 973 program. The project was renewed in 2012, from Jan. 2013 to Aug. 2017. Beijing Jiaotong University

7. 2. Design Goals We aim at designing a future Internet architecture that satisfies the following design goals: Being information centric: While the current Internet was designed centered on hosts, its current majority usage is data retrieval. Accordingly, there is an increasing consensus that the future Internet should be information-centric. That is, content should be addressed independent of its hosted location. Efficient support for mobility:With the rapid increase in the number of mobile devices, the future Internet architecture should efficiently support mobility. Until March 2013, the number of mobile users in China is 1.146 billion; 71.34% of them have access to the Internet. Cisco predicted that: “traffic from wireless devices will exceeds traffic wired devices by 2014. Beijing Jiaotong University

8. 2. Design Goals Efficient support for multi-homing: In multi-homing, a host (or network) is simultaneously attached to multiple networks. While the current Internet is cumbersome to support multi-homing since it causes serious routing scalability issue, the future internet architecture is expected to efficient support multi-homing. Encouraging innovation: The future internet architecture should allow each network to use its preferred network architecture and routing mechanism so that different network technologies can be simultaneously deployed and contest, thus encouraging innovation. Enhanced security: The current Internet employs a default-on model and any host is able to send packets to a remote host, which makes the current Internet fragile to distributed denial-of-service attacks. Therefore, the future Internet should offer receivers the ability to control incoming traffic, especially to refuse unwanted traffic. Beijing Jiaotong University

9. 2. Design Goals Enhanced scalability: The future Internet should provide better routing scalability over the current Internet. The routing table size should be significantly less than that in the current Internet. Ease of traffic matrix estimation: It is difficult to estimate traffic matrices in the current Internet. However, since traffic matrices are critical inputs to many aspects of network management such as traffic engineering and network provisioning, the future Internet should makes it easy to precisely estimate traffic matrices in real time. Deployability: Although we aim at a clean-slate design, the future Internet architecture should be deployed without incurring significant cost. Beijing Jiaotong University

10. 3. Design details Basic ideas: Using four namespaces: Service identifiers (SIDs): used to name contents. They are flat, self-certifying. Node identifiers (NIDs): used to identify the identity of network nodes. They are flat, self-certifying and 128 bits long. Intra-domain routing locators: used for intra-domain routing. Every domain can choose its preferred intra-domain routing architecture and routing locators. Path identifiers (PIDs): used for inter-domain routing. Two domains can negotiate a set of PIDs, as long as the PIDs are unique in each domain. PIDs are not advertised throughout the Internet, but are local to the two domains. Beijing Jiaotong University

11. 3. Design details Basic ideas: Using name-based routing for service location. Inter-domain routing for data packet forwarding is determined during the service location process. Intra-domain routing may or may not be determined during the service location process. We leave this for domains’ local policy. End-to-end data packet forwarding is based on loose source routing. While some of the ideas are borrowed from existing literature, we believe that: one can see further by standing on the shoulders of giants. Beijing Jiaotong University

12. 3. Design details Network topology As the current Internet, CoLoR assumes that the future Internet will still centered around domains. Domains have the AS-level provider/ customer/peer relationship. Beijing Jiaotong University

13. 3. Design details Intra-domain routing Domain 3 uses MPLS for intra-domain routing. Domain 1 uses IPv6 for intra-domain routing. Domain 4 uses OpenFlow for intra-domain routing. A domain can freely choose its preferred intra-domain routing architecture, without considering other domains. Beijing Jiaotong University

14. 3. Design details Inter-domain routing Inter-domain routing relies on paths negotiated by two neighbor domains. P7 R4 D2 P5 P1 R2 P6 D1 D6 P5 R5 P6 P7 P4 Nodes in a domain maintains the end point of every path that connects the domain to a neighboring domain. Beijing Jiaotong University

15. 3. Design details Service registration RM6 Every domain has a logical resource manager. P5 P1 The service registration process is similar to that in DONA. RM5 RM3 Content sources register SIDs to their local RMs, which registers the SID to their peers or provider RMs. P6 P7 RM1 P4 RM2 RM4 Beijing Jiaotong University

16. 3. Design details Service location and inter-domain routing RM6 Users send requests to their local RMs when they want a content represented by an SID. (iii) P5 P1 (iv) RM5 RM3 RMs forward requests to either the closest copy of the content, or their provider RMs. P6 (v) P7 (ii) RM1 P4 RM2 (i) RM4 (vi) Beijing Jiaotong University

17. 3. Design details Service location and routing RM6 (iii) P5 P1 (iv) RM5 RM3 Every time a RM forwards a request to a neighboring RM, it appends the path between the two domains onto the request. P6 (v) P7 (ii) RM1 P4 RM2 (i) RM4 (vi) Beijing Jiaotong University

18. 3. Design details Packet forwarding (d) Intra-domain packet forwarding is based on the routing mechanism of each domain. (b) (a) (c) Inter-domain packet forwarding is based on PIDs that are determined during the service location process. Every time a border router receives an incoming packet, it strips out the outer most PID in the packet header. Beijing Jiaotong University

19. 3. Design details Content caching A domain can freely choose whether or not to cache a content, based on its local policy and network status. In addition, caching may or may not be en-route. For example, domain D5 can cache a content at R9 instead of R10, though R10 is en-route. For example, if D5 uses LIPSIN for local routing, we may compute a zFilter including the red links. Beijing Jiaotong University

20. 3. Design details Content caching RM6 The node caching the content should register the cached content to its local RM. ? When the local RM receives requests for the content, it forwards the requests to the node caching the content, thus improving resource and energy efficiency. register RM5 We leave it as local policy that whether or not a RM registers a cached content to its provider or peer RMs. requests Beijing Jiaotong University

21. 4. Benefits of CoLoR Inter-domain traffic engineering Inter-domain traffic engineering is easy to implement. RM6 When the RM5 in D5 forwards requests to the RM6 in D6, RM5 can choose the preferred path (e.g., P1 or P2) to carry packets corresponding to the request, based on domain D5’s local policy. P1 (iii) P2 RM5 Inter-domain traffic engineering could be done at fine granularity (e.g., per-request level), which makes it easier to achieve better load balancing. Beijing Jiaotong University

22. 4. Benefits of CoLoR Intra-domain traffic engineering CoLoR makes it easier to achieve fine-grained intra-domain traffic engineering. RM6 For example, when RM5 forwards request A and B to RM6, it assigns intra-domain path P and P’ for transmitting data packets corresponding to requests A and B, respectively. Requests A and B RM5 P’ P Note that intra-domain traffic engineering could be implemented at per-request level, or some coarse granularity, depending the domain’s local policy. Beijing Jiaotong University

23. 4. Benefits of CoLoR Multi-homing Add an inter-domain path 10.10.3.0/23 10.10.3.0/24 Add an inter-domain path 10.10.3.0/23 In the current Internet, Multi-homing often accompanies with prefix deaggregation, affecting the DFZ routing table size. By contrast, multi-homing in CoLoR only increases one or several local inter-domain paths and does not affect other domains. Beijing Jiaotong University

24. 4. Benefits of CoLoR Precise Estimation of Traffic Matrices An ingress border router (IBR) knows the egress border router (EBR) of a data packet when the IBR forwards the data packet. P7 R4 D2 R2 P6 D1 D6 P5 R5 packet To estimate the traffic matrix from the IBR to the EBR, the IBR only needs to count the number of packets (or bytes) when it forwards packets to the EBR, which could be implemented in real time at line rate. Beijing Jiaotong University

25. 4. Benefits of CoLoR Multicast Support RM6 The content caching mechanism and the content registration primitive make CoLoR efficient in supporting multicast. register RM5 requests Beijing Jiaotong University

26. 4. Benefits of CoLoR Encouraging Innovation Every domain is free to choose its preferred network architecture. This encourages domains to adopt novel networking technologies if they feel that the new solution is significantly better than the one in use, without caring whether the new one is almost perfect. AS D3 uses MPLS. AS D5 uses OPENFLOW AS D2 uses IPv6 AS D4uses IPv4 Beijing Jiaotong University

27. 4. Benefits of CoLoR Mobility Support RM6 Inter-domain movement: with the content caching and content registration primitives, a mobile host can re-request a content, which will be routed to a nearby copy of the content. register RM5 Intra-domain movement: addressed by using identifier/locator split. requests movement Beijing Jiaotong University

28. 4. Benefits of CoLoR Middleboxes RM6 When the RM in a domain forwards GET messages, it may also assign the intra-domain path for the flow corresponding to the GET message. Accordingly, the RM can direct different flows to different middleboxes, based on its local policy, which makes CoLoR efficient in support middleboxes. RM5 P’ P Beijing Jiaotong University

29. 4. Benefits of CoLoR Efficient support for SDN In existing Internet architectures with SDN, the flow entry is setup when the border router receives the first data packet of a flow. requests Forwarding rules Data packets There is a flow setup delay of about 10 ms at an SDN domain, which will add delay to the flow. Beijing Jiaotong University

30. 4. Benefits of CoLoR Efficient support for SDN In CoLoR, when a RM forwards a GET message to a next hop domain, the border router in the same domain with the RM will receive the corresponding data packets after a certain period. t The controller can use this period to setup flow entries onto the switches for the flow, thus reducing (avoiding) the flow setup delay. How Long is it? Beijing Jiaotong University

31. 4. Benefits of CoLoR Efficient support for SDN Mean = 24.5 ms The cumulative probability density function of t Beijing Jiaotong University

32. 4. Benefits of CoLoR Routing scalability The inter-domain routing tables size depends on the number of neighboring domains and the number of paths between them. Assuming that the number of paths between two neighboring domains is 10, the maximum inter-domain routing table size is less than 40,000 ( > 3777 * 10). The intra-domain routing table size depends on the routing mechanism of the domain and should be within control of the domain. Beijing Jiaotong University

33. 4. Benefits of CoLoR Security 1. Receiver controls incoming traffic by sending out GET messages. 2. Packets from a source node can be sent to a destination node only if the inter-domain path identifiers carried in the packet header are correct. 3. However, path identifiers are local to neighboring domains and it is difficult to guess the PID between two neighboring domains. If PIDs are 32-bit long, the probability to correctly guess a PID between two domains is ½^{32}. Beijing Jiaotong University

34. 4. Benefits of CoLoR Security 4. A node receiving a request can send plenty of packets to a destination node. But this could be dealt with by using mechanisms such as TVA [2]. [2] X. Yang, D. Wetherall, T. Anderson, “TVA: a DoS-limiting network architecture,” IEEE/ACM Transactions on Networking, vol. 16, no. 6, Dec. 2008, pp. 1267 – 1280. In summary, CoLoR is significantly more secure than the current Internet. [3] reports a preliminary analysis on CoLoR’s security. [3] Z. Chen, H. Luo, J. Cui, M. Jin, “Security analysis of a future Internet architecture,” in Proc. 8th Workshop on Secure Network Protocols (NPSec’13), Oct. 2013, Gottingen, Germany. Beijing Jiaotong University

35. 4. Benefits of CoLoR Deployment 1. CoLoR may not be incrementally deployable as end hosts need to be updated in order to send registration and GET messages. 2. Existing networks are only required to update their border routers and build a RM in order to accommodate CoLoR, thus significantly reducing the cost in deploying CoLoR. Beijing Jiaotong University

36. 5. Feasibility Analysis Prototype 30% 1000 GET messages per second 70% 50% 50% We have implemented CoLoR’s basic features in the prototype shown above. The implementation demonstrates that CoLoR’s is feasible. Beijing Jiaotong University

37. 5. Feasibility Analysis Click modules of RMs Click modules of RMs Beijing Jiaotong University

38. 5. Feasibility Analysis Click modules of the client and the servers Click modules of the client and the servers Beijing Jiaotong University

39. 5. Feasibility Analysis Click modules of routers Click modules of routers Beijing Jiaotong University

40. 5. Feasibility Analysis The processing delay of GET messages The delay of processing GET messages at RM1 Beijing Jiaotong University

41. 5. Feasibility Analysis Traffic Matrices R3 – R1 R6 – R4 R2 – R1 The estimated traffic matrices Beijing Jiaotong University

42. 5. Feasibility Analysis Load balancing PID1, 70% PID2, 30% The effect of load balancing Beijing Jiaotong University

43. 5. Feasibility Analysis Large scale deployment The RMs in tier-1 network needs to deal with the GET messages, which may limit the performance of CoLoR. 1. Resource handlers (RHs) in DONA are capable of processing REGISTER messages and FIND messages if DONA is deployed at the scale of the current Internet. Since RM in CoLoR is similar to RHs in DONA, RMs in CoLoR is also able to deal with GET messages. 2. Dannewitz et al. [1] pointed out that it is possible to design a distributed hash table based name resolution system for flat SIDs, with an average resolution delay below 100 ms. Therefore, CoLoR is feasible at the scale of the current Internet. [1] C. Dannewitz, M. D’Ambrosio, V. Vercellone, “Hierarchical DHT-based name resolution for information-centric networks,” Computer Communications, vol. 36, no. 7, April 2013, pp. 736 – 749. Beijing Jiaotong University

44. 6. Conclusions We have proposed CoLoR that couples service location with inter-domain routing while decoupling from forwarding. Our implementation and analysis demonstrates that CoLoR has many benefits and is feasible. Beijing Jiaotong University

45. Thanks! Beijing Jiaotong University