Name-Based Content Routing in Information Centric Networks Using Distance Information

Size: px

Start display at page:

Download "Name-Based Content Routing in Information Centric Networks Using Distance Information"

Roy Gregory
6 years ago
Views:

1 Name-Based Content Routing in Information Centric Networks Using Distance Information J.J. Garcia-Luna-Aceves Palo Alto Research Center UC Santa Cruz

q Hot- potato heuris4c rou4ng doctrine : Primordial distance- vector rou9ng.

2 Origins of Routing for Packet Switching Original documents: Paul Baran s On Distributed Communica9ons Series, RAND Corpora9on, q Packet headers, store- and- forward opera9on, sta9s9cal mul9plexing of links. q Hot- potato heuris4c rou4ng doctrine : Primordial distance- vector rou9ng. q Addresses for sender and des9na9on of packet. q Rou9ng of packets is based on des9na9on- based rou9ng tables using node addresses. q Des4na4ons are routers that originate rou4ng updates.

k) i l k p k D j i h2 = l(h i, h i+1 ) hop h i P j i.

3 State of The Art in Shortest-Path Routing Algorithms q q Problem: Compute the path of minimum length from each router to each destination node in the network G(N, E) is the network of N nodes and E links i P j N i q i (i, k) i l k p k D j i h2 = l(h i, h i+1 ) hop h i P j i. Each destination node has a single instance in the graph hx j Distances: Node starts by sending distance to itself Link-state: Node shares the state of adjacent links

4 The Brave New World of Information Centric Networks (ICN) q Many papers have stated that routing in ICNs is inherently different than routing in the old IP Internet. Content objects are copied opportunistically in the network. Multiple instances of the same destination. q However, existing proposals for routing in ICNs assume routing algorithms designed for single-instance destinations.

5 Remembering Multi-Homing and John McQuillan q McQuillan worked on the old and the new ARPANET rou9ng protocols: McQuillan J. M., et. al., The New Rou9ng Algorithm for the ARPANET, IEEE Trans. Comm q He and others also studied mul9- homed des9na9ons: J.M. McQuillan, Enhanced Message Addressing Capabili9es for Computer Networks, Proc. IEEE, Vol. 66, No. 11, Nov q Same rou4ng algorithms for single- instance des4na4ons. q A directory is used to map iden4fier of group to iden4fier of each instance.

6 How Can We Route to Multiple Instances of The Same Thing with Today s Routing Algorithms? q Source does not know all instances q Source knows all instances q Something in between

7 How Can We Route to Multiple Instances of The Same Thing with Today s Routing Algorithms? q Source does not know all instances Flood the network towards instances and prune as needed q Source knows all instances Have each node know the topology and the location of each instance. Instances flood the network. Source can compute source trees to instances q Something in between Designate a representative node between sources and destination instances Compute a routing tree rooted at the representative

8 Prior Results in ICN Routing q No information: Intanagonwiwat et al., [Directed Diffusion (Trans. Networking 2002)]; many DTN routing schemes. q All information: Mahmudul-Hoque et al., [NLSR (ACM ICN 13)]. DHT approaches need an underlay to build the DHT. q Representative: Carzaniga, Rosenblum, and Wolf [Content-Based Networking (Infocom 04)].

9 Distance-based Content Routing (DCR): Basic Approach q Establish a lexicographic ordering of distances to instances of destination. q The name of a router speaking for a destination instance (called anchor) is an attribute that must be used in the ordering. q Routers choose what information to share with their peers to preserve ordering (e.g., the best distance to any instance of content ). q Lexicographic ordering among all instances determines which instance can be the root of a DAG spanning all instances.

10 DCR Routing to nearest instances of destination: q N i = set of neighbors of node i q S i = set of neighbors of node i that are closer to des9na9on (next hops). q A i = set of known anchors of des9na9on at node i. q An update from i about a des9na9on states [d i, a i, sn i ]: distance (d i ), an anchor (a i ), a seq. number created by the anchor (sn i ). q d i = Min{d i k +li k k Si } d i k = distance reported by k; li k = cost of link to k

11 DCR Routing to nearest instances of destination: q Successor- Set Ordering Condi4on (SOC) used by node i to select a next hop k for des4na4on: ü Node k reports up- to- date informa9on ( m A i (a i k m si k sn(m) ) ü If node i has a finite distance: Node k is closer (d i k < di ) or is at the same distance and has a smaller name (d i k = di k < i ) ü If node i does not have a finite distance: Node k offers the smallest distance or has smallest name among neighbors offering the same smallest distance ( q N i {k}(d i k < di q ( di k = di q k < i ) )

12 s t (o, 1) l o (o, 0) r (o, 1) (u, 1) k p (u, 3) (d, 3) (o, 3) c u DCR (u, 0) (d, 2) e g w (u, 1) (d, 3) b (d, 2) h i (d, 3) (o, 4) a m (o, 3) (d, 4) v (d, 2) y (d, 3) (o, 4) d f (d, 0) z x (d, 2) Three instances (at d, o, and u) Lexicographic ordering based on: hop count to instance, ID of instance s anchor, and sequence number from anchor. Route to nearest instance w/o knowing all.

13 DCR Routing to some or all instances of destination: 1. Select the anchor with the smallest iden9fier as the root anchor (ra) of the des9na9on. 2. Select neighbor k as next hop to root anchor if node k: ü Reports up- to- date anchor informa9on ü Has the smallest sequence number for ra or is closest to ra among all neighbors with a valid sequence number for ra. 3. Ensure that root anchor informa9on is known by all anchors of des9na9on: ü Node i sends update about ra to neighbor q if i is an anchor or q is the lexicographically smallest way to reach another anchor: a i q rai j rai q > rai k N i {i} [ a i k = i (a i k rai q (di q < di k [di q = di k q < k ] ) ) ]

14 s t (o, 1) (d, 5) l o (o, 0) (d, 6) r (o, 1) (u, 1) (d, 4) k p (u, 3) (d, 4) (d, 3) (o, 3) c DCR u (u, 0) (d, 4) (d, 2) One instance is best and hence is known by other instances (d < o < u). Updates about d are sent only along the best paths between d and o and between d and u. e (u, 1) (d, 3) g w b h (d, 2) i (d, 3) (o, 4) a m (o, 3) (d, 4) v (d, 2) y (d, 3) (o, 4) d f (d, 0) x z (d, 2)

15 The MIDST of a Destination: Multiple Instance Destination Spanning Tree s t (o, 1) (d, 5) l o (o, 0) (d, 6) r (o, 1) (u, 1) (d, 4) k p (u, 3) (d, 4) (d, 3) (o, 3) c u (u, 0) (d, 4) (d, 2) Only Anchors send joins towards root anchor to form MIDST. Messages sent towards nearest instance, broadcast over MIDST to reach all or some instances. e g w (u, 1) (d, 3) b h (d, 2) i (d, 3) (o, 4) a m (o, 3) (d, 4) v (d, 2) y (d, 3) (o, 4) d f (d, 0) z x (d, 2)

16 Why Is This Important? APain orders of magnitude improvement in the control plane CC LSR = O(ERC + lne) SC LSR = O(RC + E) CC DHT = O(dRC + lne) CC DHT = O(C + E) CC DCR = O(EC); SC DCR = O(C) order communication complexity (CC) C: # des4na4ons R: # replicas E: # edges; N: # nodes d: diameter; l: node degree ~ N R = O(N) R = O(1) R = O(N) R = O(1) R = O(N) N = 500 l and d are O(log N) R = O(1) #contentobjects(c) DVR LSR DHT DCR

17 Why Is This Important? Loop- Free FIBs Loop- Prone FIBs Interest looping More efficient data plane! Looping with mul9- path rou9ng leads to searching for paths in O(N) steps w/o guarantees (DFS) and Interest 9mers of order O(N) to avoid loops not being detected! Loop- free FIBs allow searching for paths in O(x) steps and Interest 9mers of order O(d), with x < d << N

18 Summary q Just a start! Many more results are needed: DCR simula9ons in ns3 soon Real implementa9on aber that Mul9point applica9ons (e.g., mul9cast) QoS, policies, other types of termina9on detec9on. q Change the way we think about des9na9ons for all types of rou9ng; assume any des9na9on can have mul9ple instances. q Integrate unicast, mul9cast, and anycast rou9ng. q Need solu9ons that scale be6er than O(C). It can be done using distances.

19 THANK YOU! ANY QUESTIONS?

UC Santa Cruz UC Santa Cruz Previously Published Works

UC Santa Cruz UC Santa Cruz Previously Published Works Title Efficient Multi-Source Multicasting in Information Centric Networks Permalink https://escholarship.org/uc/item/7mx2w441 Author Garcia-Luna-Aceves,