Enhancing Compact Routing in CCN with Prefix Embedding and - PowerPoint PPT Presentation

Enhancing Compact Routing in CCN with Prefix Embedding and Topology-Aware Hashing Stefanie Roos 1 , Liang Wang 2 , Thorsten Strufe 1 , Jussi Kangasharju 2 1 TU Dresden, firstname.lastname@tu-dresden.de 2 University of Helsinki, firstname.lastname@helsinki.fi

Motivation • xCNs: new Internet architecture based on addressing content rather than locations (hosts) • Goal: Improve content delievery • Various challenges with regard to routing and content addressing • Source mobility: One particular hard challenge Stefanie Roos, TU Dresden Compact Routing 2

Greedy Embeddings • Assign coordinates to fixed topology such that greedy routing works • Handling mobility in CCNx [1] Source s registers at closest host h • h forwards packets to s • If s moves, only h updates its information • • Benefits Small routing table • Capability of hand- • ling simultaneous handoff Improved handoff • delay and latency [1] Wang, L., Waltari, O., & Kangasharju, J. (2013). Mobiccn: Mobility support with greedy routing 9/26/14 Compact Routing 3

Prior Work: Limitations • Limitations: • Embedding Hyperbolic space, combined Root’s ¡ ¡ with an naive content addressing area ¡ algorithm (SHA1) • Traffic and storage load is highly imbalanced Image ¡from ¡R. ¡Kleinberg: ¡ • Severe scalability issue when network Geographic ¡Rou:ng ¡using ¡ Hyperbolic ¡Space, ¡Infocom ¡2007 ¡ ¡ becomes bigger ¡ Our Contributions: 1. Changing the embedding algorithm 2. Changing the content key generation 9/26/14 Compact Routing 4

Prefix Embedding • Prefix Embedding:Isometry of spanning tree 1. Root has empty vector as ID 2. Node with ID id enumerates children 3. i-th child receives ID id||i 4. Distance between nodes in tree dist(s,t,)=|s|+|t|-2commonprefixlength(s,t) 9/26/14 Compact Routing 5

Prefix Embedding: Extensions • Virtual binary trees for bit strings as IDs • Routing modification for virtual trees: Forward to parent if not respon- sible but no closer neighbor • Content is stored on node closest to its key • Content keys are longer than IDs -> all content stored on leaves and nodes with only one child 9/26/14 Compact Routing 6

Prefix-S Embedding • Store content on internal nodes • Use two types of IDs: routing ID and storage ID • Routing IDs are IDs received from parent • Internal nodes with d children generated d +1 IDs, choose first one as their storage ID • Leaf nodes use routing id for storage • Greedy routing with slight modification is guaranteed to succeed 9/26/14 Compact Routing 7

Topology-Aware Keys • Nodes on higher levels of tree responsible for more files • Integrate topology in keys of content • Consider hash function h • Cpl = common prefix length • For content f, h_i (f XOR i) 0 ¡ 1 ¡ • i-th bit of content key b 1 =0 ¡with ¡p=3/4 ¡ => Uniform load 9/26/14 Compact Routing 8

Evaluation Evaluation (Static Simulation): 1. Generate random contents 2. Embed AS topology 3. Compute key of content and store 4. Execute queries for content 5. Metrics: Fraction of content pieces per node • Fraction of queries forwarded per node • Routing hops • 9/26/14 Compact Routing 9

Storage Load 1 Kleinberg Embedding PREFIX embedding 0.9 PREFIX embedding, TAKs PREFIX-S embedding 0.8 PREFIX-S embedding, TAKs 0.7 Fraction of Keys 0.6 • Hyperbolic: ¡more ¡than ¡95 ¡% ¡ ¡ 0.5 ¡ ¡ ¡ ¡ ¡ ¡content ¡on ¡1 ¡node ¡ 0.4 • Prefix ¡Embedding: ¡s:ll ¡ ¡ 0.3 0.2 ¡ ¡ ¡ ¡ ¡ ¡unbalanced ¡ 0.1 • Topology-­‑aware ¡keys: ¡uniform ¡ 0 1 2 3 4 5 ¡ ¡ ¡ ¡ ¡ ¡load ¡(aka ¡close ¡to ¡straight ¡line) ¡ ¡ Rank 1.1 1 0.9 Cumulative Fraction of Keys 0.8 0.7 0.6 0.5 0.4 0.3 Kleinberg Embedding PREFIX embedding 0.2 PREFIX Embedding, TAKs PREFIX-S embedding 0.1 PREFIX-S embedding, TAKs 0 0 100 200 300 400 500 600 700 Rank 9/26/14 Compact Routing 10

Forwarding load + routing length 1 Kleinberg Embedding • Hyperbolic: ¡close ¡to ¡98 ¡% ¡of ¡queries ¡pass ¡ PREFIX embedding 0.9 PREFIX embedding, TAKs root ¡ PREFIX-S embedding 0.8 PREFIX-S embedding, TAKs • Prefix ¡Embedding: ¡s:ll ¡around ¡70 ¡% ¡ 0.7 Fraction of Keys • Topology-­‑aware ¡keys: ¡not ¡uniform ¡but ¡ 0.6 0.5 beWer ¡balanced ¡ 0.4 0.3 ¡ 0.2 ¡ 0.1 1 2 3 4 5 Rank 1 Kleinberg Embedding • Rou:ng ¡length ¡is ¡increased ¡from ¡roughly ¡ PREFIX embedding PREFIX Embedding, TAKs 3-­‑4 ¡to ¡4-­‑5 ¡hops ¡by ¡using ¡topology-­‑aware ¡ PREFIX-S embedding PREFIX-S embedding, TAKs 0.1 keys ¡ Fraction of Keys ¡ 0.01 ¡ 0.001 0.0001 1 10 100 1000 Rank 9/26/14 Compact Routing 11

Conclusion • Problem: Source mobility in xCNs • Proposed Solution: Embeddings • Improved the load balancing by 1. Modifying embedding 2. Topology-aware keys Can now prevent overload, single point-of- • failure Future work: Evaluation in testbed to see • the effect on actual congestion 9/26/14 Compact Routing 12