UCLA ENGINEERING Computer Science Fragmented Data Routing Based on Exponentially Distributed Contacts in Delay Tolerant Networks Tuan Le, Qi Zhao, Mario Gerla Computer Science, UCLA {tuanle, qi.zhao, gerla}@cs.ucla.edu
UCLA ENGINEERING Computer Science Outline • Background • Motivation • Proposals • Protocol Design • Evaluation • Conclusion Slide 2 / 16 2/20/19
UCLA ENGINEERING Computer Science Background Delay-Tolerant Mobile Ad-Hoc Networks • Sparsely connected • End-to-end paths are rarely available due to node mobility Slide 3 / 16 2/20/19
UCLA ENGINEERING Computer Science Motivation • Pedestrians with hand-held devices communicate via Bluetooth • High-speed vehicles communicate via WiFi (802.11g) • Both cases have short contact duration: several seconds Slide 4 / 16 2/20/19
UCLA ENGINEERING Computer Science Motivation • Existing works assume unfragmented data • Large data not fit short contact duration • Abort entire transmission • Useless retransmission of entire data 20 MB 5 sec contact duration Slide 5 / 16 2/20/19
UCLA ENGINEERING Computer Science Proposals • Single-copy data fragmentation • Break data into small chunks • Transmitted at various contacts • Compute direct/indirect delivery probability to successfully deliver all chunks via multiple contacts C1 C2 C3 Fragments Slide 6 / 16 2/20/19
UCLA ENGINEERING Computer Science One-Hop Delivery Probability • Probability message i is successfully delivered from s to d during the n th meeting is the joint probability of 3 events • Message i does not expire before the n th meeting • d does not receive all parts of message i during the first n-1 meetings • d receives the remaining parts of message i during the n th meeting Slide 7 / 16 2/20/19
UCLA ENGINEERING Computer Science One-Hop Delivery Probability • Joint probability P i (n) n n − 1 n ⇣ ⌘ ⇣ ⌘ ⇣ ⌘ X X X X s,d Z s,d Z s,d P i ( n ) = P 0 < 0 ≤ < R i · P < W i · P ≥ W i k k k k =1 k =1 k =1 • One-hop (direct) delivery probability ∞ X P i = P i ( n ) n =1 • and – kth inter-contact time and contact duration time between s and d Y s,d X s,d k k R i – remaining lifetime of message i • Y s,d • = x B – amount of data sent from s to d during kth contact Z s,d k k • B – communication bandwidth between two nodes • W i – size of message i Slide 8 / 16 2/20/19
UCLA ENGINEERING Computer Science Two-Hop Delivery Probability • Probability message i is successfully delivered from s to v during their n th meeting and from v to d during their m th meeting n m ⇣ ⌘ X X X v,d X s,v P i ( n, m ) = P 0 < + < R i a b a =1 b =1 ⇣ n − 1 n ⌘ ⇣ ⌘ X X Z s,v Z s,v · P < W i · P ≥ W i a a a =1 a =1 ⇣ m − 1 ⇣ m ⌘ ⌘ X X Z v,d Z v,d · P < W i · P ≥ W i b b b =1 b =1 • Two-hop (indirect) delivery probability ∞ ∞ X X P i = P i ( n, m ) n =1 m =1 Slide 9 / 16 2/20/19
UCLA ENGINEERING Computer Science Routing Strategy • S computes one-hop delivery probability P s to d and two-hop delivery probability P vi to d via each neighbor v i • P v = max(P v1 , P v2 , …, P vm ) • If P v > P s and no chunk of message i resides at another node, s forwards all parts of message i to v that fit contact duration d P s Current contact Direct delivery probability P v > P s Indirect delivery probability v s P v Slide 10 / 16 2/20/19
UCLA ENGINEERING Computer Science Evaluation • ONE simulator 1.5.1 • Cabspotting trace: 536 taxis collected over 30 days in San Francisco Bay Area • Each node has 5 source messages of same size intended for random destinations • Messages have a homogenous TTL value • Nodes have infinite buffer capacity Slide 11 / 16 2/20/19
UCLA ENGINEERING Computer Science Evaluation Evaluate the following schemes: • Epidemic routing: flood messages (multi-copy, unfragmented) • PROPHET: select relay with higher delivery probability to the destination (single-copy, unfragmented) • BubbleRap: bubble up messages to node with high global ranking. Once reach community, bubble down using local ranking (single-copy, unfragmented) • MEED: select relay with lower minimum expected delay (single-copy, unfragmented) • Fragmented data routing (FDR) : select relay with max(one-hop, two- hop delivery probability) (single-copy, fragmented) Slide 12 / 16 2/20/19
UCLA ENGINEERING Computer Science Performance comparison 10KB 1 2.5 Epidemic Epidemic FDR FDR 0.8 2 Average delay (days) PROPHET PROPHET MEED MEED Delivery ratio BubbleRap BubbleRap 0.6 1.5 0.4 1 0.2 0.5 0 0 0.5 1 2 4 6 8 10 12 0.5 1 2 4 6 8 10 12 TTL (days) TTL (days) Average delay Delivery ratio Performance comparison with messages of 10 KB Slide 13 / 16 2/20/19
UCLA ENGINEERING Computer Science Performance comparison 20MB 0.8 3.5 Epidemic Epidemic 3 FDR FDR Average delay (days) PROPHET PROPHET 0.6 2.5 MEED MEED Delivery ratio BubbleRap BubbleRap 2 0.4 1.5 1 0.2 0.5 0 0 0.5 1 2 4 6 8 10 12 0.5 1 2 4 6 8 10 12 TTL (days) TTL (days) Average delay Delivery ratio Performance comparison with messages of 20 MB Slide 14 / 16 2/20/19
UCLA ENGINEERING Computer Science FDR Performance 0.9 2 0.8 1.8 1 2.5 0.7 1.6 Average delay (days) 0.8 2 Delivery ratio 0.6 1.4 0.6 1.5 0.5 1.2 0.4 1 0.4 1 0.2 0.5 0.3 0.8 0 0 20 20 0.2 0.6 15 15 15 15 10 10 10 10 0.1 0.4 5 5 5 5 Message size (MB) Message size (MB) 0 0 0 0 TTL (days) TTL (days) Average delay Delivery ratio Performance of FDR with varied message sizes from 10KB to 20MB Slide 15 / 16 2/20/19
UCLA ENGINEERING Computer Science Conclusion • Forwarding decision using message fragmentation is aware of contact duration and message size • Consider both direct and indirect delivery probability • FDR improves delivery rate and delay by up to 37% and 35%, respectively Slide 16 / 16 2/20/19
UCLA ENGINEERING Computer Science Thanks! Q & A
Recommend
More recommend