Download presentation
Presentation is loading. Please wait.
Published byBasil Richardson Modified over 8 years ago
1
Fault Tolerant Graph Structures Merav Parter ADGA 2015
2
General Theme of Network Design Design Logical Structure (on top of a given network) that possess desirable properties. Examples: Shortest-Path Trees Spanners Minimum Spanning Tree Clustered Representations: partitions, composition Distance Oracles Routing Schemes Informative Labeling Scheme ……
3
General Theme of Network Design
4
Limitation of Standard Structures Shortest-Path Tree (BFS) rooted at s. s v1v1 v2v2 v4v4 v3v3 v5v5 Not robust against edge and vertex faults.
5
Theory of Fault Tolerant Networks
6
Examples of FT-Structures Techniques for FT-Design Lower Bound Construction Outlines
7
Examples of FT-Structures FT-Connected Subgraph Replacement-Paths FT-BFS FT-Spanners FT-MST
8
FT-Connected Subgraph s t [Nardelli, Stege, Widmayer, Tech report, ‘97] [Chechik and Peleg. IEEEI 2010]
9
FT Notion of Shortest-Path: Replacement Path P(s,t) s t e P(s,t,e) P(s,t,e) : s-t shortest path in G\{e} [Malik, Mittal, Gupta 89’] [Hershberger, Suri ‘01] [Roditty, Zwick ‘05] [Gotthilf, Lewenstein, ’09] [Weimann, Yuster, ’11] [Vassilevska Williams, ’11] [Grandoni, Vassilevska Williams, ‘11] ….
10
Breadth First Search (BFS) Trees G H ss
11
FT-BFS Structures G H ss [P, Peleg, ESA’13]
12
FT-BFS Structures *P(s,t,e) : s-t shortest path in G\{e} [P, Peleg, ESA’13]
13
FT-k-Spanner 13 GH
14
FT-k-Spanner 14 G H [Chechik, Langberg, Peleg, Roddity, STOC’09]
15
FT-Spanners vs. FT-BFS trees FT-Spanners FT-BFS tree approximate exact FT-BFS’s easier FT-BFS’s easier FT-BFS’s harder FT-BFS’s harder
16
FT-Approximate BFS Structures [P, Peleg, SODA’14]
17
FT-Minimum Spanning Tree (MST) [Nardelli, Stege, Widmayer, Tech report, ‘97] [Chechik and Peleg. IEEEI 2010]
18
Examples of FT-Structures Techniques for FT-Design Lower Bound Construction Outlines
19
Techniques for FT-Design 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach. 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach.
20
Techniques for FT-Design 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach. 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach. Complexity Measure: Running time Size of FT-structure
21
The Challenge: The Naive Approach Size increases exponentially with #faults f
22
The Iterative Approach Compute(f+1)-disjoint “fault-free” solutions
23
The Iterative Approach Compute(f+1)-disjoint “fault-free” solutions
24
The Iterative Approach f+1)-disjoint Compute (f+1)-disjoint “fault-free” solutions. Examples: FT-connected subgraph FT-k-spanner [chechik at el. STOC09’]
25
The Iterative Approach Con: Doesn’t extend to vertex faults. Useful when it is sufficient to satisfy predicate on edges (instead of all pairs).
26
Techniques for FT-Design 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach
27
The Swap Edge Approach 5 10 6 14 11 8 13 17 4 12 9 Example: MST tree T
28
The Swap Edge Approach 5 10 6 14 11 8 13 17 4 12 9 Example: MST tree T
29
The Swap Edge Approach 5 10 6 14 11 8 13 17 4 12 9 Example: MST tree T
30
The Swap Edge Approach Example: MST tree T The FT-MST S’ contains MST and the n-1 swap edges 5 10 6 14 11 8 13 17 4 12 9
31
The Swap Edge Approach “Best” swap edge for minimizing: Diameter [Nardelli et al., ‘01] Distances (e.g., sum, average) [Flocchini, et al. ’05] Stretch (e.g., for tree Spanners) [Des et al., ‘08] “Reconnecting the structure in the best possible way” Main objective: fast computations
32
Pros: Useful for single edge fault + trees Con: Doesn’t extend to vertex fault. Size increases exponentially with the number of edge faults. The Swap Edge Approach
33
Techniques for FT-Design 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach. 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach.
34
The Sampling Approach: Handling Vertex Faults [Dinitz, Krauthgamer, PODC’11] Inspired by the color-coding technique of Alon, Yuster and Zwick. Randomly Randomly sample nodes to act as a fault set, Apply a generic alg’ on what remains.
35
35 The Sampling Approach: Handling Vertex Faults Example: vertex-FT-k-Spanner
36
The Sampling Approach: Handling Vertex Faults
37
The Sampling Approach: Handling Vertex Faults [Dinitz, Krauthgamer, PODC’11]
38
The Sampling Approach: Handling Vertex Faults [Dinitz, Krauthgamer, PODC’11]
39
Techniques for FT-Design 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach 1. The iterative approach 2. The swap edge approach 3. The sampling approach 4. The structural approach
40
The Structural Approach Assume single fault case. Correctness is immediate. Size analysis (i.e., showing sparseness) structure is based on graph structure.
41
The Structural Approach: FT-BFS The construction: [P, Peleg ESA’13]
42
FT-BFS Upper Bound Assume uniqueness of shortest-paths. Recall: P(s,t,e) is the s-t shortest path in G\{e} Based on analyzing structure of replacement-paths P(s,t,e)
43
Side Note: FT-BFS in Distributed Setting
44
Limitation of the Structural Approach Tedious. What about FT-BFS for more than one fault? complexity #faults 1 2 Exact Approx’
45
Dual Failure Replacement Paths P(s,t,F) : s-t shortest path in G\F, F={e,e’} One previous work on dual-failure RP s t e e’ P(s,t,F) Dual Failure Distance Oracle [Duan, Pettie, SODA’09]
46
Dual Failure FT-BFS
47
f-FT-BFS: State of Art # Faults Upper BoundLower Bound ?
48
Cost of Introducing Fault Tolerance Cost Connectivity f-Edge-FT-spanner f-FT-BFS FT-BFS FT-additive spanner
49
Examples of FT-Structures Techniques for FT-Design Lower Bound Construction Outlines
50
Lower Bound for FT-BFS Structures
51
Generalization to FT-BFS with multiple sources
52
Lower Bound for FT-BFS X Z s |Z| 6 8 [P, Peleg, ESA’13]
53
X Z s |Z| 6 8 Lower Bound for FT-BFS
54
Cl. : Every FT-BFS H must contain ALL the edges of the bipartite graph. By contradiction: Assume there exists an edge e i,j that is not in H. Consider the case where f i fails. X Z s xjxj zizi e i,j fifi Lower Bound for FT-BFS
55
The Construction X Z s fifi d(s,x j, H \{f i }) >d(s,x j, G\{f i }) Contradiction since H is an FT-BFS tree. xjxj 6 8 zizi vivi v i+1
56
Reducing the cost of FT Relaxing the predicate requirement. E.g.: approximate FT-BFS has O(n) edges. Strengthening the model Backup (redundancy) is only one FT-mechanism.
57
Quality vs. Quantity Tradeoff Quality Quantity Reinforcement (R) Backup (B) Product 3R 2R+3B 9B B<R
58
Reinforcement + Backup Subgraph s v1v1 v2v2 v4v4 v3v3 v5v5 FT subgraph H Low-cost, fault prone high-cost, fault resistant [P, Peleg, SPAA’15]
59
(b,r) FT-BFS Structures: Formal Definition H s [P, Peleg, SPAA’15]
60
Backup + Reinforcement s [P, Peleg, SPAA’15]
61
(b,r) FT-BFS Structures: Easy Cases r(n)=0 r(n)= n-1
62
The Tradeoff [P, Peleg, SPAA’15]
63
(b,r) FT-BFS Structures: The Tradeoff 1 1/2 1 3/2 Reinforcement (Quality) Backup (Quantity) [P, Peleg, SPAA’15]
64
(b,r) FT-BFS Structures: Lower Bound [P, Peleg, SPAA’15]
65
Theory of Fault Tolerant Networks: Take Home Message For distance related predicates understanding structure of replacement-paths is a crucial step.
66
Theory of Fault Tolerant Networks: Now what? Additional predicates (capacity, flows, etc.). Existential optimality vs. combinatorial optimality Multiple faults Lower bounds (e.g., FT-2-additive spanners) Distributed Distributed implementation. Online setting …
67
Thank you!
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.