Self-Protecting Multi-Paths A Simple and Efficient Protection Switching Mechanism

Transcription

1 University of Würzburg Dept. of Distributed Systems Prof. Dr. P. Tran-Gia, Jens Milbrandt, Andreas Reifert Self-Protecting Multi-Paths A Simple and Efficient Protection Switching Mechanism

2 Outline > Overview on resilience options Simplicity Efficiency Derivation of the Self-Protecting Multi-Path (SPM) > Numerical Results Required backup for the SPM Sensitivity Analysis Impact of network topology Impact of traffic matrix > Conclusions 2

3 Network Resilience > Reliable communication is business-critical Network outage financial loss ISPs give guarantees for network availability > Telephony system % ( five nines ) reliability ~ 5 minutes unavailability per year Redundant layout of network entities (links, switching centers, processors, line cards, hot standbys, ) > Reliability for IP and MPLS networks Network resilience against link and node failures required Traffic rerouting to another path Backup capacity required for deviated traffic cost factor 3

4 Reaction or Proaction? > Path restoration IP rerouting Reconfiguration of LSPs Reaction time 400ms 40s > Protection switching Primary LSP and preconfigured backup LSP Reaction time: <100 ms Too slow! 4

5 Hot or Cold Standby? > 1+1 protection Live backup path carries traffic simultaneously > 1:1 protection Backup path carries traffic only if primary path fails Inefficient! Reduction of backup capacity possible 5

6 Local or End-to-End Protection? > Local protection One backup path per aggregate and link Local repair keeps all traffic nearby > End-to-end protection One backup path per aggregate Distant traffic deviation by end-to-end repair Too complex! Additional reduction of backup capacity? 6

7 Single- or Multi-Path Protection? > Single backup path Load balancing not possible 100% backup capacity per backup link required > Multiple backup paths Load balancing possible (100/n)% backup capacity per backup link required Additional reduction of backup capacity? 7

8 Triggering of Traffic Redistribution: Locally by Path Failure Recognition or Globally by Signalling? > Traffic redistribution only if source recognizes path failure Only relocation of affected paths possible > Traffic redistribution triggered by failure signalling Relocation of any paths possible Possibly more efficient Too complex! 8

9 Path Protection or Self-Protecting Multi-Path (SPM)? > Path Protection Single primary paths Multiple (disjoint) backup paths > Self-Protecting Multi-Paths No distinction between primary and backup paths Multiple (disjoint) paths > Action in a failure case Source recognizes (set of) failed paths Traffic deflection to working paths according to failure specific load balancing function Can be mimicked by SPM Additional reduction of backup capacity? 9

10 Testbed London Amsterdam Copenhagen > Topology of optical core network of COST239 (European research network) Brussels Luxembourg Berlin > Homogeneous traffic matrix Paris Zürich Prague Vienna > Protected failure scenarios S : all single link and node failures Milan > Performance metric: required backup capacity (1/2)

11 Performance Comparison of Resilience Mechanisms Required Backup Capacity 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% SPM-E OSPF SPM-R Path Protection Multi-path SPM-O routing helps a lot! Maximum Number of Disjoint Paths SPM outperforms the concept of distinct primary and backup paths! SPM requires only 17% backup capacity to protect all single link and node failures! 11

12 Comparison in Sample Networks Required Backup Capacity [%] Networks with nodes of degree 2 are not suitable! Protection Schemes OSPF 5DSP-4DSP-O 5DSP-OPT MT-4DSP-O MT-OPT 5SPM-O c h d bc fa g BD HFG Ae Lab03 COST Avg. Number of Disjoint Parallel Paths k* Backup capacity depends strongly on network topology. SPM outperforms all feasible methodsbyat least 10%. 12

13 Impact of Network Characteristics Required Backup Capacity [%] Number of disjoint paths reduces backup capacity Ne twork Cha ra cte ristics n = 10 n = 15 n = 20 n = 25 deg avg = 3 deg avg = 4 deg avg = 5 deg avg = 6 deg dev,max = 1 deg dev,max = 2 deg dev,max = 3 No significant influence of network size Avg. node degree reduces backup capacity Avg. Num. of Topologically Disjoint Paths k* (2/2)

14 Comparison of SPM and IP Rerouting Required Backup Capacity [%] (OSPF Normalized) n = 10 n = 15 n = 20 n = 25 deg avg = 3 deg avg = 4 deg avg = 5 deg avg = 6 deg dev,max = 1 deg dev,max = 2 deg dev,max = 3 Network Characteristics Avg. Number of Disjoint Parallel Paths k* SPM can take advantage of a suitable network topology. Plain IP rerouting can not. 14

15 Impact of Traffic Matrix on the Required Backup Capacity (COST239) Copenhagen London Amsterdam Brussels Berlin Paris Luxembourg Prague Vienna Zürich Milan 15

16 Conclusions > Self-Protecting Multi-Path (SPM) Traffic distribution over disjoint multiple paths Optimization of the load balancing function > Simplicity Little path configuration: one structure fits all scenarios. Load balancing only at the ingress LSP No signalling: local recognition of partial path failures sufficient > Efficiency 17% backup capacity required for COST239 network (OSPF: 72%) 1/3 of network capacity can be saved! Takes advantage of suitable resilient network topologies > Outlook: configuration of SPM in networks with given link capacities 16