Abstract – This paper studies the unavailability of mesh-restorable networks from the standpoint of dual-span failures as the main source of outage in networks that are efficiently designed to fully restore any single failure. Computer emulation of the de
Computational and Design Studies on theUnavailability of Mesh-restorable Networks
Matthieu Clouqueur, Wayne D. Grover
TRLabs, #800 10611 – 98th Avenue, Edmonton, Alberta, Canada T5K 2P7
Department of Electrical and Computer Engineering, University of Alberta, Edmonton, Alberta
Contact: {mclouque, grover}@edm.trlabs.ca
Abstract – This paper studies the
unavailability of mesh-restorable networks from thestandpoint of dual-span failures as the main source ofoutage in networks that are efficiently designed tofully restore any single failure. Computer emulation ofthe detailed routing behavior of three progressivelymore adaptive restoration protocols is used to find thedual-failure restorability (R2) under modular and non-modular capacity assumptions. This is supplementedwith two new optimization formulations to study theimplications of designing for Rthe trade-off of achievable R2=100% and to observe2 versus the total allowedinvestment in spare capacity.
I. INTRODUCTION
There has been considerable work to date on thedesign of span- and path-restorable mesh networks[1,2] and on the design of ring-based networks [3].Both types of network are nominally designed toachieve 100% restorability against any single span cutat near minimum cost. Making networks "fullyrestorable" in this sense greatly improves theavailability of service relative to a non-restorabletransmission network, but it does not mean that “100%availability” is also achieved. In fact there has beenvery little characterization of what level of R2 mightbe obtained as a built-in side-effect of designing for R1=100%. Especially in modular capacity cases, andhighly adaptive mechanisms, R2 might be surprisinglyhigh. Alternately one might even begin designingmesh networks in a bi-objective sense, i.e.: formax(R2) subject to (min capacity for R1 =100%).
Accordingly recent interests have evolved past theproblem of restorability design itself, to quantifyingthe availability of services through such "fullyrestorable" networks. Some recent work [4,5,6]considered ring-based networks in this light. A closedform model is provided in [5] and augmented in [6]for the availability of a path through a ring-basednetwork, including all details of two-failurecombinations and dual-ring interconnectarrangements. Related work in [13] has alsoconsidered multi-layer availability analysis methods.
In contrast to rings, the availability analysis ofmesh-restorable networks is generally not as amenableto a completely analytical approach. The flexiblenature of a mesh-restorable network –its routingadaptability and its extensive sharing of sparecapacity- make it far less clear how to directly
enumerate the outage-causing failure combinations,as is evidently feasible with rings [4,5,6]. On theother hand, mesh networks lend themselves morereadily to standard optimization methods, so theremay be more scope to evolve the design methods soas to directly enhance R2 by design. Thus, theresearch questions we have set out are along thefollowing lines:
(i) what is the availability benefit of the basicinvestment for R1 = 100% in a mesh network?,
(ii) can the same investment for R1 be furtherexploited to minimize the two-failure unavailabilitye.g, maximize R2?
(iii) what are the effects of modularity in thepresence of a highly adaptive DRA1 on meshavailability?
(iv) what would be the capacity cost of strictlydesigning for R2 =100% ? 2
(v) what is the shape of the trade-off betweentotal capacity required and the achievable R2 level?A. Overview
The first part of the paper studies the two-failurerestorability (Rnominally to support only single failure restoration.2) of mesh networks that are designedThis is based on an essentially experimentalcomputational approach. We find it quite feasible tosimply conduct every dual failure restorationexperiment explicitly, so there is little reason not toadopt this approach at present, as opposed toattempting an analytical model with inherentapproximations. The experimental approach lets usexactly model the topology, the working and sparecapacity distributions, modularity, and detaileddifferences in the restoration protocol, especially itslevel of adaptability to prior circumstances.
The second main part of the work presents newoptimization formulations to address questions suchas "How much spare capacity would be needed forcomplete two-failure restorability?" and "Given abudget, what is the highest R2 level (and thedistribution of spare capacity) that can be achieved?As a practical matter that we think should encounterlittle objection, the entire study is based on dual-failure sources of outage. We do not expect any of
1
DRA = distributed restoration algorithm
2
A method for this design problem has also previously appearedin [14], although it was not there accompanied by quantitativeresults.
Design of a Meta-Mesh of... 暂无评价 15页 免费 Computational and Design... 暂无评价 6页 免费 Span-Restorable Mesh Net... 4页 免费 MAINTENANCE-IMMUN...
Design of a Meta-Mesh of... 暂无评价 15页 免费 Computational and Design... 暂无评价 6页 免费 Span-Restorable Mesh Net... 4页 免费 MAINTENANCE-IMMUN...
热门文档