related presentation - University of Alberta by zhangsshaohui123

VIEWS: 0 PAGES: 16

									Mesh Restorable Networks with Complete
   Dual Failure Restorability and with
   Selectvely Enhanced Dual-Failure
        Restorability Properties


  Matthieu Clouqueur, Wayne D. Grover (presenter)
               clouqueur@trlabs.ca, grover@trlabs.ca
                  TRLabs and University of Alberta
                      Edmonton, AB, Canada


 web site for other related papers: www.ee.ualberta.ca/~grover


                       OptiComm 2002
                         Boston, MA, USA
                          30/July/2002
Outline


  • Background on Dual Failure Restorability

  • Ideas and Motivations

  • Research Methods

  • Experimental Results

  • Conclusions and Impacts




                                        Matthieu Clouqueur and Wayne D. Grover
                           2            OptiComm 2002 - Boston, MA, July 2002
Dual Failures - Really ?

Not as “academic” a consideration as we first thought:
• Sheer fiber route miles
   – Hermes RailTel estimate of one one cable cut /4 days

• Span maintenance and upgrade effects
   – can be much like a first failure in network equivalent effects

• Span SRLG and nodal bypass effects
   – cause logical dual failures

• Availability of paths through single-failure restorable
  networks:
   – unavailability doesn’t just vanish...
            Becomes limited by dual failures

                                                    Matthieu Clouqueur and Wayne D. Grover
                                   3                OptiComm 2002 - Boston, MA, July 2002
Background re: Dual Failure Restorability

• Prior work on dual failure restorability analysis of
  span-restorable mesh networks: (refs: DRCN 01, JSAC 02)
   – concept of “first failure protection, second failure
     restoration”
                          (pre-planned reaction)        (adaptive reaction)
   – method for dual failure restorability analysis

• Some key findings:
   – 1) Span restorable (or “link-protected”) mesh networks
     designed for R1 = 100%, give very high average R2 values
     as a side-effect !
   – 2) Service path availability has far more to do with
                       restorability to dual failures,
               not the speed of response to a single failure
   – and …3) Explicit design for R2=100% is very capacity-
     expensive                  4
                                                 Matthieu Clouqueur and Wayne D. Grover
                                                 OptiComm 2002 - Boston, MA, July 2002
Background: Determination of “R2”


Case 1: Two failures but no spatial interactions      no outage


Case 2: Two failures and spatial interactions (competition for spare
capacity)
                                       may be outage


Case 3: Two failures with second failure hitting the first restoration
pathset                              may be outage

Case 4: Two failures isolating a degree-2 node
                                       certain outage
-> Use computer emulation of all dual failure pairs to analyze R2
                                                      Matthieu Clouqueur and Wayne D. Grover
                                  5                   OptiComm 2002 - Boston, MA, July 2002
     Prior Finding of High Dual-failure Restorability in Networks
     Designed for Single Failure Protection / Restoration ...
             100 %
                                                Between 50 %               70 % to
                                                      and               90 % network
                                                99 % R2(i j) on          average R2
                                                   individual
                                                   scenarios

              R1                              R2
(Single failure restorability)    (Dual failure restorability)

                     R2 Results
                                                Non-modular                   Modular
                     for 5 test networks:       environment           Environment
                   Static behavior              0.53 to 0.75          0.69 to 0.83
                   First-failure adaptive       0.55 to 0.79          0.87 to 0.91
                   Fully-adaptive               0.55 to 0.80          0.91 to 0.99

                                                                  Matthieu Clouqueur and Wayne D. Grover
                                            6                     OptiComm 2002 - Boston, MA, July 2002
Research Questions

• Is it possible to enhance the dual span-failure restorability of an
  R1=1 network design:
    – purely by a redistribution of the spare capacity ?
    – to maximize R2 subject to a given budget limit ?
• Can we structure or allocate the finite R2 levels that are
  obtained to support a super-high availability service class ?
     “Platinum                                                  Dual-failure
     service class” =            new                            restorable
     assured dual-               gold                           service class
     failure                     silver
     restorability                                            Existing QoP
                                 bronze                       paradigm
                                 (economy)

                                                           Matthieu Clouqueur and Wayne D. Grover
                                     7                     OptiComm 2002 - Boston, MA, July 2002
Methods to Investigate these Questions

Three Design Models :

• Dual Failure Minimum Capacity (DFMC):
       Finds the minimum capacity assignment for full restorability to
         dual-failures (R2=100%)


• Dual Failure Max Restorability (DFMR)
       Finds the spare capacity placement that maximizes the average
         restorability to dual-failures for a given spare capacity budget


• Multi-service Restorability Capacity Placement
  (MRCP)
       Finds the minimum capacity assignment and routing that serves
         demands of multiple service classes including R0 (best-effort),
         R1 and R2-assured restorability service classes

                                                         Matthieu Clouqueur and Wayne D. Grover
                                    8                    OptiComm 2002 - Boston, MA, July 2002
Complete Dual Failure Restorability at Minimum
Capacity (DFMC)

 Minimize:
     Total Cost of Capacity
 Subject to:
     (1) All demands are routed
     (2) Working capacity supports (1)
     (3) Restoration flows for 100% span restoration in the
     presence of each other span failure
     (4) Spare capacity to support (3)


 Note: This is with spare capacity reuse / sharing across non-
 simultaneous failure scenarios implicit in all cases

                                                  Matthieu Clouqueur and Wayne D. Grover
                                  9               OptiComm 2002 - Boston, MA, July 2002
Dual Failure Maximum Restorability at Given
Capacity (DFMR)

 Minimize:

        Total No. of Un-restorable Working
        Channels over all dual failure scenarios
 subject to:
    (1) All demands are routed
    (2) Working capacity supports (1)
    (3) Spare capacity less than an allowed Budget
    (4) Restoration flows as feasible under (4) for all span
    failures in the presence of each other span failure



                                                   Matthieu Clouqueur and Wayne D. Grover
                                 10                OptiComm 2002 - Boston, MA, July 2002
Multi-service Restorability Design at Minimum
Capacity (MRCP)

Define:
   “R1” , “R2” (and also “R0”)-restorable service demand matrices

Minimize:

                Total Capacity
subject to:
   (1) All demands are routed, (2) Working capacity supports (1)
   (3) Restoration flows for all dual span failure scenarios for
   “R2” demands
   (4) Restoration flows for all single span failures for all “R1”
   demands
   (5) Spare capacity to support (3) and (4)
                                                     Matthieu Clouqueur and Wayne D. Grover
                                 11                  OptiComm 2002 - Boston, MA, July 2002
Results with DFMC (Cost for R2=1 by design)

• BENCHMARK: Cost of designing for full dual-failure restorability

                                       R2        Total Cap.
            Network   Nodal deg.   Redundancy   Increase / R1
             6n14s1     4.67        116.8 %       50.5 %                       Large
                                                                              capacity
            11n20s1     3.63        258.9 %       87.3 %                     increases
            11n20s2     3.63        161.3 %       77.1 %
                                                                                 are
                                                                            required to
            12n18s1       3         268.6 %       84.5 %                      provide
                                                                               strictly
            12n24s1       4         145.9 %       65.7 %
                                                                              100% R2
            16n26s1     3.25        248.2 %       94.7 %


  – Interpretation: Although average dual-failure restorability
    levels are quite high with a R1 design, the capacity cost for
    making the network restorable to all dual failures is extremely
    high, (~ 3 x in spare capacity relative to R1=1 design)
                                                           Matthieu Clouqueur and Wayne D. Grover
                                    12                     OptiComm 2002 - Boston, MA, July 2002
Results with DFMR (Acheivable R2 vs. Cost)

• Trade-off between capacity and best acheivable dual-failure
  restorability:
                                       high capacity requirement as R2 =1 is
            Pure Redistribution        approached (confirms DFMC results)
            of capacity




                                                                     “Budget
                                                                     amount”
                                                          Matthieu Clouqueur and Wayne D. Grover
                                  13                      OptiComm 2002 - Boston, MA, July 2002
Results with MRCP
(MultiRestorability Service Class Design)
• Results of MRCP confirm that R2 restorability can be
  guaranteed end to end for selected service paths:



                      Up to about 20% of demands can be guaranteed R2 =1
                      restorability for a small or negligible capacity increase




                                                      Matthieu Clouqueur and Wayne D. Grover
                               14                     OptiComm 2002 - Boston, MA, July 2002
Concluding Insights and Comments

• Designing for 100% Dual-failure restorability is feasible but very
   expensive

• DFMR design method can maximize the network average dual
   failure restorability (R2) given any total budget for capacity.

• MRCP design can structure and enhance the R2 ability of an R1-
   designed network onto specific priority paths:
    – 20 to 40% of all demands per O-D pair could be in this “platinum”
      service class at very little or no extra capacity cost.

• And note ! Such R2-restorable service paths will have availability
   that exceeds that of 1+1 APS...

                                                            Matthieu Clouqueur and Wayne D. Grover
                                     15                     OptiComm 2002 - Boston, MA, July 2002
   A Key Insight: why priority services in a “mesh-restorable”
   will network get better than 1+1 APS availability
                 1+1 APS                    “1F-P 2F-R” mesh (for a priority path)


Normal                                                                            Normal



First failure                                                                First failure
 -> protection                                                                -> protection



Second                                                                       Second failure
failure ->                                                                     -> restoration !
outage                                                                       (adaptive)

R2(ij) =0                                                                    no outage yet

                 “Takes a licking and keeps on ticking” :-)                  R2(ij) >0
                                                                Matthieu Clouqueur and Wayne D. Grover
                                       16                       OptiComm 2002 - Boston, MA, July 2002

								
To top