E E Module 17 W. D. Grover TRLabs & University of Alberta © Wayne D. Grover 2002, 2003 ATM VP-based (or MPLS path) Restoration with Controlled Over- subscription of Restoration Capacity based on work by Y. Zheng, W.D. Grover

E E Module 17 © Wayne D. Grover 2002, Differences between STM & ATM or MPLS STM (Synchronous Transfer Mode) –Transfer mode traditionally used for the Transport Layer –Example: SONET/SDH –Switching is done on OC-n level carrier signals (“switching the containers”) –Demands are served using reserved time slots (synchronous) ATM –Technology developed for the B-ISDN network –Supports multiple Classes of Service (with guaranteed Quality of Service) –Allows more efficient use of bandwidth by exploiting the statistical nature of data traffic –Switching is done on Virtual Paths Connections (VPC or VP) –Traffic is packetized and transmitted asynchronously MPLS –IETF Technology for realizing connection-oriented functions similar to ATM VP / VC constructs in an IP-based network –Uses label-switching concepts pioneered by ATM.

E E Module 17 © Wayne D. Grover 2002, STM & ATM Differences from a Restoration Viewpoint Demand decomposition: –STM: Demand bundles are broken apart for restoration –ATM: VPs must have single equal-sized backup VP Bandwidth replacement: –STM: “Perfect bandwidth replacement” is required –ATM: Some flow convergence overloads (over subscription of bandwidth) can be tolerated during restoration Impact on capacity requirement

E E Module 17 © Wayne D. Grover 2002, STM/ATM Comparison: Restoration Granularity STM: “fine grain” rerouting ATM: single backup VP AZ AZ 10 STS Before failure Possible STM restoration pattern AZ VP BW = 10 STS1 AZ Backup VP BW =10 STS1 ATM Backup VP on single route only

E E Module 17 © Wayne D. Grover 2002, STM/ATM Comparison: Bandwidth Replacement STM: B D ZA 1 3 B D ZA ATM (or MPLS): /3: 33% restoration-induced flow convergence overload ATM: Restoration flow convergence technically possible (at the expenses of a higher packet loss probability) STM: Perfect bandwidth replacement

E E Module 17 © Wayne D. Grover 2002, ATM Capacity Design: KST Algorithm An algorithm proposed in the literature (KST algorithm * ) can be used to do the capacity design of an ATM network for restoration of any failure of a single VP Each VP has a pre-assigned disjoint backup VP for which capacity is reserved Until a failure arises on the working VP, no traffic is sent on the backup VP (Corresponding bandwidth can be used for working traffic). Backup VPs are therefore referred to as “Zero bandwidth backup VPs” * R. Kawamura, K-I Sato, I. Tokizawa, “Self-healing ATM networks based on virtual path concept,” IEEE Journal on Selected Areas in Communications, vol. 12, no. 1, January ATM “pipe” Virtual Path Zero Bandwidth backup VPs

E E Module 17 © Wayne D. Grover 2002, The Problem of Backup Flow Over-Subscription The KST algorithm does not consider the fact that a physical failure can simultaneously affect several VPs When multiple VP’s fail simultaneously, restoration overload can be experienced on some spans: –Restoration over subscription factor on span j for failure of span i: * R. Kawamura, K-I Sato, I. Tokizawa, “Self-healing ATM networks based on virtual path concept,” IEEE Journal on Selected Areas in Communications, vol. 12, no. 1, January Over-subscription factors are not controlled in the design process and therefore can potentially reach high values

E E Module 17 © Wayne D. Grover 2002, Uncontrolled Over-subscription With KST algorithm the resulting over-subscription factors are not controlled in the design process As a result, worst case over-subscription factors can be very high Source of Data: Y. Zheng, W. Grover, M. MacGregor, “Broadband network design with controlled exploitation of flow convergence overloads in ATM VP-based restoration,” Proc. Canadian Conference on Broadband Research (CCBR’98). Uncontrolled peaks

E E Module 17 © Wayne D. Grover 2002, ATM Design with Controlled Over-Subscription IP1: Minimum Spare Capacity with Design Limit on Maximum Restoration Over-Subscription –Objective: –Subject to: 1) Sparing is sufficient to keep restoration over-subscription below the design limit for all failures 2) Backup VPs are sufficient to meet the target restoration for all working VPs 3) Only one backup VP can be used for each working VP, i.e. VP flows are not split

E E Module 17 © Wayne D. Grover 2002, ATM Design with Controlled Over-Subscription IP2: Minimum Over-Subscription for given Spare Capacity allocation –Objective: –Subject to: 1) Backup VPs are sufficient to meet the target restoration for all working VPs 2) Only one backup VP can be used for each working VP, i.e. VP flows are not split

E E Module 17 © Wayne D. Grover 2002, Results of IP1 with X tol = 1.0 Comparative Spare Capacity Requirements Limiting X j,i to 1.0 means: No over subscription allowed (equivalent to STM) The consequence is: Higher spare capacity requirements to guarantee full bandwidth replacement available for each physical failure The KST Algorithm has low spare capacity requirements but has no control over the maximum X j,i.

E E Module 17 © Wayne D. Grover 2002, IP1 Results: Spare Capacity Requirements vs. X tol Very high reduction of spare capacity requirement is observed when the maximum over- subscription ratio is increased The question then is: “What over-subscription ratio can we allow without too much degradation of service during span-failure states?”

E E Module 17 © Wayne D. Grover 2002, Statistics of X j,i for a given X tol X tol is only the design-limiting maximum over-subscription In a design with a worst case over-subscription X tol, what is the experience of most spans that are not at the worst case?

E E Module 17 © Wayne D. Grover 2002, Relating Over-subscription to cell-level effects How much over-subscription might be tolerable without undue cell-level effects? Criterion: “Allow the worst case over-subscription ratio to be such that a pre-failure cell- loss-probability (CLP) of would do degrade to more than 10 -5, during the restored state” Note: even this worst-case outcome would occur only for span j, upon failure of span i, where X j,i = X tol is the design-limiting instance AND the failure occurs at the design busy traffic hour AND at the growth build-out horizon of the installed equipment. Method: FIFO B (R 1,  1, b 1 ) (R 3,  3, b 3 ) (R 2,  2, b 2 ) (R N,  N, b N ) c -9 = equivalent bandwidth at which CLP=10 -9 for buffer size B Simulation Queuing Model Simulation parameters: (n,R, , b) n: number of sources R: peak rate per source  : utilization per source b: mean burst length

E E Module 17 © Wayne D. Grover 2002, Results of Equivalent Bandwidth Simulations Simulation determines the equivalent bandwidth c -5 at which CLP=10 -5 c -5 / c -9 corresponds to the over-subscription factor for which a CLP of will degrade to Results show that in many case, an over-subscription of 5 to 10% would be acceptable based on the CLP increase criterion This corresponds to a 10 to 20% reduction of spare capacity according to previous results (slide 12)

E E Module 17 © Wayne D. Grover 2002, Conclusion ATM is an alternative approach to transport networking that allows QoS control and a better exploitation of statistical nature of data traffic for higher capacity utilization However, we have seen that: –Capacity design for ATM transport networks needs to consider the possibility of multiple VP failures resulting from single physical failures otherwise potentially high overload situations could be experienced during restoration –10-20 % capacity savings can still be obtained (relative to STM) when controlled limited over-subscription is allowed in the design process (based on a CLP limit of when the original CLP is ) These concepts also apply to more recent IP-based transport networks