Avoiding Routing Instability in ASTN-Based Multi-Layer Transport Networks

1. Avoiding Routing Instability in ASTN-Based Multi-Layer Transport Networks Stefan Bodamer, Jan Späth, Christoph Glingener Partly funded by the German government in the context of the MultiTeraNet research programme

2. Visibility Peak of Inflated Technology Trough of Slope of Plateau of Expectations Trigger Disillusionment Enlightenment Productivity Time ASTN/GMPLS: Hype, Depression, Recreation peak of hype: Supercomm 2001 many signs of recreation • big expectations: • wavelength demands • IP paradigm less demands than expected

3. Field Trials Real Network Implementations Interoperability Tests ASTN/GMPLS: Hype, Depression, Recreation • however: • no experience with real networks under load • no experience with dynamic services • careful design to avoid “unforeseen problems” • routing strategy • network upgrade strategy • “squeeze out” network as long as possible? • addressed by network simulation in the following

4. OXC/OPP EXC Network Model • 17 nodes German network • SDH/WDM multi-layer scenario • hybrid nodes (optical bypass and intermediate grooming) • network dimensioning • 10 Gbit/s line rate on all lightpaths • lightpaths obtained from given traffic matrix (“planned traffic demand”) • traffic modelling • dynamic arrival of SDH connection requests (Poisson process) • mean values according to given traffic matrix • traffic with either STM-1, STM-16, or STM-64 granularity • routing • fixed routing (FR) along pre-calculated shortest paths • dynamic routing (DR) using shortest available path

5. Instability EffectBlocking Probability • well-known effects • low load: DR better • high load: FR better • coarser granularity causes higher blocking • new observation: instability/“bi-stability” • sudden increase of blocking probability • emphasised in case of fine granular demands • emphasised in transparent networks

6. Instability EffectWhat happens in the network? • dynamic routing tries to avoid blocking by choosing longer paths • inefficient resource usage • “chain reaction”: overload is spread over the network • finally blocking occurs

7. blocking probability offered load Instability EffectImpact and Potential Solutions • highly relevant for network operators using ASTN • strategy to upgrade as soon as blocking occurs could be dangerous (network may be already down) • sudden load increase due to on-the-fly restoration events could trigger instability effect • lessons can be learned from telephone networks • similar observations in telephone networksreported by Nakagome/Mori in 1973 • hysteresis effect: significant load reduction required to return from high blocking state • proposals to avoid instability based on “early dropping” of connections

8. Avoiding InstabilityA Simple Mechanism: Utilisation Limit • referred to as “trunk reservation” in telephone networks • threshold q for overall network utilisation (e.g., q = 95%) • incoming connections are blocked if q is exceeded • “bi-stable” behaviour no longer present • high load blocking much lower than for DR (similar to FR) • no increase of blockingin low load region

9. Conclusions • instability effect under dynamic traffic • known in the context of telephone networks • but so far not considered in ASTN networks • potentially severe impact on network operators • solution • pro-active rejection of incoming connections in overload situations • example: global network utilisation limit • further investigation in the context of the IST project NOBEL, e.g. • optimal value for utilisation limit? • how to avoid regional instability?

10. Thank you! Marconi Communications ONDATA GmbH Stuttgarter Straße 139 D-71522 Backnang Germany www.marconi.de Stefan Bodamer Senior Engineer Network Architecture • Telefon +49 (0) 7191 13-3320 • Fax +49 (0) 7191 13-63320 • E-Mail Stefan.Bodamer@marconi.com