To save costs, corporations are increasingly using the shared infrastructure of network service providers to connect their various locations. Subscription to a so called 'virtual private network' service is an economical alternative to leasing dedicated lines. Similarly, e-commerce companies can have their web-sites hosted on 'server farms' of web hosting companies, instead of maintaining the servers and internet connections themselves.
In such cases, the service provider and customer sign a contract outlining the service quality to be delivered, the so called Service Level Agreement (SLA). SLAs can be quite comprehensive, encompassing not only network performance metrics, but also metrics for service reliability, disaster recovery, security, customer care and trouble handling. When SLAs are not met, customers are typically entitled to rebates. In order to determine whether SLA requirements are violated, careful monitoring is necessary. In fact, SLAs should contain for each metric: a precise definition, the acceptable value range, and the measurement technique.
Mathematical modeling can help build a framework for determining appropriate SLAs. In this talk, I will review several approaches taken to model network phenomena influencing the structure of SLAs. In particular, I will focus on the use of statistical techniques to help set achievable thresholds for SLA metrics.