Impact of outages on GPRS service availabilityGeneral Packet Radio Service has been developed to enhance the GSM telecommunication system with the addition of packet-switched capabilities, as the key enabler for the integration between wireless transmission and Internet data services. Analyses of the GPRS expected behavior have been performed essentially focusing on measures like throughput and delay for the end-to-end packet transmission, and packet loss probability. Classical availability is defined as the property of "readiness of usage", measured as the delivery of correct service with respect to alternation of correct-incorrect service. However, in the case of GPRS, whose services are continuously required by users, the mere estimation of availability in terms of intervals of times the system is operative with respect to those of outage is not a satisfactory measure to know. During outages, there is an accumulation of users waiting for service. As soon as the GPRS becomes available anew, the high number of requests overloads the slotted Aloha mechanism used to get access to system resources. This collision generates a negative effect that lengthens the period of service provision interruption. The system requires some time to absorb the congestion before getting back to the state in which the average level of service quality is provided to the users. A study on the impact of outages on GPRS service availability is being conducted in the framework of the collaboration between the Modeling and Simulation Team at Motorola Technology Center Italy and the Dependable Computing Group at CNUCE, an institute of the Italian National Research Council based in Pisa. The investigation being performed explores the behavior of GPRS through a stochastic modeling approach based on Stochastic Activity Network models [1]. This work contributes to the analysis of GPRS in two ways:1. The behavior of the GPRS in presence of outages is modeled and analyzed [2], in order to estimate the degradation of the service quality as perceived by users and to understand the relevant phenomena. The curves in Figure 1 plot the time required for the system to go back to its steady-state conditions for various outage durations. As it can be observed, the congestion after an outage may lead to an additional service unavailability whose length is of the same order of magnitude as that of the outage itself. 2. An availability model of GPRS infrastructure is built, which allows deriving the distribution of the unavailability periods during system lifetime. Since the seriousness of the post-outage degradation is sensitive to the outage duration distribution, results drawn from the availability model are used to provide accurate outage characterization to be included in the model at point 1) above.