Local and Metropolitan Area Networks are often modeled as Polling Systems with a single server cycling between the stations. Many works in the literature studied various aspects of those models, but only a few tackled the problem of station breakdowns. Recently Kofman and Yechiali [7] analyzed polling systems with failing nodes under the Gated and the Exhaustive protocols. In this work we investigate such a system under the Globally Gated service regime, where each station is subject to a random failure process. If a node fails during a visit of the server, the breakdown is discovered only at the end of the service (e.g. transmission of a message), whereupon the work at that station stops and the server moves to the next node. The service of the interrupted job will be re-attempted during the next (or a future) visit of the server. We present an analysis of this important model and derive various performance measures needed for efficient design, development and operation of such networks.