Optical grid systems have been viewed as a promising virtual computing environment to support distributed real-time DAG (directed acyclic graph) applications. For such a system involving many heterogeneous computing and network resources, faults seem to be inevitable. Therefore, a fault-tolerant DAG scheduling scheme is necessary to improve the performance of the optical grid system. However, existing joint task scheduling schemes for real-time DAG applications generally do not consider the availability issues when making scheduling decisions. In this paper, we develop an availability-driven scheduling (ADS) scheme that improves the DAG availability iteratively by allocating two copies of one communication task to two disjoint lightpaths for data transfer while satisfying application deadline requirements. Extensive simulation results demonstrate the effectiveness and the feasibility of the proposed scheduling scheme.