This paper presents the Meta Task Graph (MTG) as an intermediate model of effective expressing parallelism between various problems and efficient algorithms for solving these problems for volunteer computing. One of the goals of this work is to develop effective techniques to deal with the heterogeneities of potential applications and system architectures. Another goal of MTG is to provide a tuple-matching mechanism, which supports fast synchronization of data-dependence and crash tolerance. Ongoing state collection of the underlying system is also a goal. For scheduling to be optimal, various pieces of information are needed. This feature provides full optimisation potential for the design of task assignment and scheduling for load balance to deal with dynamic environment of volunteer computing. To demonstrate MTG's applicability and flexibility, this paper presents a real-world problem and its MTG implementation on our simulation platform GNet. The future research on MTG is presented at the end of this paper.