Abstract: A method for check pointing a global state of a distributed system with one or more distributed applications organized in a directed acyclic graph topology includes, upon receiving a marker in an active input channel of a first task application, putting an active input channel on hold, performing check pointing by saving an internal state of the first task application when all input channels have received a marker and are put on hold, forwarding the marker via all output channels of the first task application to at least one other task application of the one or more task applications, and reactivating all input channels of the first task application, wherein the global state is a union of all internal states of the task applications after each of the one or more task applications has been check pointed.