Scheduling multiple streams with (m,k)-firm deadlines having different importance over Markovian channels

There are several (not only industrial) applications which can tolerate some losses. A concise way to distinguish between allowed and forbidden loss patterns are (m,k)-firm deadlines, which state that at most m out of k consecutive packets may get lost. An obstacle to the fulfillment of even the relaxed (m,k)-firm deadlines are channel errors. In this paper we consider the case where multiple periodic streams having the same period and their own (m_i,k_i)-firm deadlines are to be scheduled over a common, error-prone communications link. For each new packet transmission a scheduler has to decide which stream to serve next. A number of scheduling policies is investigated over a range of simple channel models of varying burstiness. It shows up that already in this simple case none of the scheduling policies is best (in the sense of having the least costs incurred when violating (m_i ,k_i)-firm deadlines) for all different values of the burstiness, but the best policy depends on the actual channel burstiness. One approach to improve this would be to design decision rules like "if burstiness is so and so, use policy A, otherwise use policy B". However, since it cannot be expected that such rules carry over to more elaborate channels, we have designed an adaptive "meta-scheduler", which continuously selects out of a set of candidate policies a policy that might be the best in the current channel situation because it would have been the best one in the immediate past. Depending on its parameters, the meta-scheduler achieves close-to-optimal performance