Evaluation and estimation of the availability of p-cycle protected connections

Abstract

The availability is an important QoS metric of network connections. We can estimate the availability of a connection based on the availability parameters of the network components. There are several resilience schemes that can be applied to enhance the availability of the connections. However, an accurate approximation can be achieved with simple and fast calculation only for the most basic protection schemes. The more complex the applied protection scheme is the more complex calculation is required to get the exact results. Fortunately, heuristic algorithms have been developed which can give a fast approximation on the price of approximation error, and this error is negligible in the most cases of practical usage.

In this paper we examine and compare two methods that can be used to evaluate or to estimate the availability of connections protected by p-cycle protection scheme. The first method, that we propose, focuses on the accuracy, and by exploiting the special properties of p-cycles provides accurate results without enumerating all the possible network or protection configuration states. The second method is the well-known Serial-Parallel availability modeling and calculation method, which is known to be fast (O(n) steps in case of n components), however, its results can have approximation error since the model does not take into account the overlapping of connection components, i.e., when a component is member of more different series.

Setting the advantages and disadvantages of these methods against each other, first, we show, that even though, theoretically, the complexity of the proposed algorithm is still exponential (O(2^l)), the index (l) does not get high, moreover, it can be kept moderate low if the size and extent of the cycles is constrained. Next, we analyze the approximation error of the Serial-Parallel method. We prove that the estimated availability is always less than the exact one, define an upper bound onto the approximation error of the estimated unavailability and show where does this inaccuracy converge by increasing the availability of the network components.