Allocation of end-to-end delay objectives for networks supporting SMDS

The end-to-end delay objective allocation problem for networks supporting the Switched Multi-megabit Data Service (SMDS) is considered. Traditionally end-to-end service objectives are allocated to network elements in such a way that, if the allocated service objective for each network element is satisfied then the end-to-end service objectives are satisfied. Such an objective allocation strategy is referred to as a feasible objective allocation strategy. For networks supporting SMDS, the delay objectives state that 95% of the packets delivered from the origin subscriber network interface (SNI) to the destination SNI should be within a given time threshold. From the network monitoring point of view, this percentile type of delay objectives makes it complicated to compute feasible allocation strategies so that network elements instead of each origin-destination pair should be monitored. From network planning point of view, these end-to-end percentile-type delay objectives usually impose an excessively large number of nonconvex and complicated constraints. The emphasis of the paper is (i) to propose an efficient generic approach to replacing the set of end-to-end percentile-type delay constraints by a simpler set of network element utilization constraints (small size and convex), (ii) to investigate how this approach could be adopted in conjunction with a number of possible allocation schemes and (iii) to compare the relative effectiveness (in terms of the utilization thresholds determined) using M/M/1 and M/D/1 queueing models