Dual-Queue Scheduling and an Engineering Delay Bound Under Overload

One shared queue under overload is considered. Seven of every ten extractions are taken from the tail and three from the head. An engineering upper bound on the wait of jobs removed from the head is derived for a sustained hybrid and a constant share θ. With eight servers, mean service 40 ms and a full buffer of 1000 jobs, the bound is about 16.7 s and sits above a 3 s threshold. A 10 Hz guard is therefore included in the sizing: it replaces LIFO with the hybrid when the head has waited more than 3 s. A measured P99 is not reported.