TABLE OF CONTENTS This section considers how best to implement overload controls that satisfy the requirements discussed in the preceding sections. It is organised into sub-sections which discuss the following issues:
why both internal and external overload controls are needed (section 7.6.1);
factors pertinent to internal overload control design types of internal workload and their performance requirements (section 7.6.2);
factors pertinent to external overload control design (section 7.6.3);
feedback control responsiveness versus stability (section 7.6.4);
why controlling reject rate is better than controlling reject proportion (section 7.6.5);
an example of good design (Automatic Call Restriction section 7.6.6).
An overload control has to restrict demand, whether by returning an indication explicitly rejecting it, or discarding it without notifying the sender. In a scheme which uses feedback (i.e. a closed loop control), this is done according to signals or measurements obtained from the overloaded resource. It is useful to distinguish between an internal control, where the system in overload itself does the restriction, and external control, where the system performing the restriction is remote from the overloaded resource.
Shedding demand will always take some non-zero processing effort. This can be made small, but some effort is necessary in order to:
discriminate so that subsequent demand (after the initial request) is not lost;
discriminate in favour of priority demand;
(possibly) obtain additional information such as called address.
It follows that an internal control alone is not sufficient, because as the load increases...
