|
Recommendation ITU-T H.248.11 describes a package for media gateway (MG) overload control for use with the H.248.1 gateway control protocol. It serves to protect an MG from processing overload that prevents the timely execution of H.248.1 transactions. In summary, in this Recommendation, overload protection is achieved as follows: 1) An MG (or virtual MG) detects that it is in overload and notifies its media gateway controller (MGC) of that fact whenever it receives an ADD command. 2) The MGC adaptively throttles the rate it sets up calls using that MG (or virtual MG) to maximize the MG's effective throughput whilst bounding its response times. It does this by throttling the rate at which transactions that set up new calls or that new call legs are sent to the overloaded MG, so as to cause the rate of overload notifications the MGC receives from the overloaded MG (or virtual MG) to converge to a suitably low level. A separate instance of the overload control shall be initiated at an MGC for each of its dependent MGs (or virtual MGs) that is overloaded. These separate instances should run independently (that is, they do not explicitly exchange information). Their overload control parameters shall be separately configurable, for example, by means of a proprietary management interface, or the use of SNMP to invoke configuration functions. The most general overload scenario the control can handle is where one or more MGCs are jointly overloading a single MG that has several virtual MGs (virtual MG 'i' interacting only with MGC "i'). The control does not need to know how many MGCs are causing the MG to be overloaded, nor what the MG capacity is. Informative reference [b-Oftel] provides a full explanation of one way this can be achieved, and informative reference [b-whitehead] provides further material on designing overload controls. The overload control is largely specified by saying how it shall behave, but not how it should be implemented to achieve that behaviour. This has two important consequences. As a first consequence, part of the package (see clause 8.5) defines a set of overload scenarios, and any fully-compliant implementation of the package must automatically (i.e., without the need for operator intervention to adjust parameter values from one overload scenario to another) satisfy all the requirements for each of the scenarios. As a second consequence, not all configurable parameters can be known to this package since they depend upon specific implementations of the control. Nevertheless, there is a requirement that the implementation shall provide a means by which an operator can change all parameters which affect the performance of the control. See clause 9 for the management requirements associated with this package. It is expected that they will be realized by a proprietary management interface, or the use of SNMP. Corrigendum 1 to Recommendation ITU-T H.248.11 clarifies that, when the MG is overloaded, overload events may be sent either only following the first ADD.request which creates a new context, or following all ADD.request commands. These two options result in different normalizations of the overload event rate as an indicator of the level of MG overload. The concept of normalization is explained. |