Overload Management Through Selective Data

During system and network overload periods, excessive delay or even data loss may occur. To maintain the quality of control of an NCS, the implementation system (including both computer and network) overload must be correctly handled. As we can see in the previous chapters, a common approach to dealing with this overload problem is to dynamically change the sampling period of the control loops. In this chapter, as an alternative to the explicit sampling period adjustment, we present an indirect sampling period adjustment approach which is based on selective sampling data dropping according to the (m, k)-firm model [HAM 94]. The interest of this alternative is its easy implementation despite having less adjustment quality, since only the multiples of the basic sampling period can be exploited. Upon overload detection, the basic idea is to selectively drop some samples according to the (m, k)-firm model to avoid long consecutive data drops. The consequence is that the shared network and processor will be less loaded. However, the control stability and performance must still be maintained to an acceptable level. This can be achieved by keeping either the total control tasks on a same processor or the messages sharing a same network bandwidth schedulable under the (m, k)-firm constraint. In this chapter, we first give a sufficient condition for scheduling a set of control tasks under (m, k)-firm constraint (section 5.2). Then, through several examples, we