Dynamic QOS control based on real-time threads

Performance analysis of dynamic soft real-time systems

“…Our soft real-time system [3,4,8] is based on the notion of Quality of Service (QoS) levels [25]. The soft real-time resource and timing requirements of an application are abstracted into a table relating the value of a computation to a level of resource allocation.…”

“…Many researchers have investigated mechanisms for supporting soft real-time processing [3,7,11,12,19,20,25,26]. The basic assumptions in these systems are that some or all of the applications can miss deadlines and still perform satisfactorily, and that situations can occur in which the worst-case resource needs of all of the running applications may not simultaneously be satisfiable.…”

“…The basic assumptions in these systems are that some or all of the applications can miss deadlines and still perform satisfactorily, and that situations can occur in which the worst-case resource needs of all of the running applications may not simultaneously be satisfiable. To support this type of processing, these systems variously provide a static resource guarantee with the possibility of renegotiation of those guarantees [13,20], or dynamic assignment of resources based on a specified application importance, utility or benefit (hereinafter called "benefit") for each application [4,7,11,17,20,24,25]. Those systems that provide a static resource guarantee must also have some notion of the relative benefit of the applications if they are to have any meaningful renegotiation of the guarantees when the resources become oversubscribed [14].…”

“…Some systems provide direct support for the applications to do this [1,2,17,24,25], while others provide information to the applications to allow them to do this, but without providing any direct support [7,13,19,20]. In all cases, the applications may simply run slower, which is the result of running the same algorithm with less resources.…”

“…In all cases, the applications may simply run slower, which is the result of running the same algorithm with less resources. However, many of these systems support and expect that applications will choose to actively modify their resource usage in other ways based on the resources that they have been granted [4,13,17,19,20,24,25].…”