Abstract
Adaptation in time-triggered systems can be motivated by energy efficiency, fault recovery, and changing environmental conditions. Adaptation in time-triggered systems is achieved by preserving temporal predictability through metascheduling techniques. Nevertheless, utilising existing metascheduling schemes for time-triggered network-on-chip architectures poses design time computation and run-time storage challenges for adaptation using the resulting schedules. In this work, an algorithm for path reconvergence in a multi-schedule graph, enabled by a reconvergence horizon, is presented to manage the state-space explosion problem resulting from an increase in the number of scenarios required for adaptation. A meta-scheduler invokes a genetic algorithm to solve a new scheduling problem for each adaptation scenario, resulting in a multi-schedule graph. Finally, repeated nodes of the multi-schedule graph are merged, and further exploration of paths is terminated. The proposed algorithm is evaluated using various application model sizes and different horizon configurations. Results show up to 56% reduction of schedules necessary for adaptation to 10 context events, with the reconvergence horizon set to 50 time units. Furthermore, 10 jobs with 10 slack events and a horizon of 40 ticks result in a 23% average sleep time for energy savings. Furthermore, the results demonstrate the reduction in the state-space size while showing the trade-off between the size of the reconvergence horizon and the number of nodes of the multi-schedule graph.
Highlights
The processing capacity of recent VLSI technology has considerably grown as several Processing Elements (PEs), tensor cores, memory elements and Intellectual Property (IP) cores are integrated onto a single chip
An algorithm for path reconvergence in a multi-schedule graph, enabled by a reconvergence horizon, is presented to manage the state-space explosion problem resulting from an increase in the number of scenarios required for adaptation
These new schedules are computed by the metascheduler, which results in directed acyclic graph (DAG) as illustrated in Figure 4 to cover the occurrence of the given context events
Summary
The processing capacity of recent VLSI technology has considerably grown as several Processing Elements (PEs), tensor cores, memory elements and Intellectual Property (IP) cores are integrated onto a single chip. Fault tolerance at a lower cost than active redundancy, energy efficiency, and adaptation to changing environmental conditions necessitates adaptation services to accommodate run-time changes These run-time changes may arise from an execution slack (slack events), failures in system resources (failure events) or from a change in an operational mode requiring different application services [2]. The context of hybrid scheduling discussed in this work is such that schedule changes are made at run time from multiple pre-computed static scheduling solutions obtained at design time, referred to as metascheduling. The metascheduler pre-computes the MSG at design time using an application, platform and context model. The exponential growth of the MSG caused by the increasing number of context events results in the state-space explosion.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
