Fast Context Switching in Real - time

The trend to increasingly capable and aaordable control processors has generated an explosion of embedded real-time gadgets that serve almost every function imaginable. The daunting task of programming these gadgets is greatly alleviated with real-time de-ductive engines that perform all execution and monitoring functions from a single core model. Fast response times are achieved using an incremental propo-sitional deductive database (an LTMS). Ideally the cost of an LTMS's incremental update should be linear in the number of labels that change between successive contexts. Unfortunately an LTMS can expend a signiicant percentage of its time working on labels that remain constant between contexts. This is caused by the LTMS's conservative approach: a context switch rst removes all consequences of deleted clauses, whether or not those consequences hold in the new context. This paper presents a more aggressive in-cremental TMS, called the ITMS, that avoids processing a signiicant number of these consequences that are unchanged. Our empirical evaluation for spacecraft control shows that the overhead of processing unchanged consequences can be reduced by a factor of seven.