Nowdays we know a lot of information systems that store and provide different information. To make it easier for remote clients to get information from those systems, we connect them with the help of an interoperable information system, which expands the search to every connected system and returns a composed response. That way remote clients don't have to worry about specific information systems to get combined information from them. Interoperable informative system is challenged by the avaliability and resposiveness of information system to which it connects. They can sometimes only be partly responsive or even unaccesible. To prevent interoperability system to overload remote information system, we would like to throttle requests to that system. We built a throttling subsystem that is capable of recognizing an inoperative or overloaded system and in that case lower its number of demands. That's why we collected production data, analysed it and prepared it for learning prediction models. The results show that throttle can reliably recognise or predict pending inoperative source and lower its demands.
|
