An Internet-based distributed manufacturing system utilizing a recurrent neural network controller for dynamic buffer size tuning to prevent user-level buffer overflow and shorten the service roundtrip time

摘要

An Internet-based distributed manufacturing system usually consists of a network/grid of autonomous centers that collaborate together. Each center may receive world-wide customer orders and collaborates with other friendly manufacturers, sub-contractors, agents, and suppliers to maximize the sales and quality of the goods. The centers together strike a consistent balance in terms of manufacturing speed, quality, material costs, and workload. The queue maintained by the front-end server/coordinator in each center usually contains requests from different internal and external sources. The merged traffic from these sources can inundate the queue buffer and cause overflow easily in high loading situations. As a result this leads to request retransmissions, unreliable collaborations, and unhappy customers. One way to eliminate buffer overflow in non-persistent (transient) situations is to tune the buffer size on the fly to ensure that it always cover the queue length. The recurrent NNC (neural network controller) or R-NNC proposed in this paper can achieve efficacious dynamic buffer size tuning at the user level for autonomous centers in an Internet-based distributed manufacturing system.