An Approach to Reduce Application-to-Application Latency in a Real-Time Distributed Architecture

The great interest raised by the advantages of using Ethernet technology to support real-time communications has led to various proposals in the literature, involving both Shared and Switched Ethernet, to obtain predictable behavior. However, to reduce application-to-application transfer time, latency within the end-nodes has to be addressed too. This paper illustrates a solution to improve the average application-to-application delay in a distributed architecture featuring real-time data exchanges. The approach is implemented via a Middleware located between the UDP/IP and the network driver. The paper presents an implementation of the proposed solution in the SHaRK kernel and a performance evaluation on a real system.