How to Reduce Network Interrupt Latency for Real-Time Processing

Abstract

There are efforts to transform the Linux system to a real-time system as it becomes very popular and stable. Linux divides the interrupt service into two steps: Top Half and Bottom Half. When the interrupt occurs, Because of this Bottom Half which is not handled immediately, Linux is not a real-time system. Transforming Linux system into a real-time system, however, is a very difficult and time-consuming task because we have to modify the kernel in many places to ensure its correctness under repeated reentering events. We suggest limiting the effort to a certain specific interrupt, for example network interrupt, instead of trying to provide real-time handling for all interrupts. The scope is limited, and we can control the amount of shared data structures among the reentering kernel threads. We identify a special packet that requires real-time treatment and process it inside the Top Half without delaying. This thesis explains the algorithm and shows how it is implemented in Linux.