Kernel Level Distributed Inter-Process Communication System (KDIPC)

, 2004

Category: Operating Systems

Overall Rating

2.3/5 (16/35 pts)

This paper is primarily a historical account of a specific, flawed attempt to provide a simple, transparent distributed shared memory and semaphore interface by implementing it deep within the Linux 2.4 kernel and using a single-copy sequential consistency model.
...the high obscurity and the general idea of kernel-level IPC interception for transparent state distribution could serve as minor inspiration for highly niche modern work on low-latency interconnects like CXL...
...the KDIPC system itself is obsolete, brittle, and fundamentally limited by its performance model for concurrent workloads.
It offers no concrete, actionable blueprint for modern research beyond a conceptual pattern...

...the specific design point of implementing a traditional, simple local IPC API (System V) with strict sequential consistency at the kernel level, primarily using a single-copy/ownership model, was not the dominant direction...
...offers a distinct trade-off space (simplicity + strictness vs. potential contention) that isn't heavily explored in modern distributed state management...
modern network technologies... especially low-latency, high-bandwidth interconnects like RDMA and emerging technologies like CXL... fundamentally change the performance profile of simple protocols like KDIPC's single-copy/ownership model.
...its kernel-level interception of standard local IPC calls... and the adoption of a simple sequential consistency protocol... offers an unconventional blueprint for managing distributed state on CXL-attached memory pools.

The core of KDIPC is tied to a specific, now ancient, operating system context: the Linux 2.4 kernel and its Virtual Memory (VM) system.
Kernel-level hooks and data structure manipulations... designed for Linux 2.4 would be incompatible and require a complete rewrite for any modern kernel.
...the chosen implementation strategy of maintaining only a single active copy of each shared object... is a performance bottleneck.
Building a complex distributed system inside the kernel is inherently more risky, harder to maintain, and less flexible than user-space or hybrid approaches...

Watch