Error Management in the Pluggable File System

Distributed computing continues to be an alphabet-soup of services and protocols. No single system for managing CPUs or I/O devices has emerged (or is likely to emerge) as a universal solution. Therefore, distributed applications require adapters in order to plug themselves into existing systems. The difficulty of building such adapters lies not in normal operations, but in the complications of failures and other unusual situations. We demonstrate this with the Pluggable File System, an adapter for connecting POSIX applications to remote I/O services. We offer a detailed discussion of the construction of the system while dealing with failures and other events that are not trivially mapped into the application’s expectations. The key insight is that correct I/O management requires coordination with CPU management. We conclude with some practical advice for others constructing similar software. 1