This project started in 2001 when a new approach to parallel programming of master-worker applications using an abstraction for interprocess communication was proposed and called NetFiles. See
The programmer writes the parallel application as a collection of sequential modules that communicate with each other by reading and/or writing files. When run on a sequential machine, the data is written to and read from conventional files. But, when run on a parallel platform, these file operations are actually implemented using message passing. Our approach makes it possible to develop a parallel program in two phases. In the first phase, the user is concerned with how to decompose the data and code, but without concern for the details of the parallel platform. In the second phase, the program can be configured to run on a parallel environment with little or no modification to the sequential code. We demonstrate the effectiveness of our approach on two parallel master/worker programs: parallel matrix multiplication and parallel genetic algorithms.
By 2005, the research was linked to FAbric architecture as described inand described as an "Enhanced Stream Based Communication Mechanism". The paper abstract said:
Netfiles is an alternative API for message passing on distributed memory machines that is based on pipes. It provides enhanced capabilities such as broadcasts and gather operations. Because Netfiles overload conventional file I/O operations, parallel programs can be developed and tested on a file system before execution on a parallel machine. Netfiles is part of a parallel programming system called FAbrIC.The design and implementation of the FAbrIC architecture gives an effective approach to parallel applications, for example,a parallel shallow water model application and parallel Jacobi method.