Position Statement

Peter Corbett (IBM Research, corbett@watson.ibm.com)

Kai Li (Princeton University, li@cs.princeton.edu)

Six current trends will drive computer science research and product development, including I/O research, in new directions over the next ten years.

• Ever increasing network connectivity, both within and among organizations, will require transparent interoperation of applications and services across multiple platforms. The operating system, network protocol, and hardware must become less limiting as the user's interaction with the computer more and more is through multiplatform networked applications. The desktop computer will become an access point to many multi-tiered servers, connected via application interfaces that run locally. Current examples are web browsers, and Lotus Notes.
• Disconnected and mobile operation will become more important as users demand that their computing environment move with them, with a minimum of luggage.
• The distinction between programs and data will disappear as the two are bundled together in objects that are dynamically transported over networks.
• Consumer electronics and entertainment will merge with home computing. Home consumers will demand interactive services with comparable video and audio quality to current broadcast media (including satellite and cable distribution).
• Inexpensive ways of storing data will emerge in the mass market. Examples include writable CDs, DVDs and high density drives. The new storage methods will enable new applications.
• NonUniform Shared Memory Multiprocessors will become the enterprise servers, enhancing and replacing the current installed base of mainframe systems.

These trends have some direct implications on I/O research:

• Traditionally, I/O services have been intrinsically connected to the operating system. However, the recent trend to web based computing has provided a new level of cross-platform I/O interface over IP networks. Researchers should be thinking about how to separate I/O from the operating system to provide cross-platform networked I/O services to all applications in a heterogeneous environment.
• Research on file systems, such as CODA, should continue and expand, to enable greater levels of mobility and disconnected operation to users, even in environments where data is write shared.
• Object based, distributed client-server computing will become the dominant paradigm of computing. As a result, I/O systems will have to serve both programs and data, bundled as objects, efficiently. Some of the issues this implies are recognizing what executable needs to be moved, efficiently caching and storing common executables and their multiple data instances, and meeting requirements placed on the I/O system based on the type of object being moved. For example, multimedia objects may have real-time constraints on their distribution.
• There will be increasing consumer demand for I/O for "entertainment" purposes, including interactive viewing, on-demand viewing, gaming, communication, and education. I/O systems will have to keep pace with this demand, giving service providers a cost effective back-end for their distribution systems. Research on using new storage methods will be important.
• Research to improve I/O subsystem performance will continue to be important. Examples include research on new architectures to take advantage of high-performance networking and new operating system structures to streamline I/O operations.
• The common I/O interfaces used for the heterogeneous, networked, disconnected environments should be implemented in a very efficient way on shared memory multicomputers. Research should be done on the most efficient connection and session semantics to exploit computers that efficiently run many threads.