High Performance I/O for Parallel Computers: 
                    Problems and Prospects
           (Submitted: IEEE Computer, July 1993)

         Juan Miguel del Rosario and Alok Choudhary

           Northeast Parallel Architectures Center 
                 111 College Place, RM 3-201
                     Syracuse University 
                   Syracuse, NY 13244-4100


                           ABSTRACT

Along with the increasing use of parallel computers has come 
an increased demand for I/O systems support. Data movement to temporary 
storage, archival storage, visualization systems, or across the network 
to other computing centers has become a necessity for high-performance 
computing. The research and development of I/O systems for this type of 
environment is at a very early stage of its evolution. Although research 
in I/O systems have been conducted in the past, only recently have there 
been efforts to form a comprehensive characterization the I/O problem 
encompassing various perspectives (e.g., I/O in parallel machines, 
distributed computing, mass storage, etc.).

The aim of this paper is to present an overview of the many issues related 
to high-performance I/O in parallel computing environments. We discuss I/O
requirements for Grand Challenge applications, issues related to performance
characterization, I/O architecture alternatives, operating and file systems,
compiler and runtime support, checkpointing, network I/O and others. In most
cases, we present the current state of understanding and research, discuss
outstanding problems, and describe some alternative solutions.