I/O granularity transformations?

Abstract

In compiling I/O intensive operations, it is important to pay attention to the operating system features available for enhancing I/O performance. One such useful feature implemented by several current operating system is the write-back facility, in which the output data is cached in main memory buffers and then slowly written to disks. Since the buffers are of limited size, this feature can be made much more effective if the compiler can transform very large output operations into a sequence of smaller operations. In this paper, we present a static analysis technique through which the compiler can perform such transformations automatically.