Why is distributed system evolution not better supported?

摘要

This position paper takes the view that modern programming languages, their tools and software architectures do not adequately support the programmer in their day-to-day task of evolving large, long-lived, distributed systems. Evolving programs is the dominant cost on these kinds of system projects and the programmer is not as well supported in this task as they should be. This is argued by presenting what these three technologies do well and less well at code development time and at system run-time. We discuss the various degrees of support that are offered and then describe what can be improved for the tractable problems and what can be done for a particular intractable problem.