Conundrum of Systems
Speaker: Dr. Alfred Spector , Vice President of Services and Software IBM Research Division
Date: February 6 2003
Relevant URL: http://www.csail.mit.edu/events/DLStalks/dlsarchive.html
Most software systems are very complicated, by most any metric. As computer scientists, we have delighted in measuring the number of add operations or memory fetches, but the most telling metric would be one that measures the reactions (or pain level) of users and systems administrators, a reaction of frustration, befuddlement, and annoyance. Even programmers often view systems on which they work as ungainly and run amuck. While we computer scientists might like to justify these problems by the newness of our discipline, it is now over 50-years old, and many overly complex systems have been built using the best, widely prescribed techniques of modularity and layering. We try hard to build systems "right" yet they still turn out too complex.
In this talk, I discuss the need for computer scientists to undertake a new study of complexity, not the one focused on time and space, but rather one that defines (perceived) system complexity (to create, to maintain, and to use), objectively quantifies it, and and then seeks to reduce it. I motivate the problem with examples and explain why some well-considered approaches may not have been right. I describe the fascinating challenges in attacking complexity, (1) ranging from agreeing on the meaning of it, (2) learning how to measure it, (3) advancing the science and technology (as, for example, autonomic computing), and perhaps (4) even effecting cultural change within our field. I hope to engender lively discussion that does not end when my lecture is over.
See other events that are part of Dertouzos Lecturer Series 2002-2003
See other events happening in February 2003