PRESENTATION: Pursuing Performance in Store: Algorithms, Queries, Schemas
Production Smalltalk systems tend to be large. Smalltalk’s adaptability means that they have long lives. An effect of this size and longevity is lots of large bodies of code … and lots of versions of that code. Many of Cincom’s customers manage codebases substantially larger than the Cincom® ObjectStudio® or Cincom® VisualWorks® products, even with their bundled class libraries. For Store to adequately manage these large bodies of code, performance must not deteriorate, even in the face of very large modules with lots of versions. In the past two years, significant work has been done to eliminate Store bottlenecks discovered by customers and Cincom engineers.
Tom will talk about improvements that have already been made to Store algorithms, query improvements that let the database do more of the work and exploration of schema changes to allow further performance improvements. In addition, he will talk about using Store while changing it, the experience of being Cincom’s most performance-sensitive Glorp user, changes made for testing and performance benchmarking and the additional complexity and side effects involved in working to improve Store (and Glorp) performance on multiple, different databases at the same time.