We have a relatively simple Apache Mahout implementation (multiple similarity-based recommendations based on data exported from a CMS and held in memory, queried via Tomcat based API requests by a Ruby on Rails web app), which has been in operation for several years without any update.
During this time the volume of user data has increased significantly, and the speed and memory consumption of the Mahout instance have started to deteriorate (now at 14GB+ of RAM consumption), so scalability is becoming a concern.
If possible, we would like to enlist specialist assistance to bring our Mahout stack up to date and hopefully reduce memory consumption significantly, preferably without introducing further complexity to the stack.