Adaptive structuring of distributed databases
K. Dan Levin · 1982
This paper reports the study of an adaptive strategy for allocating programs and data files in computer networks when only imperfect information on access request rates is available. The adaptive system is based on a set of computer procedures to collect access statistics, to estimate frequencies, to find optimal file assignment based on expected improvements in performance, and to redistribute the files over the network nodes. These computer modules are synchronized with each other in order to predict the best assignment policy on the basis of where access requests are expected to arise in the future; and as new information becomes available, the forecast is revised and the assignment policy adapts itself to the new information. Transition costs incurred by file transmission from one node to another are taken into account, and the decision to reassign files is triggered only when net gains can be realized.