Waddell D FEMC2007

Parallel Processing as a Technical Solution to Large Havest Scheduling Problems

Dave Waddell, Ministry of Forests and Range, P.O. Box 9512 Stn. Prov. Govt , Victoria, BC, V8W 9C3, Canada, Dave.Waddell@gov.bc.ca

Consider a spatial harvest scheduling problem that encompasses a very large geographic area. Add to this a high level of complexity in forest management and a long planning horizon. Moreover, add the requirement that solution times be derived in minutes rather than hours. To address this problem, the BC Forest Service has developed a simulation model that takes advantage of parallel processing. The approach presented here is based on a spatial simulation model that pre-blocks based on size distribution requirements then roads the block centroids. The management unit is then broken into smaller spatially contiguous units which are scheduled simultaneously on a cluster of parallel processors. As a demonstration of the parallel concept, the model is implemented on a baewolf cluster with 7 slave nodes and has shown good results in preliminary testing.

However, parallel processing has added some new twists to an old problem. Solutions are presented regarding: hardware issues in terms of cluster configuration; operating systems limits and programming languages that allow “easy” parallel processing; communications and data transfer between nodes; harvest queuing to allow for setting harvest priorities and harvest targets; and the effects of incorporating additional features such as a Mountain Pine Beetle component.