TMODEL Procedure

PROC TMODEL uses concurrent computation threads to reduce the time it takes to perform estimation and simulation tasks. Because the characteristics of the input data, model program, and task can vary, PROC TMODEL uses different strategies for breaking its calculations into pieces that can be executed concurrently. For example, when you are estimating a simple model by using a data set with many observations, the most efficient multithreading strategy is to partition the observations and to evaluate the objective function concurrently across the partitions. However, for a highly nonlinear estimation problem with many parameters and fewer observations, the best strategy might be to execute the minimization problem concurrently for multiple regions of the parameter space and then choose the region that yields the optimal estimates.

PROC TMODEL automatically determines which threading strategy to use for each estimation or simulation modeling task by default. Alternatively, you can specify the threading strategy manually by specifying priority options in the PERFORMANCE statement. For each modeling task, the threading strategy is determined by the number of threads allocated to each job in the task. A job is an aspect or dimension of the task that can be executed concurrently with another job. The total number of threads that are used to complete each modeling task is determined as follows,

where is the actual number of threads used, n is the number of threads specified, , is the priority specified for the ith job, and J is the set of all jobs in the modeling task. The allocation of threads to jobs in modeling tasks creates the same number or more threads than the number specified in the PERFORMANCE statement and CPUCOUNT= system option.