Working Paper: NBER ID: w18714
Authors: Yongyang Cai; Kenneth L. Judd; Greg Thain; Stephen J. Wright
Abstract: We implement a dynamic programming algorithm on a computational grid consisting of loosely coupled processors, possibly including clusters and individual workstations. The grid changes dynamically during the computation, as processors enter and leave the pool of workstations. The algorithm is implemented using the Master-Worker library running on the HTCondor grid computing platform. We implement value function iteration for several large dynamic programming problems of two kinds: optimal growth problems and dynamic portfolio problems. We present examples that solve in hours on HTCondor but would take weeks if executed on a single workstation. The use of HTCondor can increase a researcher's computational productivity by at least two orders of magnitude.
Keywords: Dynamic Programming; Computational Grid; Parallel Computing
JEL Codes: C61; C63; G11
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| htcondor MW system (C88) | computational time reduction (C63) |
| parallelization (C59) | computational time reduction (C63) |
| htcondor system (C88) | task execution efficiency (H21) |
| master-worker paradigm (J54) | computational time reduction (C63) |