An evaluation of parallel simulated annealing strategies with application to standard cell placement

Abstract

Simulated annealing, a methodology for solving combinatorial optimization problems, is a very computationally expensive algorithm, and as such, numerous researchers have undertaken efforts to parallelize it. In this paper, we investigate three of these parallel simulated annealing strategies when applied to standard cell placement, specifically the TimberWolfSC placement tool. We have examined a parallel moves strategy, as well as two new approaches to parallel cell placement, multiple Markov chains and speculative computation. These algorithms have been implemented in ProperPLACE, our parallel cell placement application, as part of the ProperCAD II project. We have constructed ProperPLACE so that it is portable across a wide range of parallel architectures. Our parallel moves algorithm uses novel approaches to dynamic message sizing, message prioritization, and error control. We show that parallel moves and multiple Markov chains are effective approaches to parallel simulated annealing when applied to TimberWolfSC, yet, speculative computation is wholly inadequate.

Exploring Asynchronous MMC-Based Parallel SA Schemes for Multiobjective Cell Placement on a Cluster of Workstations

“…However, the work done by Chandy et al, shows that particularly for the standard cell placement problem, speculative execution SA succumbs to a very high overhead and thus is not a feasible option [8].…”

“…Sun and Sechen [16] have shown results achieving near linear speedup on a network of workstations, also using this approach. Chandy and Bannerjee [8] implemented this method for standard cell placement on both a shared-memory Sun 4/690MP as well as a distributed-memory Intel iPSC/860, with the former exhibiting a speedup of approximately 2 on 4 processors, and the latter achieving a maximum speedup of 3.75 on 8 processors. It is important to note at this point that virtually all of the parallel methods listed above exhibited degradation of final solution quality as more processors were added.…”

“…They include move-acceleration, parallel-moves, and speculative annealing and are generally more suitable for shared-memory environments. Approaches based on Multiple Markov chains call for the concurrent execution of separate simulated annealing chains with periodic exchange of solutions [7,8]. This approach is particularly promising since it has the potential to use parallelism to increase the quality of the solution rather than simply accelerate the algorithm.…”

“…We experiment with different versions of the Asynchronous Multiple-Markov Chain Parallel SA (or AMMC PSA) approach described in [7], as this scheme has been found to be well suited to solving this problem in a distributed-memory environment [8]. Our goal is to develop parallel SA implementations that:…”

“…It involves a close interaction between processors, and has less potential for parallelism in terms of the amount of parallel work performed and the number of processors that can be employed. Such methodologies are largely restricted to shared memory architectures [8] and preserve all the properties of the serial algorithm. Kravitz and Rutenbar [11] implemented this parallel SA method for cell placement on a shared memory multiprocessor, achieving a speedup of 2 on 4 processors.…”