Efficient large-scale replica-exchange simulations on production infrastructure.

Replica-exchange (RE) algorithms are used to understand physical phenomena--ranging from protein folding dynamics to binding affinity calculations. They represent a class of algorithms that involve a large number of loosely coupled ensembles, and are thus amenable to using distributed resources. We develop a framework for RE that supports different replica pairing (synchronous versus asynchronous) and exchange coordination mechanisms (centralized versus decentralized) and which can use a range of production cyberinfrastructures concurrently. We characterize the performance of both RE algorithms at an unprecedented number of cores employed--the number of replicas and the typical number of cores per replica--on the production distributed infrastructure. We find that the asynchronous algorithms outperform the synchronous algorithms, even though details of the specific implementations are important determinants of performance.

Adaptive distributed replica-exchange simulations.

Owing to the loose coupling between replicas, the replica-exchange (RE) class of algorithms should be able to benefit greatly from using as many resources as available. However, the ability to effectively use multiple distributed resources to reduce the time to completion remains a challenge at many levels. Additionally, an implementation of a pleasingly distributed algorithm such as replica-exchange, which is independent of infrastructural details, does not exist. This paper proposes an extensible and scalable framework based on Simple API for Grid Applications that provides a general-purpose, opportunistic mechanism to effectively use multiple resources in an infrastructure-independent way. By analysing the requirements of the RE algorithm and the challenges of implementing it on real production systems, we propose a new abstraction (BigJob), which forms the basis of the adaptive redistribution and effective scheduling of replicas.