Full text loading...
-
Scalable numerical modelling patterns for the cloud
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 79th EAGE Conference and Exhibition 2017, Jun 2017, Volume 2017, p.1 - 5
Abstract
The ready availability of cloud computing resources presents an opportunity for rapid turnaround in large numerical modelling tasks. This opens up new possibilities for interactive modelling in immersive applications. However, it is only feasible for numerical modelling algorithms that scale well over large computing clusters. This is relatively straightforward for algorithms that are embarrassingly parallel, but achieving linear scaling for algorithms such as coupled numerical modelling problems is much more challenging. We explore the improvements that can be achieved in scalability for this type of algorithm by moving away from a sequential programming approach as conventionally used with the Message Passing Interface (MPI), which encourages large-scale synchronisation across a parallel system. Instead we propose an approach based on the actor model that removes all unnecessary synchronisation during inter-node communication. It avoids synchronisation by introducing flexibility in the order of computation. It also spreads communication evenly in time and therefore significantly reduces network contention. We use a theoretical model to examine the scalability characteristics of the new system, which gives improvements of more than an order of magnitude in scalability compared with the sequential approach.