Research On Molecular Dynamics Load Balancing Key Technology Based On OpenFOAM

As a kind of computer simulation method, molecular dynamics simulations have widely used in physics, chemistry, bioscience, materials science, medicine, etc. With the continuous search for knowledge, using high performance computers to simulate largescale molecular systems has become an inevitable trend. However, as the high performance computer brings the powerful computing ability, it is also facing various problems,such as load imbalance, communication, reliability, power consumption, etc. Among the problems, load imbalance will lead to more performance loss. To solve this problem, this thesis proposes two load balancing methods for parallel molecular dynamics and implements them in Open FOAM. The main work and innovations of this thesis are as follows:? Building the molecular dynamics load balancing framework in Open FOAM(Chapter 2)This thesis deeply analyzes the program framework and implementation of OpenFOAM. Based on the analysis results, we find load imbalance will hardly affect the molecular dynamics simulation efficiency. To overcome that difficulty, based on the framework of Open FOAM, we build the static and dynamic load balancing mechanism, the corresponding data organization, program flow, framework, and the coupled method with other modules. The load balancing framework proposed in this thesis can offer references for other particle methods.? Proposing a static load balancing method for parallel molecular dynamics——MDSLB(Chapter 3)This thesis proposes MDSLB, a new molecular dynamics static load balancing method. By analyzing the characteristics of the short-range force of molecular dynamics programs running in parallel, we divide the short-range force into three kinds of force models, and then package the computations of each force model into many tiny computational units called “cell loads”, which provide the basic data structures for our load balancing method. In MDSLB, the spatial region is separated into sub-regions called “local domains”, and the cell loads of each local domain are allocated to every processor in turn. MDSLB can guarantee load balance byexecuting the algorithm only once at program startup without migrating the loads dynamically.? Proposing a dynamic load balancing method for parallel molecular dynamics——MDDLB(Chapter 4)For simulating small-scale molecular systems, this thesis proposes a dynamic load balancing method——MDDLB. Concretely, we introduce the basic idea and coordination mechanism of MDDLB, design the load migrating unit and bring in the threshold detection mechanism for monitoring and measuring the load of each processor. When the load imbalance level is beyond a certain value, it is necessary to migrate the load from heavily loaded processors to lightly loaded processors.MDDLB can effectively solve the load balancing problem of small-scale molecular dynamics simulations.? Verifying the correctness and effectiveness of the load balancing framework, MDSLB, and MDDLB(Chapter 5)This thesis designs and implements the molecular dynamics load balancing framework in Open FOAM, which is proposed in Chapter 2. Based on this framework, we implement MDSLB and MDDLB which are proposed in Chapter 3 and Chapter 4.We use the diffusion model to test the above framework and methods on a Tian He-1A subsystem, and the results show that the framework and the two methods are correct and effective. Compared with the traditional method, MDSLB can obtain 34%-64% acceleration, and MDDLB can get closer to the ideal speedup.