Mesh partitioning and load balancing for distributed memory parallel systems

A method is outlined for optimising graph partitions which arise in mapping unstructured mesh calculations to parallel computers. The method employs a relative gain iterative technique to both evenly balance the workload and minimise the number and volume of interprocessor communications. A parallel graph reduction technique is also briefly described and can be used to give a global perspective to the optimisation. The algorithms work efficiently in parallel as well as sequentially and when combined with a fast direct partitioning technique (such as the Greedy algorithm) to give an initial partition, the resulting two-stage process proves itself to be both a powerful and flexible solution to the static graph-partitioning problem. Experiments indicate that the resulting parallel code can provide high quality partitions, independent of the initial partition, within a few seconds. The algorithms can also be used for dynamic load-balancing, reusing existing partitions and in this case the procedures are much faster than static techniques, provide partitions of similar or higher quality and, in comparison, involve the migration of a fraction of the data.

Automatic implementation of dynamic load balancing strategies for structured computational mechanics codes

This paper presents a new dynamic load balancing technique for structured mesh computational mechanics codes in which the processor partition range limits of just one of the partitioned dimensions uses non-coincidental limits, as opposed to using coincidental limits in all of the partitioned dimensions. The partition range limits are 'staggered', allowing greater flexibility in obtaining a balanced load distribution in comparison to when the limits are changed 'globally'. as the load increase/decrease on one processor no longer restricts the load decrease/increase on a neighbouring processor. The automatic implementation of this 'staggered' load balancing strategy within an existing parallel code is presented in this paper, along with some preliminary results.

Mesh partitioning: a multilevel balancing and refinement algorithm

Multilevel algorithms are a successful class of optimization techniques which addresses the mesh partitioning problem. They usually combine a graph contraction algorithm together with a local optimization method which refines the partition at each graph level. In this paper we present an enhancement of the technique which uses imbalance to achieve higher quality partitions. We also present a formulation of the Kernighan-Lin partition optimization algorithm which incorporates load-balancing. The resulting algorithm is tested against a different but related state-of-the-art partitioner and shown to provide improved results.

Mesh partitioning: A multilevel balancing and refinement algorithm

Multilevel algorithms are a successful class of optimisation techniques which address the mesh partitioning problem. They usually combine a graph contraction algorithm together with a local optimisation method which refines the partition at each graph level. In this paper we present an enhancement of the technique which uses imbalance to achieve higher quality partitions. We also present a formulation of the Kernighan-Lin partition optimisation algorithm which incorporates load-balancing. The resulting algorithm is tested against a different but related state-of the-art partitioner and shown to provide improved results.

Quality balancing for parallel adaptive FEM

We present a dynamic distributed load balancing algorithm for parallel, adaptive finite element simulations using preconditioned conjugate gradient solvers based on domain-decomposition. The load balancer is designed to maintain good partition aspect ratios. It can calculate a balancing flow using different versions of diffusion and a variant of breadth first search. Elements to be migrated are chosen according to a cost function aiming at the optimization of subdomain shapes. We show how to use information from the second step to guide the first. Experimental results using Bramble's preconditioner and comparisons to existing state-ot-the-art load balancers show the benefits of the construction.

Multiphase mesh partitioning

We consider the load-balancing problems which arise from parallel scientific codes containing multiple computational phases, or loops over subsets of the data, which are separated by global synchronisation points. We motivate, derive and describe the implementation of an approach which we refer to as the multiphase mesh partitioning strategy to address such issues. The technique is tested on several examples of meshes, both real and artificial, containing multiple computational phases and it is demonstrated that our method can achieve high quality partitions where a standard mesh partitioning approach fails.

Multiphase mesh partitioning for parallel computational mechanics codes

We consider the load-balancing problems which arise from parallel scientific codes containing multiple computational phases, or loops over subsets of the data, which are separated by global synchronisation points. We motivate, derive and describe the implementation of an approach which we refer to as the multiphase mesh partitioning strategy to address such issues. The technique is tested on example meshes containing multiple computational phases and it is demonstrated that our method can achieve high quality partitions where a standard mesh partitioning approach fails.

The development of parallel implementation for a CFD based fire field model utilising conventional office based PCs

Parallel processing techniques have been used in the past to provide high performance computing resources for activities such as fire-field modelling. This has traditionally been achieved using specialized hardware and software, the expense of which would be difficult to justify for many fire engineering practices. In this article we demonstrate how typical office-based PCs attached to a Local Area Network has the potential to offer the benefits of parallel processing with minimal costs associated with the purchase of additional hardware or software. It was found that good speedups could be achieved on homogeneous networks of PCs, for example a problem composed of ~100,000 cells would run 9.3 times faster on a network of 12 800MHz PCs than on a single 800MHz PC. It was also found that a network of eight 3.2GHz Pentium 4 PCs would run 7.04 times faster than a single 3.2GHz Pentium computer. A dynamic load balancing scheme was also devised to allow the effective use of the software on heterogeneous PC networks. This scheme also ensured that the impact between the parallel processing task and other computer users on the network was minimized.

Parallel algorithms of the Purcell method for direct solution of linear systems

In this paper, we first demonstrate that the classical Purcell's vector method when combined with row pivoting yields a consistently small growth factor in comparison to the well-known Gauss elimination method, the Gauss–Jordan method and the Gauss–Huard method with partial pivoting. We then present six parallel algorithms of the Purcell method that may be used for direct solution of linear systems. The algorithms differ in ways of pivoting and load balancing. We recommend algorithms V and VI for their reliability and algorithms III and IV for good load balance if local pivoting is acceptable. Some numerical results are presented.

Parallel partitioning of unstructured meshes

A method is outlined for optimising graph partitions which arise in mapping un- structured mesh calculations to parallel computers. The method employs a combination of iterative techniques to both evenly balance the workload and minimise the number and volume of interprocessor communications. They are designed to work efficiently in parallel as well as sequentially and when combined with a fast direct partitioning technique (such as the Greedy algorithm) to give an initial partition, the resulting two-stage process proves itself to be both a powerful and flexible solution to the static graph-partitioning problem. The algorithms can also be used for dynamic load-balancing and a clustering technique can additionally be employed to speed up the whole process. Experiments indicate that the resulting parallel code can provide high quality partitions, independent of the initial partition, within a few seconds.

DRAMA: Dynamic re-allocation of meshes for parallel finite element applications

In many areas of simulation, a crucial component for efficient numerical computations is the use of solution-driven adaptive features: locally adapted meshing or re-meshing; dynamically changing computational tasks. The full advantages of high performance computing (HPC) technology will thus only be able to be exploited when efficient parallel adaptive solvers can be realised. The resulting requirement for HPC software is for dynamic load balancing, which for many mesh-based applications means dynamic mesh re-partitioning. The DRAMA project has been initiated to address this issue, with a particular focus being the requirements of industrial Finite Element codes, but codes using Finite Volume formulations will also be able to make use of the project results.

Load sharing in cluster service provision

The factors that are driving the development and use of grids and grid computing, such as size, dynamic features, distribution and heterogeneity, are also pushing to the forefront service quality issues. These include performance, reliability and security. Although grid middleware can address some of these issues on a wider scale, it has also become imperative to ensure adequate service provision at local level. Load sharing in clusters can contribute to the provision of a high quality service, by exploiting both static and dynamic information. This paper is concerned with the presentation of a load sharing scheme, that can satisfy grid computing requirements. It follows a proactive, non preemptive and distributed approach. Load information is gathered continuously before it is needed, and a task is allocated to the most appropriate node for execution. Performance and reliability are enhanced by the decentralised nature of the scheme and the symmetric roles of the nodes. In addition, the scheme exhibits transparency characteristics that facilitate integration with the grid.

Implementation of a proactive load sharing scheme

This paper presents a proactive approach to load sharing and describes the architecture of a scheme, Concert, based on this approach. A proactive approach is characterized by a shift of emphasis from reacting to load imbalance to avoiding its occurrence. In contrast, in a reactive load sharing scheme, activity is triggered when a processing node is either overloaded or underloaded. The main drawback of this approach is that a load imbalance is allowed to develop before costly corrective action is taken. Concert is a load sharing scheme for loosely-coupled distributed systems. Under this scheme, load and task behaviour information is collected and cached in advance of when it is needed. Concert uses Linux as a platform for development. Implemented partially in kernel space and partially in user space, it achieves transparency to users and applications whilst keeping the extent of kernel modifications to a minimum. Non-preemptive task transfers are used exclusively, motivated by lower complexity, lower overheads and faster transfers. The goal is to minimize the average response-time of tasks. Concert is compared with other schemes by considering the level of transparency it provides with respect to users, tasks and the underlying operating system.