United Kingdom, June 12, 2013

The workshop aims at introducing the notion of mesh
partitioning or graph partitioning as a pre-processing load
balancing step for applications that rely on High
Performance Computing. Both static and dynamic partitioning
approaches will be discussed based on the needs of the
underlying physical applications.

Description:

 

Course Objectives 

 

Effective use of any High Performance Computing (HPC)  

Infrastructure is important and often a mandatory pre- 

requisite to gain access to larger HPC facilities.  

Partitioning of the underlying graph, or mesh, is a crucial  

pre-processing step towards a balanced work split amongst  

the processors that will later perform the heavy  

computational work. This pre-processing task is particularly  

challenging when dealing with larger unstructured meshes or  

when the computational objectives are complex or many-fold.  

Practice has shown that careful consideration of the choice  

of the partitioning strategy leads to significant  

computational gains and hence effective use of computational  

infrastructure. 

 

Course Topics 

 

The course features lectures on: 

 

Graph Partitioning 

Mesh Partitioning 

Load Balancing. 

 

Who Should Attend? 

 

Practitioners involved in numerical simulations in a  

parallel computing environment (ranging from smaller  

Computing Clusters to larger High Performance Computing  

infrastructures). In particular Engineers involved in any  

Computational Fluid Dynamics (CFD) or other simulations  

where the computational domain is described through an  

unstructured mesh or the computational load is not static. 

 

Practitioners whose data can be represented in a graph  

format (social networks for example), where this graph is  

too large to be analysed in traditional desktops and thus  

the use of Parallel Computing is mandatory.