File Input and Output ( I/O ) often become a severe bottleneck when parallel applications run on large numbers of processors.
In this talk, we will show how to use I/O profiling tools , such as Recorder and Darshan, to understand and increase the I/O performance of an application. We will investigate a large portion of the I/O software stack, including Lustre, POSIX calls, MPI-IO and HDF5 libraries.
As an example, we will investigate the I/O performance of a simple application writing a 3D array to the filesystem and the I/O behaviour of a real application, such as Nektar++.
This online session is open to all. It will use the Blackboard Collaborate platform.