Why might you want to use Panda?

First, Panda provides high-performance collective array I/O. High-performance I/O can be achieved in Panda by utilizing I/O parallelism and Panda's unique I/O architecture called server-directed I/O. In server-directed I/O, I/O processors actively direct the entire I/O process rather than just passively respond to file read and write requests from the compute processors.

Second, Panda supports flexible in-memory and on-disk layouts. In Panda, applications can use different in-memory and on-disk layouts. If the user specifies different layouts in memory and on disk, Panda automatically and efficiently reorganize the data at I/O time. It supports both regular and irregular distribution of arrays. For regular distribution, Panda follows High-Performance Fortran (HPF) style distribution. The IOPanda thorn use (BLOCK, *, *) distribution as a default when writing out grid functions in Cactus, which makes the process to create a single array file much easier when multiple I/O processors creates multiple data files.

Finally, Panda supports many other useful features such as adaptive mesh refinement (AMR) and automatic data migration. For the IOPanda thorn, these features will be included soon.