The present work deals with an optimization procedure developed in the full-f global GYrokinetic SEmi-LAgrangian code (GYSELA). Optimizing the writing of the restart files is necessary to reduce the computing impact of crashes. These files require a very large memory space, and particularly so for very large mesh sizes. The limited bandwidth of the data pipe between the comput- ing nodes and the storage system induces a non-scalable part in the GYSELA code, which increases with the mesh size. Indeed the transfer time of RAM to data depends linearly on the files size. The necessity of non synchronized writing-in-file procedure is therefore crucial.A new GYSELA module has been developed. This asynchronous procedure allows the frequent writ- ing of the restart files, whilst preventing a severe slowing down due to the limited writing bandwidth. This method has been improved to generate a checksum control of the restart files, and automatically rerun the code in case of a crash for any cause.
展开▼
