One of the main challenges in the HPC clusters is the I/O bottleneck between the main memory (disk storage system) and the node memory. This bottleneck is a consequence of the fact that the disk access speed is thousands-fold slower than the processing speed of the parallel co-processors. We propose a strategy to reduce the impact of this bottleneck in a heterogeneous cluster.
Año: 2018
ISSN: 00000
Grupo(s):