Improving collectives by user buffer relocation

According to top 500 list, in November 2010 processors with four or more cores were used in the 92% of the HPC systems, while systems using six or more cores per processor have increased from 19% to 62% until November 2011. In the next years it is expected hundreds of cores per processor competing for memory bandwidth and aggravatting the problem known as memory wall, the decrease of memory bandwidth relative to a processor that make applications to reach just a few percent of its peak performance.