Vivek Thakkar wrote: > Hi Vijay, > > I have also used move_pages() in my thesis to move very high number of > pages ( >100 K). Apparently, there is another way to move pages, which > is through mbind() syscall with MPOL_MF_MOVE_PAGE flag. In a > micro-benchmark, I found that mbind takes really less time as compared > to move_pages(). The simple test case allocates a large number of pages That is surprising since they both trigger a call to the same migration function. > (linearly contiguous) on one node and then uses either move_pages or > mbind to initiate migration on a different note. I found that for high > number of pages (> 20 K), the difference is of the order of 100x. I > don't have an answer to this. I did some theoretical calculations (based > on the NUMA interconnect bandwidth and the amount of data to be > migrated) and it looks that mbind is probably a truer representation of > the migration time. Huh? There must be some other effects here. The method of gathering the pages is even similar. Do you just move_pages to move pages of the currently executing process or of pages from some other process? -- To unsubscribe from this list: send the line "unsubscribe linux-numa" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
