On Mon, Oct 24, 2011 at 11:10:35AM -0500, Dimitri Sivanich wrote: > Avoid false sharing of the vm_stat array. > > This was found to adversely affect tmpfs I/O performance. > I think this fix is overly simplistic. It is moving each counter into its own cache line. While I accept that this will help the preformance of the tmpfs-based workload, it will adversely affect workloads that touch a lot of counters because of the increased cache footprint. 1. Is it possible to rearrange the vmstat array such that two hot counters do not share a cache line? 2. Has Andrew's suggestion to alter the per-cpu threshold based on the value of the global counter to reduce conflicts been tried? (I'm at Linux Con at the moment so will be even slower to respond than usual) -- Mel Gorman SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>