Hi Dave, I retested the will-it-scale/read2 case with perf profile enabled, and here are the new comparison results. It shows that there are increased overheads in shmem_getpage_gfp(). If you'd like to collect more data, feel free to tell me.
9a0bb2966efbf30 0f6934bf1695682e7ced973f6 --------------- ------------------------- 26460 ~95% +136.3% 62514 ~ 1% numa-vmstat.node2.numa_other 62927 ~ 0% -85.9% 8885 ~ 2% numa-vmstat.node1.numa_other 8363465 ~ 4% +81.9% 15210930 ~ 2% interrupts.RES 3.96 ~ 6% +42.8% 5.66 ~ 4% perf-profile.cpu-cycles.find_lock_page.shmem_getpage_gfp.shmem_file_aio_read.do_sync_read.vfs_read 209881 ~11% +35.2% 283704 ~ 9% numa-vmstat.node1.numa_local 1795727 ~ 7% +52.1% 2730750 ~17% interrupts.LOC 7 ~ 0% -33.3% 4 ~10% vmstat.procs.b 18461 ~12% -21.1% 14569 ~ 2% numa-meminfo.node1.SUnreclaim 4614 ~12% -21.1% 3641 ~ 2% numa-vmstat.node1.nr_slab_unreclaimable 491 ~ 2% -25.9% 363 ~ 6% proc-vmstat.nr_tlb_remote_flush 14595 ~ 8% -17.1% 12093 ~16% numa-meminfo.node2.AnonPages 3648 ~ 8% -17.1% 3025 ~16% numa-vmstat.node2.nr_anon_pages 277 ~12% -14.4% 237 ~ 8% numa-vmstat.node2.nr_page_table_pages 202594 ~ 8% -20.5% 161033 ~12% softirqs.SCHED 1104 ~11% -14.0% 950 ~ 8% numa-meminfo.node2.PageTables 5201 ~ 7% +21.0% 6292 ~ 3% numa-vmstat.node0.nr_slab_unreclaimable 20807 ~ 7% +21.0% 25171 ~ 3% numa-meminfo.node0.SUnreclaim 975 ~ 8% +16.7% 1138 ~ 5% numa-meminfo.node1.PageTables 245 ~ 7% +16.5% 285 ~ 5% numa-vmstat.node1.nr_page_table_pages 109964 ~ 4% -16.7% 91589 ~ 1% numa-numastat.node0.local_node 20433 ~ 4% -16.3% 17104 ~ 2% proc-vmstat.pgalloc_dma32 112051 ~ 4% -16.4% 93676 ~ 1% numa-numastat.node0.numa_hit 273320 ~ 8% -14.4% 234064 ~ 3% numa-vmstat.node2.numa_local 31480 ~ 4% +13.9% 35852 ~ 5% numa-meminfo.node0.Slab 917358 ~ 2% +12.5% 1031687 ~ 2% softirqs.TIMER 513 ~ 0% +37.7% 706 ~33% numa-meminfo.node2.Mlocked 8404395 ~13% +256.9% 29992039 ~ 9% time.voluntary_context_switches 157154 ~17% +201.7% 474102 ~ 8% vmstat.system.cs 36948 ~ 3% +67.7% 61963 ~ 2% vmstat.system.in 2274 ~ 0% +13.7% 2584 ~ 1% time.system_time 769 ~ 0% +13.5% 873 ~ 1% time.percent_of_cpu_this_job_got 4359 ~ 2% +13.6% 4951 ~ 3% time.involuntary_context_switches 104 ~ 3% +10.2% 115 ~ 2% time.user_time Thanks, Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/