Re: [linus:master] [sched/deadline] 14a8570564: stress-ng.mmaprandom.ops_per_sec 18.9% improvement
Christian Loehle <[email protected]> Tue, 14 Jul 2026 22:36:32 +0100
| Newsgroups | dev.linux.lists.oe-lkp,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
On 7/12/26 10:25, kernel test robot wrote:
>
>
> Hello,
>
> kernel test robot noticed a 18.9% improvement of stress-ng.mmaprandom.ops_per_sec on:
>
>
> commit: 14a857056466be9d3d907a94e92a704ac1be149b ("sched/deadline: Use revised wakeup rule for dl_server")
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
>
>
> testcase: stress-ng
> config: x86_64-rhel-9.4
> compiler: gcc-14
> test machine: 192 threads 2 sockets Intel(R) Xeon(R) 6740E CPU @ 2.4GHz (Sierra Forest) with 256G memory
> parameters:
>
> nr_threads: 100%
> testtime: 60s
> test: mmaprandom
> cpufreq_governor: performance
>
>
>
> Details are as below:
> -------------------------------------------------------------------------------------------------->
>
>
> The kernel config and materials to reproduce are available at:
> https://download.01.org/0day-ci/archive/20260710/[email protected]
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/nr_threads/rootfs/tbox_group/test/testcase/testtime:
> gcc-14/performance/x86_64-rhel-9.4/100%/debian-13-x86_64-20250902.cgz/lkp-srf-2sp2/mmaprandom/stress-ng/60s
>
> commit:
> v7.0-rc7
> sched-urgent-2026-04-12
>
> v7.0-rc7 sched-urgent-2026-04-12
> ---------------- ---------------------------
> %stddev %change %stddev
> \ | \
> 7.42 ± 6% +13.7% 8.44 ± 4% stress-ng.mmaprandom.munmap_lo/hi_addr_ops/sec
> 3.268e+08 ± 3% +18.8% 3.884e+08 ± 3% stress-ng.mmaprandom.ops
> 5266977 ± 3% +18.9% 6261751 ± 3% stress-ng.mmaprandom.ops_per_sec
> 502294 ± 4% +18.9% 597382 ± 5% stress-ng.time.involuntary_context_switches
> 42134946 ± 4% +15.8% 48772691 ± 3% stress-ng.time.minor_page_faults
> 11559 ± 2% +23.9% 14321 stress-ng.time.percent_of_cpu_this_job_got
> 7227 +23.2% 8905 stress-ng.time.system_time
> 1834057 ± 3% +18.6% 2175623 ± 3% stress-ng.time.voluntary_context_switches
> 109.70 ± 52% +161.8% 287.20 ± 30% perf-c2c.DRAM.local
> 10.88 ± 12% -24.7% 8.19 ± 11% perf-sched.wait_time.max.ms.perf_trace_sched_switch.exit_to_user_mode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe.shmdt
> 5.149e+09 ± 4% -33.2% 3.441e+09 ± 16% cpuidle..time
> 3590818 ± 4% +13.6% 4079292 ± 2% cpuidle..usage
> 89858 ± 4% +12.6% 101156 ± 3% perf-stat.i.context-switches
> 78474 ± 4% +19.4% 93707 ± 4% perf-stat.ps.context-switches
> 40.88 ± 3% -18.9 21.97 ± 51% turbostat.C1%
> 40.77 ± 3% -46.5% 21.83 ± 51% turbostat.CPU%c1
> 41.69 ± 2% -32.1% 28.32 ± 10% vmstat.cpu.id
> 114.21 ± 3% +17.4% 134.07 ± 5% vmstat.procs.r
> 75991 ± 4% +17.6% 89367 ± 4% vmstat.system.cs
> 931.70 ± 15% -21.5% 731.03 ± 6% meminfo.Inactive
> 931.70 ± 15% -21.5% 731.03 ± 6% meminfo.Inactive(file)
> 3323 ± 17% -92.7% 241.56 ± 41% meminfo.ShmemHugePages
> 3167 ± 17% -93.0% 223.03 ± 46% meminfo.ShmemPmdMapped
> 1958 ± 28% -94.4% 109.27 ± 64% numa-meminfo.node0.ShmemHugePages
> 1883 ± 27% -95.2% 90.32 ± 80% numa-meminfo.node0.ShmemPmdMapped
> 1464 ± 25% -90.2% 143.09 ± 73% numa-meminfo.node1.ShmemHugePages
> 1380 ± 26% -90.7% 127.76 ± 65% numa-meminfo.node1.ShmemPmdMapped
> 29299405 ± 3% +16.7% 34179169 ± 3% numa-numastat.node0.local_node
> 29397753 ± 3% +16.7% 34300051 ± 3% numa-numastat.node0.numa_hit
> 29074958 ± 3% +17.0% 34029990 ± 3% numa-numastat.node1.local_node
> 29175702 ± 3% +16.9% 34110422 ± 3% numa-numastat.node1.numa_hit
> 29398448 ± 3% +16.7% 34299565 ± 3% numa-vmstat.node0.numa_hit
> 29300100 ± 3% +16.7% 34178683 ± 3% numa-vmstat.node0.numa_local
> 29175972 ± 3% +16.9% 34110183 ± 3% numa-vmstat.node1.numa_hit
> 29075150 ± 3% +17.0% 34029751 ± 3% numa-vmstat.node1.numa_local
> 40.14 ± 2% -14.1 26.05 ± 11% mpstat.cpu.all.idle%
> 0.33 ± 4% +0.1 0.39 ± 6% mpstat.cpu.all.irq%
> 0.26 ± 6% +0.1 0.32 ± 3% mpstat.cpu.all.soft%
> 57.71 ± 2% +13.8 71.52 ± 4% mpstat.cpu.all.sys%
> 87.62 ± 6% +12.6% 98.66 mpstat.max_utilization_pct
> 22227 +17.6% 26146 sched_debug.cfs_rq:/.avg_vruntime.avg
> 82637 ± 12% -26.0% 61123 ± 15% sched_debug.cfs_rq:/.avg_vruntime.max
> 18397 +16.4% 21416 ± 5% sched_debug.cfs_rq:/.avg_vruntime.min
> 6674 ± 7% -26.8% 4887 ± 8% sched_debug.cfs_rq:/.avg_vruntime.stddev
> 829.95 ± 16% +37.0% 1136 ± 9% sched_debug.cfs_rq:/.util_est.max
> 22226 +17.6% 26146 sched_debug.cfs_rq:/.zero_vruntime.avg
> 82637 ± 12% -26.0% 61123 ± 15% sched_debug.cfs_rq:/.zero_vruntime.max
> 18397 +16.4% 21416 ± 5% sched_debug.cfs_rq:/.zero_vruntime.min
> 6674 ± 7% -26.8% 4887 ± 8% sched_debug.cfs_rq:/.zero_vruntime.stddev
> 142314 ± 12% +56.2% 222249 ± 11% sched_debug.cpu.avg_idle.stddev
> 14910 ± 3% +15.0% 17153 ± 2% sched_debug.cpu.nr_switches.avg
> 58577011 ± 3% +16.8% 68413496 ± 3% proc-vmstat.numa_hit
> 58377918 ± 3% +16.8% 68212181 ± 3% proc-vmstat.numa_local
> 3081605 ± 2% +19.3% 3677408 ± 3% proc-vmstat.pgactivate
> 69802610 ± 3% +17.2% 81798112 ± 3% proc-vmstat.pgalloc_normal
> 6196217 ± 2% +19.9% 7429815 ± 3% proc-vmstat.pgdeactivate
> 42798950 ± 4% +15.5% 49451147 ± 3% proc-vmstat.pgfault
> 68727309 ± 3% +17.4% 80696135 ± 3% proc-vmstat.pgfree
> 627355 ± 3% +19.7% 750783 ± 3% proc-vmstat.pglazyfree
> 4638 ± 3% +20.0% 5565 ± 4% proc-vmstat.thp_collapse_alloc
> 15260 ± 2% +19.2% 18197 ± 3% proc-vmstat.thp_fault_alloc
> 2729 ± 3% +20.6% 3292 ± 4% proc-vmstat.thp_file_mapped
> 1214 ± 3% +20.9% 1468 ± 3% proc-vmstat.thp_split_page
> 1214 ± 3% +20.9% 1468 ± 3% proc-vmstat.thp_swpout_fallback
> 38553874 ± 3% +18.6% 45724835 ± 3% proc-vmstat.unevictable_pgs_culled
> 38620357 ± 3% +18.6% 45806796 ± 3% proc-vmstat.unevictable_pgs_mlocked
> 37844607 ± 3% +18.6% 44877490 ± 3% proc-vmstat.unevictable_pgs_munlocked
> 37775054 ± 3% +18.6% 44792535 ± 3% proc-vmstat.unevictable_pgs_rescued
> 775098 ± 4% +19.8% 928721 ± 4% proc-vmstat.unevictable_pgs_stranded
>
>
>
>
> Disclaimer:
> Results have been estimated based on internal Intel analysis and are provided
> for informational purposes only. Any difference in system hardware or software
> design or configuration may affect actual performance.
>
>
Okay I'll bite, how does the revised wakeup rule for fair_server result in this
improvement? The above is mostly "more CPU time used by fair tasks instead of idle"
right?