sched/numa: Enhance vma scanning
Subsystems (4)
v3 cover sched/numa: Enhance vma scanning
The patchset proposes one of the enhancements to numa vma scanning
suggested by Mel. This is continuation of [3].
Existing mechanism of scan period involves, scan period derived from per-thread stats. Process Adaptive autoNUMA [1] proposed to gather NUMA fault stats at per-process level to capture aplication behaviour better.
During that course of discussion, Mel proposed several ideas to enhance current numa balancing. One of the suggestion was below
Track what threads access a VMA. The suggestion was to use an unsigned long pid_mask and use the lower bits to tag approximately what threads access a VMA. Skip VMAs that did not trap a fault. This would be approximate because of PID collisions but would reduce scanning of areas the thread is not interested in. The above suggestion intends not to penalize threads that has no interest in the vma, thus reduce scanning overhead.
V3 changes are mostly based on PeterZ comments (details below in changes)
Summary of patchset:
Current patchset implements:
1. Delay the vma scanning logic for newly created VMA's so that additional overhead of scanning is not incurred for short lived tasks (implementation by Mel)
2. Store the information of tasks accessing VMA in 2 windows. It is regularly cleared in (4*sysctl_numa_balancing_scan_delay) interval. The above time is derived from experimenting (Suggested by PeterZ) to balance between frequent clearing vs obsolete access data
3. hash_32 used to encode task index accessing VMA information
4. VMA's acess information is used to skip scanning for the tasks which had not accessed VMA
Things to ponder over:
==========================================
- Improvement to clearing accessing PIDs logic (discussed in-detail in patch3 itself (Done in this patchset by implementing 2 window history)
- Current scan period is not changed in the patchset, so we do see frequent tries to scan. Relaxing scan period dynamically could improve results
further.
[1] sched/numa: Process Adaptive autoNUMA
[2] RFC V1 Link:
https://lore.kernel.org/all/[email protected]/
[3] V2 Link:
https://lore.kernel.org/lkml/[email protected]/
Changes since V2:
patch1:
- Renaming of structure, macro to function,
- Add explanation to heuristics
- Adding more details from result (PeterZ)
Patch2:
- Usage of test and set bit (PeterZ)
- Move storing access PID info to numa_migrate_prep()
- Add a note on fainess among tasks allowed to scan (PeterZ)
Patch3:
- Maintain two windows of access PID information (PeterZ supported implementation and Gave idea to extend to N if needed)
Patch4:
- Apply hash_32 function to track VMA accessing PIDs (PeterZ)
Changes since RFC V1:
- Include Mel's vma scan delay patch
- Change the accessing pid store logic (Thanks Mel)
- Fencing structure / code to NUMA_BALANCING (David, Mel)
- Adding clearing access PID logic (Mel)
- Descriptive change log ( Mike Rapoport)
Results:
Summary: Huge autonuma cost reduction seen in mmtest. Kernbench and dbench improvement is around 5% and huge system time (80%+) improvement from mmtest autonuma.
kernbench
=============
6.1.0-base 6.1.0-patched Amean user-256 22437.65 ( 0.00%) 22622.16 * -0.82%* Amean syst-256 9290.30 ( 0.00%) 8763.85 * 5.67%* Amean elsp-256 159.36 ( 0.00%) 157.44 * 1.20%*
Duration User 67322.16 67876.18 Duration System 27884.89 26306.28
Duration Elapsed 498.95 494.42
Ops NUMA alloc hit 1738904367.00 1738882062.00 Ops NUMA alloc local 1738904104.00 1738881490.00 Ops NUMA base-page range updates 440526.00 272095.00 Ops NUMA PTE updates 440526.00 272095.00 Ops NUMA hint faults 109109.00 55630.00
Ops NUMA hint local faults % 5474.00 196.00
Ops NUMA hint local percent 5.02 0.35
Ops NUMA pages migrated 103400.00 55434.00
Ops AutoNUMA cost 550.59 281.11
autonumabench
===============
6.1.0-base 6.1.0-patched Amean syst-NUMA01 252.55 ( 0.00%) 27.71 * 89.03%* Amean syst-NUMA01_THREADLOCAL 0.20 ( 0.00%) 0.23 * -12.77%* Amean syst-NUMA02 0.91 ( 0.00%) 0.76 * 16.22%* Amean syst-NUMA02_SMT 0.67 ( 0.00%) 0.67 * -1.07%* Amean elsp-NUMA01 269.93 ( 0.00%) 309.44 * -14.64%* Amean elsp-NUMA01_THREADLOCAL 1.05 ( 0.00%) 1.07 * -1.36%* Amean elsp-NUMA02 3.26 ( 0.00%) 3.29 * -0.79%* Amean elsp-NUMA02_SMT 3.73 ( 0.00%) 3.52 * 5.64%*
Duration User 318683.69 330084.06
Duration System 1780.77 206.14
Duration Elapsed 1954.30 2233.06
Ops NUMA alloc hit 62237331.00 49179090.00 Ops NUMA alloc local 62235222.00 49177092.00 Ops NUMA base-page range updates 85303091.00 29242.00 Ops NUMA PTE updates 85303091.00 29242.00 Ops NUMA hint faults 87457481.00 8302.00 Ops NUMA hint local faults % 66665145.00 6064.00
Ops NUMA hint local percent 76.23 73.04
Ops NUMA pages migrated 9348511.00 2232.00
Ops AutoNUMA cost 438062.15 41.76
dbench
========
dbench -t 90 <nproc>
Throughput
#clients base patched %improvement
1 842.655 MB/sec 922.305 MB/sec 9.45
16 5062.82 MB/sec 5079.85 MB/sec 0.34 64 9408.81 MB/sec 9980.89 MB/sec 6.08 256 7076.59 MB/sec 7590.76 MB/sec 7.26
Mel Gorman (1):
sched/numa: Apply the scan delay to every new vma
Raghavendra K T (3):
sched/numa: Enhance vma scanning logic sched/numa: implement access PID reset logic sched/numa: Use hash_32 to mix up PIDs accessing VMA
include/linux/mm.h | 30 +++++++++++++++++++++ include/linux/mm_types.h | 9 +++++++ kernel/fork.c | 2 ++ kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++++++ mm/memory.c | 3 +++ 5 files changed, 101 insertions(+)
Signature · 1 line
2.34.1
PATCH v3 1/4 sched/numa: Apply the scan delay to every new vma
From: Mel Gorman <[email protected]>
Currently whenever a new task is created we wait for sysctl_numa_balancing_scan_delay to avoid unnessary scanning overhead. Extend the same logic to new or very short-lived VMAs.
(Raghavendra: Add initialization in vm_area_dup())
--- include/linux/mm.h | 16 ++++++++++++++++ include/linux/mm_types.h | 7 +++++++ kernel/fork.c | 2 ++ kernel/sched/fair.c | 19 +++++++++++++++++++ 4 files changed, 44 insertions(+)
include/linux/mm.h +16 -0
@@ -29,6 +29,7 @@#include <linux/pgtable.h> #include <linux/kasan.h> #include <linux/memremap.h> +#include <linux/slab.h> struct mempolicy ;struct anon_vma ;
@@ -611,6 +612,20 @@ struct vm_operations_struct { unsigned long addr); }; +#ifdef CONFIG_NUMA_BALANCING +static inline void vma_numab_state_init (struct vm_area_struct *vma) +{ + vma->numab_state =NULL ; +} +static inline void vma_numab_state_free (struct vm_area_struct *vma) +{ +kfree (vma->numab_state); +} +#else +static inline void vma_numab_state_init (struct vm_area_struct *vma) {} +static inline void vma_numab_state_free (struct vm_area_struct *vma) {} +#endif /* CONFIG_NUMA_BALANCING */ +static inline void vma_init (struct vm_area_struct *vma,struct mm_struct *mm) {static const struct vm_operations_struct dummy_vm_ops = {};
@@ -619,6 +634,7 @@ static inline void vma_init(struct vm_area_struct *vma, struct mm_struct *mm) vma->vm_mm = mm; vma->vm_ops = &dummy_vm_ops;INIT_LIST_HEAD (&vma->anon_vma_chain); +vma_numab_state_init (vma); }static inline void vma_set_anonymous (struct vm_area_struct *vma)
include/linux/mm_types.h +7 -0
@@ -435,6 +435,10 @@ struct anon_vma_name {char name[]; }; +struct vma_numab_state { +unsigned long next_scan; +}; +/* * This struct describes a virtual memory area. There is one of these * per VM-area/task. A VM area is any part of the process virtual memory
@@ -504,6 +508,9 @@ struct vm_area_struct {#endif #ifdef CONFIG_NUMA struct mempolicy *vm_policy;/* NUMA policy for the VMA */ +#endif +#ifdef CONFIG_NUMA_BALANCING +struct vma_numab_state *numab_state;/* NUMA Balancing state */ #endif struct vm_userfaultfd_ctx vm_userfaultfd_ctx; } __randomize_layout;
kernel/fork.c +2 -0
@@ -474,6 +474,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)*/ *new =data_race (*orig);INIT_LIST_HEAD (&new->anon_vma_chain); +vma_numab_state_init (new);dup_anon_vma_name (orig, new); }return new;
@@ -481,6 +482,7 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)void vm_area_free (struct vm_area_struct *vma) { +vma_numab_state_free (vma);free_anon_vma_name (vma);kmem_cache_free (vm_area_cachep, vma); }
kernel/sched/fair.c +19 -0
@@ -3015,6 +3015,25 @@ static void task_numa_work(struct callback_head *work)if (!vma_is_accessible (vma))continue ; +/* Initialise new per-VMA NUMAB state. */ +if (!vma->numab_state) { + vma->numab_state =kzalloc (sizeof (struct vma_numab_state ), +GFP_KERNEL ); +if (!vma->numab_state) +continue ; + + vma->numab_state->next_scan = now + +msecs_to_jiffies (sysctl_numa_balancing_scan_delay); + } + +/* +* Scanning the VMA's of short lived tasks add more overhead. So +* delay the scan for new VMAs. +*/ +if (mm->numa_scan_seq &&time_before (jiffies, + vma->numab_state->next_scan)) +continue ; + do { start = max(start, vma->vm_start); end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
Signature · 1 line
2.34.1
PATCH v3 2/4 sched/numa: Enhance vma scanning logic
In reply to Raghavendra K T During the Numa scanning make sure only relevant vmas of the
tasks are scanned.
Before:
All the tasks of a process participate in scanning the vma
even if they do not access vma in it's lifespan.
Now:
Except cases of first few unconditional scans, if a process do
not touch vma (exluding false positive cases of PID collisions) tasks no longer scan all vma
Logic used:
1) 6 bits of PID used to mark active bit in vma numab status during fault to remember PIDs accessing vma. (Thanks Mel)
2) Subsequently in scan path, vma scanning is skipped if current PID had not accessed vma.
3) First two times we do allow unconditional scan to preserve earlier behaviour of scanning.
Acknowledgement to Bharata B Rao <[email protected]> for initial patch to store pid information and Peter Zijlstra <[email protected]> (Usage of test and set bit)
--- include/linux/mm.h | 14 ++++++++++++++ include/linux/mm_types.h | 1 + kernel/sched/fair.c | 19 +++++++++++++++++++ mm/memory.c | 3 +++ 4 files changed, 37 insertions(+)
include/linux/mm.h +14 -0
@@ -1388,6 +1388,16 @@ static inline int xchg_page_access_time(struct page *page, int time) last_time =page_cpupid_xchg_last (page, time >>PAGE_ACCESS_TIME_BUCKETS ); return last_time << PAGE_ACCESS_TIME_BUCKETS; } + +static inline void vma_set_access_pid_bit (struct vm_area_struct *vma) +{ +unsigned int pid_bit; + + pid_bit = current->pid %BITS_PER_LONG ; +if (vma->numab_state && !test_bit (pid_bit, &vma->numab_state->access_pids)) { +__set_bit (pid_bit, &vma->numab_state->access_pids); + } +}#else /* !CONFIG_NUMA_BALANCING */ static inline int page_cpupid_xchg_last (struct page *page,int cpupid) {
@@ -1437,6 +1447,10 @@ static inline bool cpupid_match_pid(struct task_struct *task, int cpupid) {return false; } + +static inline void vma_set_access_pid_bit (struct vm_area_struct *vma) +{ +}#endif /* CONFIG_NUMA_BALANCING */ #if defined(CONFIG_KASAN_SW_TAGS ) || defined(CONFIG_KASAN_HW_TAGS )
include/linux/mm_types.h +1 -0
@@ -437,6 +437,7 @@ struct anon_vma_name {struct vma_numab_state {unsigned long next_scan; +unsigned long access_pids; };/*
kernel/sched/fair.c +19 -0
@@ -2916,6 +2916,21 @@ static void reset_ptenuma_scan(struct task_struct *p) p->mm->numa_scan_offset =0 ; } +static bool vma_is_accessed (struct vm_area_struct *vma) +{ +/* +* Allow unconditional access first two times, so that all the (pages) +* of VMAs get prot_none fault introduced irrespective of accesses. +* This is also done to avoid any side effect of task scanning +* amplifying the unfairness of disjoint set of VMAs' access. +*/ +if (READ_ONCE (current->mm->numa_scan_seq) <2 ) +return true; + +return test_bit (current->pid %BITS_PER_LONG , + &vma->numab_state->access_pids); +} +/* * The expensive part of numa migration is done from task_work context. * Triggered from task_tick_numa().
@@ -3034,6 +3049,10 @@ static void task_numa_work(struct callback_head *work) vma->numab_state->next_scan))continue ; +/* Do not scan the VMA if task has not accessed */ +if (!vma_is_accessed (vma)) +continue ; + do { start = max(start, vma->vm_start); end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
mm/memory.c +3 -0
@@ -4698,6 +4698,9 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma, {get_page (page); +/* Record the current PID acceesing VMA */ +vma_set_access_pid_bit (vma); +count_vm_numa_event (NUMA_HINT_FAULTS );if (page_nid ==numa_node_id ()) {count_vm_numa_event (NUMA_HINT_FAULTS_LOCAL );
Signature · 1 line
2.34.1
PATCH v3 4/4 sched/numa: Use hash_32 to mix up PIDs accessing VMA
In reply to Raghavendra K Tbefore: last 6 bits of PID is used as index to store information about tasks accessing VMA's.
after: hash_32 is used to take of cases where tasks are created over a period of time, and thus improve collision probability.
Result:
The patch series overall improving autonuma cost by a huge margin.
Kernbench anbd dbench showed around 5% improvement and system time in mmtest autonuma showed 80% improvement
--- include/linux/mm.h | 2 +- kernel/sched/fair.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
include/linux/mm.h +1 -1
@@ -1393,7 +1393,7 @@ static inline void vma_set_access_pid_bit(struct vm_area_struct *vma) {unsigned int pid_bit; - pid_bit = current->pid %BITS_PER_LONG ; + pid_bit =hash_32 (current->pid,ilog2 (BITS_PER_LONG ));if (vma->numab_state && !test_bit (pid_bit, &vma->numab_state->access_pids[1 ])) {__set_bit (pid_bit, &vma->numab_state->access_pids[1 ]); }
kernel/sched/fair.c +1 -1
@@ -2929,7 +2929,7 @@ static bool vma_is_accessed(struct vm_area_struct *vma)return true; pids = vma->numab_state->access_pids[0 ] | vma->numab_state->access_pids[1 ]; -return test_bit (current->pid %BITS_PER_LONG , &pids); +return test_bit (hash_32 (current->pid,ilog2 (BITS_PER_LONG )), &pids); }#define VMA_PID_RESET_PERIOD (4 * sysctl_numa_balancing_scan_delay)
Signature · 1 line
2.34.1
PATCH v3 3/4 sched/numa: implement access PID reset logic
In reply to Raghavendra K T This helps to ensure, only recently accessed PIDs scan the
VMAs.
Current implementation: (idea supported by PeterZ)
1. Accessing PID information is maintained in two windows. access_pids[1] being newest.
2. Reset old access PID info i.e. access_pid[0] every
(4 * sysctl_numa_balancing_scan_delay) interval after initial scan delay period expires.
The above interval seemed to be experimentally optimum since it avoids frequent reset of access info as well as helps clearing the old access info regularly.
The reset logic is implemented in scan path.
--- include/linux/mm.h | 4 ++-- include/linux/mm_types.h | 3 ++- kernel/sched/fair.c | 23 +++++++++++++++++++++-- 3 files changed, 25 insertions(+), 5 deletions(-)
include/linux/mm.h +2 -2
@@ -1394,8 +1394,8 @@ static inline void vma_set_access_pid_bit(struct vm_area_struct *vma)unsigned int pid_bit; pid_bit = current->pid %BITS_PER_LONG ; -if (vma->numab_state && !test_bit (pid_bit, &vma->numab_state->access_pids)) { -__set_bit (pid_bit, &vma->numab_state->access_pids); +if (vma->numab_state && !test_bit (pid_bit, &vma->numab_state->access_pids[1 ])) { +__set_bit (pid_bit, &vma->numab_state->access_pids[1 ]); } }#else /* !CONFIG_NUMA_BALANCING */
include/linux/mm_types.h +2 -1
@@ -437,7 +437,8 @@ struct anon_vma_name {struct vma_numab_state {unsigned long next_scan; -unsigned long access_pids; +unsigned long next_pid_reset; +unsigned long access_pids[2 ]; };/*
kernel/sched/fair.c +21 -2
@@ -2918,6 +2918,7 @@ static void reset_ptenuma_scan(struct task_struct *p)static bool vma_is_accessed (struct vm_area_struct *vma) { +unsigned long pids;/* * Allow unconditional access first two times, so that all the (pages) * of VMAs get prot_none fault introduced irrespective of accesses.
@@ -2927,10 +2928,12 @@ static bool vma_is_accessed(struct vm_area_struct *vma)if (READ_ONCE (current->mm->numa_scan_seq) <2 )return true; -return test_bit (current->pid %BITS_PER_LONG , - &vma->numab_state->access_pids); + pids = vma->numab_state->access_pids[0 ] | vma->numab_state->access_pids[1 ]; +return test_bit (current->pid %BITS_PER_LONG , &pids); } +#define VMA_PID_RESET_PERIOD (4 * sysctl_numa_balancing_scan_delay) +/* * The expensive part of numa migration is done from task_work context. * Triggered from task_tick_numa().
@@ -3039,6 +3042,10 @@ static void task_numa_work(struct callback_head *work) vma->numab_state->next_scan = now +msecs_to_jiffies (sysctl_numa_balancing_scan_delay); + +/* Reset happens after 4 times scan delay of scan start */ + vma->numab_state->next_pid_reset = vma->numab_state->next_scan + +msecs_to_jiffies (VMA_PID_RESET_PERIOD ); }/*
@@ -3053,6 +3060,18 @@ static void task_numa_work(struct callback_head *work)if (!vma_is_accessed (vma))continue ; +/* +* RESET access PIDs regularly for old VMAs. Resetting after checking +* vma for recent access to avoid clearing PID info before access.. +*/ +if (mm->numa_scan_seq && +time_after (jiffies, vma->numab_state->next_pid_reset)) { + vma->numab_state->next_pid_reset = vma->numab_state->next_pid_reset + +msecs_to_jiffies (VMA_PID_RESET_PERIOD ); + vma->numab_state->access_pids[0 ] =READ_ONCE (vma->numab_state->access_pids[1 ]); + vma->numab_state->access_pids[1 ] =0 ; + } + do { start = max(start, vma->vm_start); end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
Signature · 1 line
2.34.1
Andrew Morton
maintainerQuoted 10 lines · On Tue, 28 Feb 2023 10:20:18 +0530 Raghavendra K T <[email protected]>…
The patchset proposes one of the enhancements to numa vma scanning
suggested by Mel. This is continuation of [3]....
include/linux/mm.h | 30 +++++++++++++++++++++ include/linux/mm_types.h | 9 +++++++ kernel/fork.c | 2 ++ kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++++++ mm/memory.c | 3 +++
It's unclear (to me) which tree would normally carry these.
But there are significant textual conflicts with the "Per-VMA locks" patchset, and there might be functional issues as well. So mm.git would be the better choice.
Please can you redo and retest against tomorrow's mm-unstable branch
(git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm)? Hopefully the sched developers can take a look and provide feedback.
Thanks.
Quoted 15 lines · Andrew Morton
Quoted 10 lines · On Tue, 28 Feb 2023 10:20:18 +0530 Raghavendra K T <[email protected]>…
The patchset proposes one of the enhancements to numa vma scanning
suggested by Mel. This is continuation of [3]....
include/linux/mm.h | 30 +++++++++++++++++++++ include/linux/mm_types.h | 9 +++++++ kernel/fork.c | 2 ++ kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++++++ mm/memory.c | 3 +++It's unclear (to me) which tree would normally carry these.
Quoted 8 lines
But there are significant textual conflicts with the "Per-VMA locks" patchset, and there might be functional issues as well. So mm.git would be the better choice.
Please can you redo and retest against tomorrow's mm-unstable branch
(git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm)? Hopefully the sched developers can take a look and provide feedback.
Thank you Andrew. Sure will do that.
Quoted 20 lines · Raghavendra K T
Quoted 19 lines · Andrew Morton
Quoted 10 lines · On Tue, 28 Feb 2023 10:20:18 +0530 Raghavendra K T <[email protected]>…
The patchset proposes one of the enhancements to numa vma scanning suggested by Mel. This is continuation of [3].
...
include/linux/mm.h | 30 +++++++++++++++++++++
include/linux/mm_types.h| 9 +++++++
kernel/fork.c | 2 ++
kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++++++ mm/memory.c | 3 +++It's unclear (to me) which tree would normally carry these.
But there are significant textual conflicts with the "Per-VMA locks" patchset, and there might be functional issues as well. So mm.git would be the better choice.
Quoted 7 lines
Quoted 4 lines
Please can you redo and retest against tomorrow's mm-unstable branch
(git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm)?Hopefully the sched developers can take a look and provide feedback.Thank you Andrew. Sure will do that.
Thanks again, Sent rebased patches,
Just to record, so that new discussion can happen in new posting