autonuma: Fix scan period updating
Subsystems (2)
PATCH RESEND 1/1 autonuma: Fix scan period updating
From: Huang Ying <[email protected]>
From the commit log and comments of commit 37ec97deb3a8 ("sched/numa:
Slow down scan rate if shared faults dominate"), the autonuma scan period should be increased (scanning is slowed down) if the majority of the page accesses are shared with other processes. But in current code, the scan period will be decreased (scanning is speeded up) in that situation.
The commit log and comments make more sense. So this patch fixes the code to make it match the commit log and comments. And this has been verified via tracing the scan period changing and /proc/vmstat numa_pte_updates counter when running a multi-threaded memory accessing program (most memory areas are accessed by multiple threads).
--- kernel/sched/fair.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)
kernel/sched/fair.c +10 -10
@@ -1940,7 +1940,7 @@ static void update_task_scan_period(struct task_struct *p, unsigned long shared, unsigned long private) {unsigned int period_slot; -int lr_ratio, ps_ratio; +int lr_ratio, sp_ratio;int diff;unsigned long remote = p->numa_faults_locality[0 ];
@@ -1971,22 +1971,22 @@ static void update_task_scan_period(struct task_struct *p,*/ period_slot =DIV_ROUND_UP (p->numa_scan_period,NUMA_PERIOD_SLOTS ); lr_ratio = (local *NUMA_PERIOD_SLOTS ) / (local + remote); - ps_ratio = (private *NUMA_PERIOD_SLOTS ) / (private + shared); + sp_ratio = (shared *NUMA_PERIOD_SLOTS ) / (private + shared); -if (ps_ratio >=NUMA_PERIOD_THRESHOLD ) { +if (sp_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are local. There is no need to -* do fast NUMA scanning, since memory is already local. +* Most memory accesses are shared with other tasks. +* There is no point in continuing fast NUMA scanning, +* since other tasks may just move the memory elsewhere. */ -int slot = ps_ratio -NUMA_PERIOD_THRESHOLD ; +int slot = sp_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot) slot =1 ; diff = slot * period_slot; }else if (lr_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are shared with other tasks. -* There is no point in continuing fast NUMA scanning, -* since other tasks may just move the memory elsewhere. +* Most memory accesses are local. There is no need to +* do fast NUMA scanning, since memory is already local. */ int slot = lr_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot)
@@ -1998,7 +1998,7 @@ static void update_task_scan_period(struct task_struct *p,* yet they are not on the local NUMA node. Speed up * NUMA scanning to get the memory moved over. */ -int ratio =max (lr_ratio, ps_ratio); +int ratio =max (lr_ratio, sp_ratio); diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot; }
Signature · 1 line
2.20.1
Quoted 9 lines · Huang, Ying
From: Huang Ying <[email protected]>
From the commit log and comments of commit
37ec97deb3a8("sched/numa:
Slow down scan rate if shared faults dominate"), the autonuma scan period should be increased (scanning is slowed down) if the majority of the page accesses are shared with other processes. But in current code, the scan period will be decreased (scanning is speeded up) in that situation.
Quoted 7 lines
The commit log and comments make more sense. So this patch fixes the code to make it match the commit log and comments. And this has been verified via tracing the scan period changing and
/proc/vmstatnuma_pte_updatescounter when running a multi-threaded memory accessing program (most memory areas are accessed by multiple threads).
Lets split into 4 modes.
More Local and Private Page Accesses:
We definitely want to scan slowly i.e increase the scan window.
More Local and Shared Page Accesses:
We still want to scan slowly because we have consolidated and there is no point in scanning faster. So scan slowly + increase the scan window. (Do remember access on any active node counts as local!!!)
More Remote + Private page Accesses:
Most likely the Private accesses are going to be local accesses.
In the unlikely event of the private accesses not being local, we should scan faster so that the memory and task consolidates.
More Remote + Shared page Accesses: This means the workload has not consolidated and needs to scan faster. So we need to scan faster.
So I would think we should go back to before 37ec97deb3a8.
i.e
int slot = lr_ratio - NUMA_PERIOD_THRESHOLD; if (!slot) slot = 1; diff = slot * period_slot;
No?
Quoted 31 lines
Fixes 37ec97deb3a8 ("sched/numa: Slow down scan rate if shared faults dominate")Signed-off-by "Huang, Ying" <[email protected]>Cc Rik van Riel <[email protected]>Cc Peter Zijlstra (Intel) <[email protected]>Cc Mel Gorman <[email protected]>Cc Ingo Molnar <[email protected]>Cc Andrew Morton <[email protected]>--- kernel/sched/fair.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)kernel/sched/fair.c +1
-2@@ -1940,7 +1940,7 @@ static void update_task_scan_period(struct task_struct *p, unsigned long shared, unsigned long private) {unsigned int period_slot; -int lr_ratio, ps_ratio; +int lr_ratio, sp_ratio;int diff;unsigned long remote = p->numa_faults_locality[0 ];@@ -1971,22 +1971,22 @@ static void update_task_scan_period(struct task_struct *p,*/ period_slot =DIV_ROUND_UP (p->numa_scan_period,NUMA_PERIOD_SLOTS ); lr_ratio = (local *NUMA_PERIOD_SLOTS ) / (local + remote); - ps_ratio = (private *NUMA_PERIOD_SLOTS ) / (private + shared);
kernel/sched/fair.c static void update_task_scan_period(struct task_struct *p, · line 1980
Quoted 10 lines
kernel/sched/fair.c +5
-3+ sp_ratio = (shared *NUMA_PERIOD_SLOTS ) / (private + shared); -if (ps_ratio >=NUMA_PERIOD_THRESHOLD ) { +if (sp_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are local. There is no need to -* do fast NUMA scanning, since memory is already local. +* Most memory accesses are shared with other tasks. +* There is no point in continuing fast NUMA scanning, +* since other tasks may just move the memory elsewhere.
With this change, I would expect that with Shared page accesses, consolidation to take a hit.
kernel/sched/fair.c static void update_task_scan_period(struct task_struct *p, · line 1989
Quoted 13 lines
kernel/sched/fair.c +3
-4*/ -int slot = ps_ratio -NUMA_PERIOD_THRESHOLD ; +int slot = sp_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot) slot =1 ; diff = slot * period_slot; }else if (lr_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are shared with other tasks. -* There is no point in continuing fast NUMA scanning, -* since other tasks may just move the memory elsewhere. +* Most memory accesses are local. There is no need to +* do fast NUMA scanning, since memory is already local.
Comment wise this make sense.
Quoted 15 lines
kernel/sched/fair.c +1
-1*/ int slot = lr_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot)@@ -1998,7 +1998,7 @@ static void update_task_scan_period(struct task_struct *p,* yet they are not on the local NUMA node. Speed up * NUMA scanning to get the memory moved over. */ -int ratio =max (lr_ratio, ps_ratio); +int ratio =max (lr_ratio, sp_ratio); diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot; }Signature · 1 line
2.20.1
Signature · 2 lines
Thanks and Regards Srikar Dronamraju
Hi, Srikar,
Quoted 28 lines · Srikar Dronamraju
Quoted 9 lines · Huang, Ying
From: Huang Ying <[email protected]>
From the commit log and comments of commit
37ec97deb3a8("sched/numa:
Slow down scan rate if shared faults dominate"), the autonuma scan period should be increased (scanning is slowed down) if the majority of the page accesses are shared with other processes. But in current code, the scan period will be decreased (scanning is speeded up) in that situation.Quoted 7 lines
The commit log and comments make more sense. So this patch fixes the code to make it match the commit log and comments. And this has been verified via tracing the scan period changing and
/proc/vmstatnuma_pte_updatescounter when running a multi-threaded memory accessing program (most memory areas are accessed by multiple threads).Lets split into 4 modes.
More Local and Private Page Accesses:
We definitely want to scan slowly i.e increase the scan window.More Local and Shared Page Accesses:
We still want to scan slowly because we have consolidated and there is no point in scanning faster. So scan slowly + increase the scan window. (Do remember access on any active node counts as local!!!)
Quoted 8 lines
More Remote + Private page Accesses:
Most likely the Private accesses are going to be local accesses.In the unlikely event of the private accesses not being local, we should scan faster so that the memory and task consolidates.
More Remote + Shared page Accesses: This means the workload has not consolidated and needs to scan faster. So we need to scan faster.
This sounds reasonable. But
lr_ratio < NUMA_PERIOD_THRESHOLD
doesn't indicate More Remote. If Local = Remote, it is also true. If there are also more Shared, we should slow down the scanning. So, the logic could be
if (lr_ratio >= NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio >= NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS - lr_ratio >= NUMA_PERIOD_THRESHOLD)
speed up scanning
else
slow down scanning
} else
speed up scanning
This follows your idea better?
Best Regards,
Huang, Ying
Quoted 89 lines
So I would think we should go back to before
37ec97deb3a8.i.e
int slot = lr_ratio - NUMA_PERIOD_THRESHOLD; if (!slot) slot = 1; diff = slot * period_slot;No?
Quoted 31 lines
Fixes 37ec97deb3a8 ("sched/numa: Slow down scan rate if shared faults dominate")Signed-off-by "Huang, Ying" <[email protected]>Cc Rik van Riel <[email protected]>Cc Peter Zijlstra (Intel) <[email protected]>Cc Mel Gorman <[email protected]>Cc Ingo Molnar <[email protected]>Cc Andrew Morton <[email protected]>--- kernel/sched/fair.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)kernel/sched/fair.c +1
-2@@ -1940,7 +1940,7 @@ static void update_task_scan_period(struct task_struct *p, unsigned long shared, unsigned long private) {unsigned int period_slot; -int lr_ratio, ps_ratio; +int lr_ratio, sp_ratio;int diff;unsigned long remote = p->numa_faults_locality[0 ];@@ -1971,22 +1971,22 @@ static void update_task_scan_period(struct task_struct *p,*/ period_slot =DIV_ROUND_UP (p->numa_scan_period,NUMA_PERIOD_SLOTS ); lr_ratio = (local *NUMA_PERIOD_SLOTS ) / (local + remote); - ps_ratio = (private *NUMA_PERIOD_SLOTS ) / (private + shared);kernel/sched/fair.c static void update_task_scan_period(struct task_struct *p, · line 1980
Quoted 10 lines
kernel/sched/fair.c +5
-3+ sp_ratio = (shared *NUMA_PERIOD_SLOTS ) / (private + shared); -if (ps_ratio >=NUMA_PERIOD_THRESHOLD ) { +if (sp_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are local. There is no need to -* do fast NUMA scanning, since memory is already local. +* Most memory accesses are shared with other tasks. +* There is no point in continuing fast NUMA scanning, +* since other tasks may just move the memory elsewhere. With this change, I would expect that with Shared page accesses, consolidation to take a hit.
kernel/sched/fair.c static void update_task_scan_period(struct task_struct *p, · line 1989
Quoted 13 lines
kernel/sched/fair.c +3
-4*/ -int slot = ps_ratio -NUMA_PERIOD_THRESHOLD ; +int slot = sp_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot) slot =1 ; diff = slot * period_slot; }else if (lr_ratio >=NUMA_PERIOD_THRESHOLD ) {/* -* Most memory accesses are shared with other tasks. -* There is no point in continuing fast NUMA scanning, -* since other tasks may just move the memory elsewhere. +* Most memory accesses are local. There is no need to +* do fast NUMA scanning, since memory is already local. Comment wise this make sense.
Quoted 15 lines
kernel/sched/fair.c +1
-1*/ int slot = lr_ratio -NUMA_PERIOD_THRESHOLD ;if (!slot)@@ -1998,7 +1998,7 @@ static void update_task_scan_period(struct task_struct *p,* yet they are not on the local NUMA node. Speed up * NUMA scanning to get the memory moved over. */ -int ratio =max (lr_ratio, ps_ratio); +int ratio =max (lr_ratio, sp_ratio); diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot; }Signature · 1 line
2.20.1
Quoted 9 lines · Huang, Ying
Hi, Srikar,
Quoted 7 lines
More Remote + Private page Accesses:
Most likely the Private accesses are going to be local accesses.In the unlikely event of the private accesses not being local, we should scan faster so that the memory and task consolidates.
Quoted 8 lines
Quoted 2 lines
More Remote + Shared page Accesses: This means the workload has not consolidated and needs to scan faster. So we need to scan faster.
This sounds reasonable. But
lr_ratio<NUMA_PERIOD_THRESHOLDdoesn't indicate More Remote. If Local = Remote, it is also true. If
less lr_ratio means more remote.
Quoted 1 line
there are also more Shared, we should slow down the scanning. So, the
Why should we slowing down if there are more remote shared accesses?
Quoted 16 lines
logic could be
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanning
else
slow down scanning
} else
speed up scanningThis follows your idea better?
Best Regards,
Huang, Ying
Signature · 2 lines
Thanks and Regards Srikar Dronamraju
Quoted 18 lines · Srikar Dronamraju
Quoted 16 lines · Huang, Ying
Hi, Srikar,
Quoted 9 lines
More Remote + Private page Accesses:
Most likely the Private accesses are going to be local accesses.In the unlikely event of the private accesses not being local, we should scan faster so that the memory and task consolidates.
More Remote + Shared page Accesses: This means the workload has not consolidated and needs to scan faster. So we need to scan faster.
This sounds reasonable. But
lr_ratio<NUMA_PERIOD_THRESHOLD
Quoted 15 lines
Quoted 1 line
doesn't indicate More Remote. If Local = Remote, it is also true. If
less
lr_ratiomeans more remote.Quoted 1 line
there are also more Shared, we should slow down the scanning. So, the
Why should we slowing down if there are more remote shared accesses?
Quoted 7 lines
logic could be
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanning
Thought about this again. For example, a multi-threads workload runs on a 4-sockets machine, and most memory accesses are shared. The optimal situation will be pseudo-interleaving, that is, spreading memory accesses evenly among 4 NUMA nodes. Where "share" >> "private", and "remote" > "local". And we should slow down scanning to reduce the overhead.
What do you think about this?
Best Regards,
Huang, Ying
Quoted 9 lines
Quoted 9 lines
else
slow down scanning
} else
speed up scanningThis follows your idea better?
Best Regards,
Huang, Ying
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanning
Quoted 8 lines
Thought about this again. For example, a multi-threads workload runs on a 4-sockets machine, and most memory accesses are shared. The optimal situation will be pseudo-interleaving, that is, spreading memory accesses evenly among 4 NUMA nodes. Where "share" >> "private", and "remote" > "local". And we should slow down scanning to reduce the overhead.
What do you think about this?
If all 4 nodes have equal access, then all 4 nodes will be active nodes.
From task_numa_fault()
if (!priv && !local && ng && ng->active_nodes > 1 && numa_is_active_node(cpu_node, ng) && numa_is_active_node(mem_node, ng)) local = 1;
Hence all accesses will be accounted as local. Hence scanning would slow down.
Signature · 2 lines
Thanks and Regards Srikar Dronamraju
Quoted 20 lines · Srikar Dronamraju
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanningQuoted 8 lines
Thought about this again. For example, a multi-threads workload runs on a 4-sockets machine, and most memory accesses are shared. The optimal situation will be pseudo-interleaving, that is, spreading memory accesses evenly among 4 NUMA nodes. Where "share" >> "private", and "remote" > "local". And we should slow down scanning to reduce the overhead.
What do you think about this?
If all 4 nodes have equal access, then all 4 nodes will be active nodes.
From
task_numa_fault()
Quoted 7 lines
if (!priv && !local && ng && ng->active_nodes > 1 && numa_is_active_node(cpu_node, ng) && numa_is_active_node(mem_node, ng)) local = 1;Hence all accesses will be accounted as local. Hence scanning would slow down.
Yes. You are right! Thanks a lot!
There may be another case. For example, a workload with 9 threads runs on a 2-sockets machine, and most memory accesses are shared. 7 threads runs on the node 0 and 2 threads runs on the node 1 based on CPU load balancing. Then the 2 threads on the node 1 will have "share" >> "private" and "remote" >> "local". But it doesn't help to speed up scanning.
Best Regards,
Huang, Ying
Mel Gorman
reviewerQuoted 30 lines · Ying
Quoted 20 lines · Srikar Dronamraju
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanningQuoted 8 lines
Thought about this again. For example, a multi-threads workload runs on a 4-sockets machine, and most memory accesses are shared. The optimal situation will be pseudo-interleaving, that is, spreading memory accesses evenly among 4 NUMA nodes. Where "share" >> "private", and "remote" > "local". And we should slow down scanning to reduce the overhead.
What do you think about this?
If all 4 nodes have equal access, then all 4 nodes will be active nodes.
From
task_numa_fault()Quoted 7 lines
if (!priv && !local && ng && ng->active_nodes > 1 && numa_is_active_node(cpu_node, ng) && numa_is_active_node(mem_node, ng)) local = 1;Hence all accesses will be accounted as local. Hence scanning would slow down.
Quoted 9 lines
Yes. You are right! Thanks a lot!
There may be another case. For example, a workload with 9 threads runs on a 2-sockets machine, and most memory accesses are shared. 7 threads runs on the node 0 and 2 threads runs on the node 1 based on CPU load balancing. Then the 2 threads on the node 1 will have "share" >> "private" and "remote" >> "local". But it doesn't help to speed up scanning.
Ok, so the results from the patch are mostly neutral. There are some small differences in scan rates depending on the workload but it's not universal and the headline performance is sometimes worse. I couldn't find something that would justify the change on its own. I think in the short term -- just fix the comments.
For the shared access consideration, the scan rate is important but so too is the decision on when pseudo interleaving should be used. Both should probably be taken into account when making changes in this area. The current code may not be optimal but it also has not generated bug reports, high CPU usage or obviously bad locality decision in the field. Hence, for this patch or a similar series, it is critical that some workloads are selected that really care about the locality of shared access and evaluate based on that. Initially it was done with a large battery of tests run by different people but some of those people have changed role since and would not be in a position to rerun the tests. There also was the issue that when those were done, NUMA balancing was new so it's comparative baseline was "do nothing at all".
Signature · 2 lines
Mel Gorman SUSE Labs
Quoted 41 lines · Mel Gorman
Quoted 30 lines · Ying
Quoted 20 lines · Srikar Dronamraju
Quoted 7 lines
Quoted 6 lines
Quoted 6 lines
if (
lr_ratio>=NUMA_PERIOD_THRESHOLD)
slow down scanning
else if (sp_ratio>=NUMA_PERIOD_THRESHOLD) {
if (NUMA_PERIOD_SLOTS-lr_ratio>=NUMA_PERIOD_THRESHOLD)
speed up scanningQuoted 8 lines
Thought about this again. For example, a multi-threads workload runs on a 4-sockets machine, and most memory accesses are shared. The optimal situation will be pseudo-interleaving, that is, spreading memory accesses evenly among 4 NUMA nodes. Where "share" >> "private", and "remote" > "local". And we should slow down scanning to reduce the overhead.
What do you think about this?
If all 4 nodes have equal access, then all 4 nodes will be active nodes.
From
task_numa_fault()Quoted 7 lines
if (!priv && !local && ng && ng->active_nodes > 1 && numa_is_active_node(cpu_node, ng) && numa_is_active_node(mem_node, ng)) local = 1;Hence all accesses will be accounted as local. Hence scanning would slow down.
Quoted 9 lines
Yes. You are right! Thanks a lot!
There may be another case. For example, a workload with 9 threads runs on a 2-sockets machine, and most memory accesses are shared. 7 threads runs on the node 0 and 2 threads runs on the node 1 based on CPU load balancing. Then the 2 threads on the node 1 will have "share" >> "private" and "remote" >> "local". But it doesn't help to speed up scanning.
Quoted 4 lines
Ok, so the results from the patch are mostly neutral. There are some small differences in scan rates depending on the workload but it's not universal and the headline performance is sometimes worse. I couldn't find something that would justify the change on its own.
Thanks a lot for your help!
Quoted 1 line
I think in the short term -- just fix the comments.
Then we will change the comments to something like,
"Slow down scanning if most memory accesses are private."
It's hard to be understood. Maybe we just keep the code and comments as it was until we have better understanding.
Quoted 4 lines
For the shared access consideration, the scan rate is important but so too is the decision on when pseudo interleaving should be used. Both should probably be taken into account when making changes in this area. The current code may not be optimal but it also has not generated bug reports,
Quoted 8 lines
high CPU usage or obviously bad locality decision in the field. Hence, for this patch or a similar series, it is critical that some workloads are selected that really care about the locality of shared access and evaluate based on that. Initially it was done with a large battery of tests run by different people but some of those people have changed role since and would not be in a position to rerun the tests. There also was the issue that when those were done, NUMA balancing was new so it's comparative baseline was "do nothing at all".
Yes. I totally agree that we should change the behavior based on testing.
Best Regards,
Huang, Ying