[v1,0/4] mm/ksm: Add ksm advisor

  What is the KSM advisor?
=========================
The ksm advisor automatically manages the pages_to_scan setting to
achieve a target scan time. The target scan time defines how many seconds
it should take to scan all the candidate KSM pages. In other words the
pages_to_scan rate is changed by the advisor to achieve the target scan
time.

Why do we need a KSM advisor?
==============================
The number of candidate pages for KSM is dynamic. It can often be observed
that during the startup of an application more candidate pages need to be
processed. Without an advisor the pages_to_scan parameter needs to be
sized for the maximum number of candidate pages. With the scan time
advisor the pages_to_scan parameter based can be changed based on demand.

Algorithm
==========
The algorithm calculates the change value based on the target scan time
and the previous scan time. To avoid pertubations an exponentially
weighted moving average is applied.

The algorithm has a max and min
value to:
- guarantee responsiveness to changes
- to avoid to spend too much CPU

Parameters to influence the KSM scan advisor
=============================================
The respective parameters are:
- ksm_advisor_mode
  0: None (default), 1: scan time advisor
- ksm_advisor_target_scan_time
  how many seconds a scan should of all candidate pages take
- ksm_advisor_min_pages
  minimum value for pages_to_scan per batch
- ksm_advisor_max_pages
  maximum value for pages_to_scan per batch

The parameters are exposed as knobs in /sys/kernel/mm/ksm.
By default the scan time advisor is disabled.

Currently there are two advisors:
- none and
- scan time.

Resource savings
=================
Tests with various workloads have shown considerable CPU savings. Most
of the workloads I have investigated have more candidate pages during
startup, once the workload is stable in terms of memory, the number of
candidate pages is reduced. Without the advisor, the pages_to_scan needs
to be sized for the maximum number of candidate pages. So having this
advisor definitely helps in reducing CPU consumption.

For the instagram workload, the advisor achieves a 25% CPU reduction.
Once the memory is stable, the pages_to_scan parameter gets reduced to
about 40% of its max value.

The new advisor works especially well if the smart scan feature is also
enabled.

How is defining a target scan time better?
===========================================
For an administrator it is more logical to set a target scan time.. The
administrator can determine how many pages are scanned on each scan.
Therefore setting a target scan time makes more sense.

In addition the administrator might have a good idea about the
memory sizing of its respective workloads.

The pages_to_scan parameter is per batch. But the administrator has
no insight how many batches are needed per scan.

Tracing
=======
A new tracing event has been added for the scan time advisor. The new
trace event is called ksm_advisor. It reports the scan time and the new
pages_to_scan setting.

Other approaches
=================

Approach 1: Adapt pages_to_scan after processing each batch. If KSM
  merges pages, increase the scan rate, if less KSM pages, reduce the
  the pages_to_scan rate. This doesn't work too well. While it increases
  the pages_to_scan for a short period, but generally it ends up with a
  too low pages_to_scan rate.

Approach 2: Adapt pages_to_scan after each scan. The problem with that
  approach is that the calculated scan rate tends to be high. The more
  aggressive KSM scans, the more pages it can de-duplicate.

There have been earlier attempts at an advisor:
  propose auto-run mode of ksm and its tests
  (https://marc.info/?l=linux-mm&m=166029880214485&w=2)



Stefan Roesch (4):
  mm/ksm: add ksm advisor
  mm/ksm: add sysfs knobs for advisor
  mm/ksm: add tracepoint for ksm advisor
  mm/ksm: document ksm advisor and its sysfs knobs

 Documentation/admin-guide/mm/ksm.rst |  45 +++++
 include/trace/events/ksm.h           |  28 ++++
 mm/ksm.c                             | 242 ++++++++++++++++++++++++++-
 3 files changed, 314 insertions(+), 1 deletion(-)


base-commit: 12d04a7bf0da67321229d2bc8b1a7074d65415a9
  

On 04.10.23 21:02, Stefan Roesch wrote:
> What is the KSM advisor?
> =========================
> The ksm advisor automatically manages the pages_to_scan setting to
> achieve a target scan time. The target scan time defines how many seconds
> it should take to scan all the candidate KSM pages. In other words the
> pages_to_scan rate is changed by the advisor to achieve the target scan
> time.
> 
> Why do we need a KSM advisor?
> ==============================
> The number of candidate pages for KSM is dynamic. It can often be observed
> that during the startup of an application more candidate pages need to be
> processed. Without an advisor the pages_to_scan parameter needs to be
> sized for the maximum number of candidate pages. With the scan time
> advisor the pages_to_scan parameter based can be changed based on demand.
> 
> Algorithm
> ==========
> The algorithm calculates the change value based on the target scan time
> and the previous scan time. To avoid pertubations an exponentially
> weighted moving average is applied.
> 
> The algorithm has a max and min
> value to:
> - guarantee responsiveness to changes
> - to avoid to spend too much CPU
> 
> Parameters to influence the KSM scan advisor
> =============================================
> The respective parameters are:
> - ksm_advisor_mode
>    0: None (default), 1: scan time advisor
> - ksm_advisor_target_scan_time
>    how many seconds a scan should of all candidate pages take
> - ksm_advisor_min_pages
>    minimum value for pages_to_scan per batch
> - ksm_advisor_max_pages
>    maximum value for pages_to_scan per batch
> 
> The parameters are exposed as knobs in /sys/kernel/mm/ksm.
> By default the scan time advisor is disabled.

What would be the main reason to not have this enabled as default?

IIUC, it is kind-of an auto-tuning of pages_to_scan. Would "auto-tuning" 
describe it better than "advisor" ?

[...]

> How is defining a target scan time better?
> ===========================================
> For an administrator it is more logical to set a target scan time.. The
> administrator can determine how many pages are scanned on each scan.
> Therefore setting a target scan time makes more sense.
> 
> In addition the administrator might have a good idea about the
> memory sizing of its respective workloads.

Is there any way you could imagine where we could have this just do 
something reasonable without any user input? IOW, true auto-tuning?

I read above:
 > - guarantee responsiveness to changes
 > - to avoid to spend too much CPU

whereby both things are accountable/measurable to use that as the input 
for auto-tuning?


I just had a family NMI, so my todo list is quite lengthy. Hoping I cna 
take a closer look next week.
  

David Hildenbrand <david@redhat.com> writes:

> On 04.10.23 21:02, Stefan Roesch wrote:
>> What is the KSM advisor?
>> =========================
>> The ksm advisor automatically manages the pages_to_scan setting to
>> achieve a target scan time. The target scan time defines how many seconds
>> it should take to scan all the candidate KSM pages. In other words the
>> pages_to_scan rate is changed by the advisor to achieve the target scan
>> time.
>> Why do we need a KSM advisor?
>> ==============================
>> The number of candidate pages for KSM is dynamic. It can often be observed
>> that during the startup of an application more candidate pages need to be
>> processed. Without an advisor the pages_to_scan parameter needs to be
>> sized for the maximum number of candidate pages. With the scan time
>> advisor the pages_to_scan parameter based can be changed based on demand.
>> Algorithm
>> ==========
>> The algorithm calculates the change value based on the target scan time
>> and the previous scan time. To avoid pertubations an exponentially
>> weighted moving average is applied.
>> The algorithm has a max and min
>> value to:
>> - guarantee responsiveness to changes
>> - to avoid to spend too much CPU
>> Parameters to influence the KSM scan advisor
>> =============================================
>> The respective parameters are:
>> - ksm_advisor_mode
>>    0: None (default), 1: scan time advisor
>> - ksm_advisor_target_scan_time
>>    how many seconds a scan should of all candidate pages take
>> - ksm_advisor_min_pages
>>    minimum value for pages_to_scan per batch
>> - ksm_advisor_max_pages
>>    maximum value for pages_to_scan per batch
>> The parameters are exposed as knobs in /sys/kernel/mm/ksm.
>> By default the scan time advisor is disabled.
>
> What would be the main reason to not have this enabled as default?
>
There might be already exisiting users which directly set pages_to_scan
and tuned the KSM settings accordingly, as the default setting of 100 for
pages_to_scan is too low for typical workloads.

> IIUC, it is kind-of an auto-tuning of pages_to_scan. Would "auto-tuning"
> describe it better than "advisor" ?
>
> [...]
>

I'm fine with auto-tune. I was also thinking about that name, but I
chose advisor, its a bit less strong and it needs input from the user.

>> How is defining a target scan time better?
>> ===========================================
>> For an administrator it is more logical to set a target scan time.. The
>> administrator can determine how many pages are scanned on each scan.
>> Therefore setting a target scan time makes more sense.
>> In addition the administrator might have a good idea about the
>> memory sizing of its respective workloads.
>
> Is there any way you could imagine where we could have this just do something
> reasonable without any user input? IOW, true auto-tuning?
>

True auto-tuning might be difficult as users might want to be able to
choose how aggressive KSM is. Some might want it to be as aggressive as
possible to get the maximum de-duplication rate. Others might want a
more balanced approach that takes CPU-consumption into consideration.

I guess it depends if you are memory-bound, cpu-bound or both.

> I read above:
>> - guarantee responsiveness to changes
>> - to avoid to spend too much CPU
>
> whereby both things are accountable/measurable to use that as the input for
> auto-tuning?
>
I'm not sure a true auto-tuning can be achieved. I think we need
some input from the user
- How much resources to consume
- How fast memory changes or how stable memory is
  (this we might be able to detect)

>
>
> I just had a family NMI, so my todo list is quite lengthy. Hoping I cna take a
> closer look next week.
  

On 06.10.23 18:17, Stefan Roesch wrote:
> 
> David Hildenbrand <david@redhat.com> writes:
> 
>> On 04.10.23 21:02, Stefan Roesch wrote:
>>> What is the KSM advisor?
>>> =========================
>>> The ksm advisor automatically manages the pages_to_scan setting to
>>> achieve a target scan time. The target scan time defines how many seconds
>>> it should take to scan all the candidate KSM pages. In other words the
>>> pages_to_scan rate is changed by the advisor to achieve the target scan
>>> time.
>>> Why do we need a KSM advisor?
>>> ==============================
>>> The number of candidate pages for KSM is dynamic. It can often be observed
>>> that during the startup of an application more candidate pages need to be
>>> processed. Without an advisor the pages_to_scan parameter needs to be
>>> sized for the maximum number of candidate pages. With the scan time
>>> advisor the pages_to_scan parameter based can be changed based on demand.
>>> Algorithm
>>> ==========
>>> The algorithm calculates the change value based on the target scan time
>>> and the previous scan time. To avoid pertubations an exponentially
>>> weighted moving average is applied.
>>> The algorithm has a max and min
>>> value to:
>>> - guarantee responsiveness to changes
>>> - to avoid to spend too much CPU
>>> Parameters to influence the KSM scan advisor
>>> =============================================
>>> The respective parameters are:
>>> - ksm_advisor_mode
>>>     0: None (default), 1: scan time advisor
>>> - ksm_advisor_target_scan_time
>>>     how many seconds a scan should of all candidate pages take
>>> - ksm_advisor_min_pages
>>>     minimum value for pages_to_scan per batch
>>> - ksm_advisor_max_pages
>>>     maximum value for pages_to_scan per batch
>>> The parameters are exposed as knobs in /sys/kernel/mm/ksm.
>>> By default the scan time advisor is disabled.
>>
>> What would be the main reason to not have this enabled as default?
>>
> There might be already exisiting users which directly set pages_to_scan
> and tuned the KSM settings accordingly, as the default setting of 100 for
> pages_to_scan is too low for typical workloads.

Good point.

> 
>> IIUC, it is kind-of an auto-tuning of pages_to_scan. Would "auto-tuning"
>> describe it better than "advisor" ?
>>
>> [...]
>>
> 
> I'm fine with auto-tune. I was also thinking about that name, but I
> chose advisor, its a bit less strong and it needs input from the user.
> 

I'm not a native speaker, but "adviser" to me implies that no action is 
taken, only advises are given :) But again, no native speaker.

>>> How is defining a target scan time better?
>>> ===========================================
>>> For an administrator it is more logical to set a target scan time.. The
>>> administrator can determine how many pages are scanned on each scan.
>>> Therefore setting a target scan time makes more sense.
>>> In addition the administrator might have a good idea about the
>>> memory sizing of its respective workloads.
>>
>> Is there any way you could imagine where we could have this just do something
>> reasonable without any user input? IOW, true auto-tuning?
>>
> 
> True auto-tuning might be difficult as users might want to be able to
> choose how aggressive KSM is. Some might want it to be as aggressive as
> possible to get the maximum de-duplication rate. Others might want a
> more balanced approach that takes CPU-consumption into consideration.
> 
> I guess it depends if you are memory-bound, cpu-bound or both.

Agreed, more below.

> 
>> I read above:
>>> - guarantee responsiveness to changes
>>> - to avoid to spend too much CPU
>>
>> whereby both things are accountable/measurable to use that as the input for
>> auto-tuning?
>>
> I'm not sure a true auto-tuning can be achieved. I think we need
> some input from the user
> - How much resources to consume
> - How fast memory changes or how stable memory is
>    (this we might be able to detect)

Setting the pages_to_scan is a bit mystical. Setting upper/lower 
pages_to_scan bounds is similarly mystical, and highly workload dependent.

So I agree that a better abstraction to automatically tune the scanning 
is reasonable. I wonder if we can let the user give better inputs that 
are less workload dependent.

For example, do we need min/max values for pages_to_scan, or can we 
replace it by something better to the auto-tuning algorithm?

IMHO "target scan time" goes into the right direction, but it can still 
be fairly workload dependent. Maybe a "max CPU consumption" or sth. like 
that would similarly help to limit CPU waste, and it could be fairly 
workload dependent.
  

David Hildenbrand <david@redhat.com> writes:

> On 06.10.23 18:17, Stefan Roesch wrote:
>> David Hildenbrand <david@redhat.com> writes:
>>
>>> On 04.10.23 21:02, Stefan Roesch wrote:
>>>> What is the KSM advisor?
>>>> =========================
>>>> The ksm advisor automatically manages the pages_to_scan setting to
>>>> achieve a target scan time. The target scan time defines how many seconds
>>>> it should take to scan all the candidate KSM pages. In other words the
>>>> pages_to_scan rate is changed by the advisor to achieve the target scan
>>>> time.
>>>> Why do we need a KSM advisor?
>>>> ==============================
>>>> The number of candidate pages for KSM is dynamic. It can often be observed
>>>> that during the startup of an application more candidate pages need to be
>>>> processed. Without an advisor the pages_to_scan parameter needs to be
>>>> sized for the maximum number of candidate pages. With the scan time
>>>> advisor the pages_to_scan parameter based can be changed based on demand.
>>>> Algorithm
>>>> ==========
>>>> The algorithm calculates the change value based on the target scan time
>>>> and the previous scan time. To avoid pertubations an exponentially
>>>> weighted moving average is applied.
>>>> The algorithm has a max and min
>>>> value to:
>>>> - guarantee responsiveness to changes
>>>> - to avoid to spend too much CPU
>>>> Parameters to influence the KSM scan advisor
>>>> =============================================
>>>> The respective parameters are:
>>>> - ksm_advisor_mode
>>>>     0: None (default), 1: scan time advisor
>>>> - ksm_advisor_target_scan_time
>>>>     how many seconds a scan should of all candidate pages take
>>>> - ksm_advisor_min_pages
>>>>     minimum value for pages_to_scan per batch
>>>> - ksm_advisor_max_pages
>>>>     maximum value for pages_to_scan per batch
>>>> The parameters are exposed as knobs in /sys/kernel/mm/ksm.
>>>> By default the scan time advisor is disabled.
>>>
>>> What would be the main reason to not have this enabled as default?
>>>
>> There might be already exisiting users which directly set pages_to_scan
>> and tuned the KSM settings accordingly, as the default setting of 100 for
>> pages_to_scan is too low for typical workloads.
>
> Good point.
>
>>
>>> IIUC, it is kind-of an auto-tuning of pages_to_scan. Would "auto-tuning"
>>> describe it better than "advisor" ?
>>>
>>> [...]
>>>
>> I'm fine with auto-tune. I was also thinking about that name, but I
>> chose advisor, its a bit less strong and it needs input from the user.
>>
>
> I'm not a native speaker, but "adviser" to me implies that no action is taken,
> only advises are given :) But again, no native speaker.
>
>>>> How is defining a target scan time better?
>>>> ===========================================
>>>> For an administrator it is more logical to set a target scan time.. The
>>>> administrator can determine how many pages are scanned on each scan.
>>>> Therefore setting a target scan time makes more sense.
>>>> In addition the administrator might have a good idea about the
>>>> memory sizing of its respective workloads.
>>>
>>> Is there any way you could imagine where we could have this just do something
>>> reasonable without any user input? IOW, true auto-tuning?
>>>
>> True auto-tuning might be difficult as users might want to be able to
>> choose how aggressive KSM is. Some might want it to be as aggressive as
>> possible to get the maximum de-duplication rate. Others might want a
>> more balanced approach that takes CPU-consumption into consideration.
>> I guess it depends if you are memory-bound, cpu-bound or both.
>
> Agreed, more below.
>
>>
>>> I read above:
>>>> - guarantee responsiveness to changes
>>>> - to avoid to spend too much CPU
>>>
>>> whereby both things are accountable/measurable to use that as the input for
>>> auto-tuning?
>>>
>> I'm not sure a true auto-tuning can be achieved. I think we need
>> some input from the user
>> - How much resources to consume
>> - How fast memory changes or how stable memory is
>>    (this we might be able to detect)
>
> Setting the pages_to_scan is a bit mystical. Setting upper/lower pages_to_scan
> bounds is similarly mystical, and highly workload dependent.
>
> So I agree that a better abstraction to automatically tune the scanning is
> reasonable. I wonder if we can let the user give better inputs that are less
> workload dependent.
>
> For example, do we need min/max values for pages_to_scan, or can we replace it
> by something better to the auto-tuning algorithm?
>
> IMHO "target scan time" goes into the right direction, but it can still be
> fairly workload dependent. Maybe a "max CPU consumption" or sth. like that would
> similarly help to limit CPU waste, and it could be fairly workload dependent.

I can look into replacing min/max values for pages_to_scan with min/max
cpu utilization. This might be easier for users to decide on. However I
still think that we need a target value like scan time to optimize for.
  

On 10.10.23 18:02, Stefan Roesch wrote:
> 
> David Hildenbrand <david@redhat.com> writes:
> 
>> On 06.10.23 18:17, Stefan Roesch wrote:
>>> David Hildenbrand <david@redhat.com> writes:
>>>
>>>> On 04.10.23 21:02, Stefan Roesch wrote:
>>>>> What is the KSM advisor?
>>>>> =========================
>>>>> The ksm advisor automatically manages the pages_to_scan setting to
>>>>> achieve a target scan time. The target scan time defines how many seconds
>>>>> it should take to scan all the candidate KSM pages. In other words the
>>>>> pages_to_scan rate is changed by the advisor to achieve the target scan
>>>>> time.
>>>>> Why do we need a KSM advisor?
>>>>> ==============================
>>>>> The number of candidate pages for KSM is dynamic. It can often be observed
>>>>> that during the startup of an application more candidate pages need to be
>>>>> processed. Without an advisor the pages_to_scan parameter needs to be
>>>>> sized for the maximum number of candidate pages. With the scan time
>>>>> advisor the pages_to_scan parameter based can be changed based on demand.
>>>>> Algorithm
>>>>> ==========
>>>>> The algorithm calculates the change value based on the target scan time
>>>>> and the previous scan time. To avoid pertubations an exponentially
>>>>> weighted moving average is applied.
>>>>> The algorithm has a max and min
>>>>> value to:
>>>>> - guarantee responsiveness to changes
>>>>> - to avoid to spend too much CPU
>>>>> Parameters to influence the KSM scan advisor
>>>>> =============================================
>>>>> The respective parameters are:
>>>>> - ksm_advisor_mode
>>>>>      0: None (default), 1: scan time advisor
>>>>> - ksm_advisor_target_scan_time
>>>>>      how many seconds a scan should of all candidate pages take
>>>>> - ksm_advisor_min_pages
>>>>>      minimum value for pages_to_scan per batch
>>>>> - ksm_advisor_max_pages
>>>>>      maximum value for pages_to_scan per batch
>>>>> The parameters are exposed as knobs in /sys/kernel/mm/ksm.
>>>>> By default the scan time advisor is disabled.
>>>>
>>>> What would be the main reason to not have this enabled as default?
>>>>
>>> There might be already exisiting users which directly set pages_to_scan
>>> and tuned the KSM settings accordingly, as the default setting of 100 for
>>> pages_to_scan is too low for typical workloads.
>>
>> Good point.
>>
>>>
>>>> IIUC, it is kind-of an auto-tuning of pages_to_scan. Would "auto-tuning"
>>>> describe it better than "advisor" ?
>>>>
>>>> [...]
>>>>
>>> I'm fine with auto-tune. I was also thinking about that name, but I
>>> chose advisor, its a bit less strong and it needs input from the user.
>>>
>>
>> I'm not a native speaker, but "adviser" to me implies that no action is taken,
>> only advises are given :) But again, no native speaker.
>>
>>>>> How is defining a target scan time better?
>>>>> ===========================================
>>>>> For an administrator it is more logical to set a target scan time.. The
>>>>> administrator can determine how many pages are scanned on each scan.
>>>>> Therefore setting a target scan time makes more sense.
>>>>> In addition the administrator might have a good idea about the
>>>>> memory sizing of its respective workloads.
>>>>
>>>> Is there any way you could imagine where we could have this just do something
>>>> reasonable without any user input? IOW, true auto-tuning?
>>>>
>>> True auto-tuning might be difficult as users might want to be able to
>>> choose how aggressive KSM is. Some might want it to be as aggressive as
>>> possible to get the maximum de-duplication rate. Others might want a
>>> more balanced approach that takes CPU-consumption into consideration.
>>> I guess it depends if you are memory-bound, cpu-bound or both.
>>
>> Agreed, more below.
>>
>>>
>>>> I read above:
>>>>> - guarantee responsiveness to changes
>>>>> - to avoid to spend too much CPU
>>>>
>>>> whereby both things are accountable/measurable to use that as the input for
>>>> auto-tuning?
>>>>
>>> I'm not sure a true auto-tuning can be achieved. I think we need
>>> some input from the user
>>> - How much resources to consume
>>> - How fast memory changes or how stable memory is
>>>     (this we might be able to detect)
>>
>> Setting the pages_to_scan is a bit mystical. Setting upper/lower pages_to_scan
>> bounds is similarly mystical, and highly workload dependent.
>>
>> So I agree that a better abstraction to automatically tune the scanning is
>> reasonable. I wonder if we can let the user give better inputs that are less
>> workload dependent.
>>
>> For example, do we need min/max values for pages_to_scan, or can we replace it
>> by something better to the auto-tuning algorithm?
>>
>> IMHO "target scan time" goes into the right direction, but it can still be
>> fairly workload dependent. Maybe a "max CPU consumption" or sth. like that would
>> similarly help to limit CPU waste, and it could be fairly workload dependent.
> 
> I can look into replacing min/max values for pages_to_scan with min/max
> cpu utilization. This might be easier for users to decide on. However I
> still think that we need a target value like scan time to optimize for.

Agreed, it can't be completely automatic.