by Stefan Roesch

[permalink] [raw]

Subject: Re: [PATCH v3 2/4] mm/ksm: add sysfs knobs for advisor

On Tue, Dec 12, 2023, at 10:07 AM, David Hildenbrand wrote:
> On 12.12.23 19:02, Stefan Roesch wrote:
>>
>>
>> On Tue, Dec 12, 2023, at 5:45 AM, David Hildenbrand wrote:
>>> On 05.12.23 00:49, Stefan Roesch wrote:
>>>> This adds four new knobs for the KSM advisor to influence its behaviour.
>>>>
>>>> The knobs are:
>>>> - advisor_mode:
>>>> none: no advisor (default)
>>>> scan-time: scan time advisor
>>>> - advisor_max_cpu: 70 (default, cpu usage percent)
>>>> - advisor_min_pages: 500 (default)
>>>> - advisor_max_pages: 30000 (default)
>>>> - advisor_target_scan_time: 200 (default in seconds)
>>>>
>>>> The new values will take effect on the next scan round.
>>>>
>>>> Signed-off-by: Stefan Roesch <[email protected]>
>>>> ---
>>>> mm/ksm.c | 127 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>> 1 file changed, 127 insertions(+)
>>>>
>>>> diff --git a/mm/ksm.c b/mm/ksm.c
>>>> index b27010fa2e946..18b7185bbc65b 100644
>>>> --- a/mm/ksm.c
>>>> +++ b/mm/ksm.c
>>>> @@ -3735,6 +3735,128 @@ static ssize_t smart_scan_store(struct kobject *kobj,
>>>> }
>>>> KSM_ATTR(smart_scan);
>>>>
>>>> +static ssize_t advisor_mode_show(struct kobject *kobj,
>>>> + struct kobj_attribute *attr, char *buf)
>>>> +{
>>>> + const char *output;
>>>> +
>>>> + if (ksm_advisor == KSM_ADVISOR_NONE)
>>>> + output = "[none] scan-time";
>>>> + else if (ksm_advisor == KSM_ADVISOR_SCAN_TIME)
>>>> + output = "none [scan-time]";
>>>> +
>>>> + return sysfs_emit(buf, "%s\n", output);
>>>> +}
>>>> +
>>>> +static ssize_t advisor_mode_store(struct kobject *kobj,
>>>> + struct kobj_attribute *attr, const char *buf,
>>>> + size_t count)
>>>> +{
>>>> + if (sysfs_streq("scan-time", buf))
>>>> + ksm_advisor = KSM_ADVISOR_SCAN_TIME;
>>>> + else if (sysfs_streq("none", buf))
>>>> + ksm_advisor = KSM_ADVISOR_NONE;
>>>> + else
>>>> + return -EINVAL;
>>>> +
>>>> + /* Set advisor default values */
>>>> + init_advisor();
>>>
>>> Is the "init_advisor()" really required?
>>>
>>
>> The init_advisor is required for the following scenario:
>> - advisor is used
>> - advisor is turned off
>> - pages_to_scan is used and run for some time
>> - advisor is turned on again
>> ==> Advisor would start with two high values for the first calculation
>
> Can't set_advisor_defaults() handle that?
>

Yes, I can move the

advisor_ctx = (const struct advisor_ctx){ 0 };

into set_advisor_defaults().

> --
> Cheers,
>
> David / dhildenb

2023-12-12 18:33:12

by Stefan Roesch

[permalink] [raw]

Subject: Re: [PATCH v3 1/4] mm/ksm: add ksm advisor

On Tue, Dec 12, 2023, at 5:44 AM, David Hildenbrand wrote:
> [...]
>
>> +
>> +/**
>> + * struct advisor_ctx - metadata for KSM advisor
>> + * @start_scan: start time of the current scan
>> + * @scan_time: scan time of previous scan
>> + * @change: change in percent to pages_to_scan parameter
>> + * @cpu_time: cpu time consumed by the ksmd thread in the previous scan
>> + */
>> +struct advisor_ctx {
>> + ktime_t start_scan;
>> + unsigned long scan_time;
>> + unsigned long change;
>> + unsigned long long cpu_time;
>> +};
>> +static struct advisor_ctx advisor_ctx;
>> +
>> +/* Define different advisor's */
>> +enum ksm_advisor_type {
>> + KSM_ADVISOR_NONE,
>> + KSM_ADVISOR_SCAN_TIME,
>> +};
>> +static enum ksm_advisor_type ksm_advisor;
>> +
>> +static void init_advisor(void)
>> +{
>> + advisor_ctx = (const struct advisor_ctx){ 0 };
>> +}
>
> Again, you can drop this completely. The static values are already
> initialized to 0.
>

It is needed for patch 2, I folded it into set_advisor_defaults

> Or is there any reason to initialize to 0 explicitly?
>
>> +
>> +static void set_advisor_defaults(void)
>> +{
>> + if (ksm_advisor == KSM_ADVISOR_NONE)
>> + ksm_thread_pages_to_scan = DEFAULT_PAGES_TO_SCAN;
>> + else if (ksm_advisor == KSM_ADVISOR_SCAN_TIME)
>> + ksm_thread_pages_to_scan = ksm_advisor_min_pages;
>> +}
>
> That function is unused?
>

I think you already saw it, it is used in patch 2, moving the function to patch 2.

>> +
>> +static inline void advisor_start_scan(void)
>> +{
>> + if (ksm_advisor == KSM_ADVISOR_SCAN_TIME)
>> + advisor_ctx.start_scan = ktime_get();
>> +}
>> +
>> +static inline s64 advisor_stop_scan(void)
>> +{
>> + return ktime_ms_delta(ktime_get(), advisor_ctx.start_scan);
>> +}
>
> Just inline that into the caller. Then rename run_advisor() into
> advisor_stop_scan(). So in scan_get_next_rmap_item)( you have paired
> start+stop hooks.
>

The next version has this change.

>> +
>> +/*
>> + * Use previous scan time if available, otherwise use current scan time as an
>> + * approximation for the previous scan time.
>> + */
>> +static inline unsigned long prev_scan_time(struct advisor_ctx *ctx,
>> + unsigned long scan_time)
>> +{
>> + return ctx->scan_time ? ctx->scan_time : scan_time;
>> +}
>> +
>> +/* Calculate exponential weighted moving average */
>> +static unsigned long ewma(unsigned long prev, unsigned long curr)
>> +{
>> + return ((100 - EWMA_WEIGHT) * prev + EWMA_WEIGHT * curr) / 100;
>> +}
>> +
>> +/*
>> + * The scan time advisor is based on the current scan rate and the target
>> + * scan rate.
>> + *
>> + * new_pages_to_scan = pages_to_scan * (scan_time / target_scan_time)
>> + *
>> + * To avoid pertubations it calculates a change factor of previous changes.
>
> s/pertubations/perturbations/

Fixed.

>
> Do you also want to describe how min/max CPU comes into play?
>

I added additional documentation for it in the next version of the patch.

>> + * A new change factor is calculated for each iteration and it uses an
>> + * exponentially weighted moving average. The new pages_to_scan value is
>> + * multiplied with that change factor:
>> + *
>> + * new_pages_to_scan *= change facor
>> + *
>> + * In addition the new pages_to_scan value is capped by the max and min
>> + * limits.
>> + */
>
>
>
> With that, LGTM
>
> Acked-by: David Hildenbrand <[email protected]>
>
> --
> Cheers,
>
> David / dhildenb