Re: [Qemu-devel] [PATCH v3 1/1] migration: calculate expected_downtime with ram_bytes_remaining()

[Laurent Vivier]

Hi,

I had time to investigate more on this problem.

On 16/05/2018 15:43, Laurent Vivier wrote:
> Hi Bala,
> 
> I've tested you patch migrating a pseries between a P9 host and a P8
> host with 1G huge page size on the P9 side and 16MB on P8 side and the
> information are strange now.
> 
> "remaining ram" doesn't change, and after a while it can be set to "0"
> and estimated downtime is 0 too, but the migration is not completed and
> "transferred ram" continues to increase.
> 
> so think there is a problem somewhere...
> 
> thanks,
> Laurent
> 
> On 01/05/2018 16:37, Balamuruhan S wrote:
>> Hi,
>>
>> Dave, David and Juan if you guys are okay with the patch, please
>> help to merge it.
>>
>> Thanks,
>> Bala
>>
>> On Wed, Apr 25, 2018 at 12:40:40PM +0530, Balamuruhan S wrote:
>>> expected_downtime value is not accurate with dirty_pages_rate * page_size,
>>> using ram_bytes_remaining would yeild it correct. It will initially be a
>>> gross over-estimate, but for for non-converging migrations it should
>>> approach a reasonable estimate later on.
>>>
>>> currently bandwidth and expected_downtime value are calculated in
>>> migration_update_counters() during each iteration from
>>> migration_thread(), where as remaining ram is calculated in
>>> qmp_query_migrate() when we actually call "info migrate". Due to this
>>> there is some difference in expected_downtime value being calculated.
>>>
>>> with this patch bandwidth, expected_downtime and remaining ram are
>>> calculated in migration_update_counters(), retrieve the same value during
>>> "info migrate". By this approach we get almost close enough value.
>>>
>>> Reported-by: Michael Roth <address@hidden>
>>> Signed-off-by: Balamuruhan S <address@hidden>
>>> ---
>>>  migration/migration.c | 11 ++++++++---
>>>  migration/migration.h |  1 +
>>>  2 files changed, 9 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/migration/migration.c b/migration/migration.c
>>> index 52a5092add..5d721ee481 100644
>>> --- a/migration/migration.c
>>> +++ b/migration/migration.c
>>> @@ -614,7 +614,7 @@ static void populate_ram_info(MigrationInfo *info, 
>>> MigrationState *s)
>>>      }
>>>
>>>      if (s->state != MIGRATION_STATUS_COMPLETED) {
>>> -        info->ram->remaining = ram_bytes_remaining();
>>> +        info->ram->remaining = s->ram_bytes_remaining;

Don't remove the ram_byte_remaining(), it is updated more often, and
give a better information about the state of memory.
(this why in my test case I have a "remaining" ram" freezed)

>>>          info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
>>>      }
>>>  }
>>> @@ -2227,6 +2227,7 @@ static void migration_update_counters(MigrationState 
>>> *s,
>>>      transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
>>>      time_spent = current_time - s->iteration_start_time;
>>>      bandwidth = (double)transferred / time_spent;
>>> +    s->ram_bytes_remaining = ram_bytes_remaining();
>>>      s->threshold_size = bandwidth * s->parameters.downtime_limit;

To have an accurate value, we must read the remaining ram just after
having updated the dirty pages count, so I think after
migration_bitmap_sync_range() in migration_bitmap_sync()

>>>
>>>      s->mbps = (((double) transferred * 8.0) /
>>> @@ -2237,8 +2238,12 @@ static void migration_update_counters(MigrationState 
>>> *s,
>>>       * recalculate. 10000 is a small enough number for our purposes
>>>       */
>>>      if (ram_counters.dirty_pages_rate && transferred > 10000) {
>>> -        s->expected_downtime = ram_counters.dirty_pages_rate *
>>> -            qemu_target_page_size() / bandwidth;
>>> +        /*
>>> +         * It will initially be a gross over-estimate, but for for
>>> +         * non-converging migrations it should approach a reasonable 
>>> estimate
>>> +         * later on
>>> +         */
>>> +        s->expected_downtime = s->ram_bytes_remaining / bandwidth;
>>>      }
>>>
>>>      qemu_file_reset_rate_limit(s->to_dst_file);
>>> diff --git a/migration/migration.h b/migration/migration.h
>>> index 8d2f320c48..8584f8e22e 100644
>>> --- a/migration/migration.h
>>> +++ b/migration/migration.h
>>> @@ -128,6 +128,7 @@ struct MigrationState
>>>      int64_t downtime_start;
>>>      int64_t downtime;
>>>      int64_t expected_downtime;
>>> +    int64_t ram_bytes_remaining;
>>>      bool enabled_capabilities[MIGRATION_CAPABILITY__MAX];
>>>      int64_t setup_time;
>>>      /*
>>> -- 
I think you don't need to add ram_byte_remaining, there is in
ram_counters a "remaining" field that seems unused.

I think this fix can be as simple as:

diff --git a/migration/migration.c b/migration/migration.c
index 1e99ec9..25b26f3 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -2712,14 +2712,7 @@ static void
migration_update_counters(MigrationState *s,
     s->mbps = (((double) transferred * 8.0) /
                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;

-    /*
-     * if we haven't sent anything, we don't want to
-     * recalculate. 10000 is a small enough number for our purposes
-     */
-    if (ram_counters.dirty_pages_rate && transferred > 10000) {
-        s->expected_downtime = ram_counters.dirty_pages_rate *
-            qemu_target_page_size() / bandwidth;
-    }
+    s->expected_downtime = ram_counters.remaining / bandwidth;

     qemu_file_reset_rate_limit(s->to_dst_file);

diff --git a/migration/ram.c b/migration/ram.c
index a500015..5f7b9f1 100644
--- a/migration/ram.c
+++ b/migration/ram.c
@@ -1164,6 +1164,7 @@ static void migration_bitmap_sync(RAMState *rs)

     trace_migration_bitmap_sync_end(rs->num_dirty_pages_period);

+    ram_counters.remaining = ram_bytes_remaining();
     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);

     /* more than 1 second = 1000 millisecons */