Re: [PATCH v3] migration/throttle: Add cpu-throttle-tailslow migration parameter

[Dr. David Alan Gilbert]

* Keqian Zhu (address@hidden) wrote:
> At the tail stage of throttling, the Guest is very sensitive to
> CPU percentage while the @cpu-throttle-increment is excessive
> usually at tail stage.
> 
> If this parameter is true, we will compute the ideal CPU percentage
> used by the Guest, which may exactly make the dirty rate match the
> dirty rate threshold. Then we will choose a smaller throttle increment
> between the one specified by @cpu-throttle-increment and the one
> generated by ideal CPU percentage.
> 
> Therefore, it is compatible to traditional throttling, meanwhile
> the throttle increment won't be excessive at tail stage. This may
> make migration time longer, and is disabled by default.
> 
> Signed-off-by: Keqian Zhu <address@hidden>

So I think this is OK; see comment below.

Reviewed-by: Dr. David Alan Gilbert <address@hidden>

> ---
> Cc: Juan Quintela <address@hidden>
> Cc: "Dr. David Alan Gilbert" <address@hidden>
> Cc: Eric Blake <address@hidden>
> Cc: Markus Armbruster <address@hidden>
> ---
>  migration/migration.c | 13 ++++++++++++
>  migration/ram.c       | 25 +++++++++++++++++-----
>  monitor/hmp-cmds.c    |  8 ++++++++
>  qapi/migration.json   | 48 +++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 89 insertions(+), 5 deletions(-)
> 
> diff --git a/migration/migration.c b/migration/migration.c
> index 187ac0410c..d478a87290 100644
> --- a/migration/migration.c
> +++ b/migration/migration.c
> @@ -785,6 +785,8 @@ MigrationParameters *qmp_query_migrate_parameters(Error 
> **errp)
>      params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
>      params->has_cpu_throttle_increment = true;
>      params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
> +    params->has_cpu_throttle_tailslow = true;
> +    params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
>      params->has_tls_creds = true;
>      params->tls_creds = g_strdup(s->parameters.tls_creds);
>      params->has_tls_hostname = true;
> @@ -1324,6 +1326,10 @@ static void 
> migrate_params_test_apply(MigrateSetParameters *params,
>          dest->cpu_throttle_increment = params->cpu_throttle_increment;
>      }
>  
> +    if (params->has_cpu_throttle_tailslow) {
> +        dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
> +    }
> +
>      if (params->has_tls_creds) {
>          assert(params->tls_creds->type == QTYPE_QSTRING);
>          dest->tls_creds = g_strdup(params->tls_creds->u.s);
> @@ -1412,6 +1418,10 @@ static void migrate_params_apply(MigrateSetParameters 
> *params, Error **errp)
>          s->parameters.cpu_throttle_increment = 
> params->cpu_throttle_increment;
>      }
>  
> +    if (params->has_cpu_throttle_tailslow) {
> +        s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
> +    }
> +
>      if (params->has_tls_creds) {
>          g_free(s->parameters.tls_creds);
>          assert(params->tls_creds->type == QTYPE_QSTRING);
> @@ -3594,6 +3604,8 @@ static Property migration_properties[] = {
>      DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
>                        parameters.cpu_throttle_increment,
>                        DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
> +    DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState,
> +                      parameters.cpu_throttle_tailslow, false),
>      DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
>                        parameters.max_bandwidth, MAX_THROTTLE),
>      DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
> @@ -3700,6 +3712,7 @@ static void migration_instance_init(Object *obj)
>      params->has_throttle_trigger_threshold = true;
>      params->has_cpu_throttle_initial = true;
>      params->has_cpu_throttle_increment = true;
> +    params->has_cpu_throttle_tailslow = true;
>      params->has_max_bandwidth = true;
>      params->has_downtime_limit = true;
>      params->has_x_checkpoint_delay = true;
> diff --git a/migration/ram.c b/migration/ram.c
> index 04f13feb2e..3317c99786 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -616,20 +616,34 @@ static size_t save_page_header(RAMState *rs, QEMUFile 
> *f,  RAMBlock *block,
>   * able to complete migration. Some workloads dirty memory way too
>   * fast and will not effectively converge, even with auto-converge.
>   */
> -static void mig_throttle_guest_down(void)
> +static void mig_throttle_guest_down(uint64_t bytes_dirty_period,
> +                                    uint64_t bytes_dirty_threshold)
>  {
>      MigrationState *s = migrate_get_current();
>      uint64_t pct_initial = s->parameters.cpu_throttle_initial;
> -    uint64_t pct_icrement = s->parameters.cpu_throttle_increment;
> +    uint64_t pct_increment = s->parameters.cpu_throttle_increment;
> +    bool pct_tailslow = s->parameters.cpu_throttle_tailslow;
>      int pct_max = s->parameters.max_cpu_throttle;
>  
> +    uint64_t throttle_now = cpu_throttle_get_percentage();
> +    uint64_t cpu_now, cpu_ideal, throttle_inc;
> +
>      /* We have not started throttling yet. Let's start it. */
>      if (!cpu_throttle_active()) {
>          cpu_throttle_set(pct_initial);
>      } else {
>          /* Throttling already on, just increase the rate */
> -        cpu_throttle_set(MIN(cpu_throttle_get_percentage() + pct_icrement,
> -                         pct_max));
> +        if (!pct_tailslow) {
> +            throttle_inc = pct_increment;
> +        } else {
> +            /* Compute the ideal CPU percentage used by Guest, which may
> +             * make the dirty rate match the dirty rate threshold. */
> +            cpu_now = 100 - throttle_now;
> +            cpu_ideal = cpu_now * (bytes_dirty_threshold * 1.0 /
> +                        bytes_dirty_period);

I worry if we need a divide-by-0 check; but that seems unlikely.
Now if that worked out as huge, then I think the MIN's guard it even
with overflow below, so I think we're OK.

> +            throttle_inc = MIN(cpu_now - cpu_ideal, pct_increment);
> +        }
> +        cpu_throttle_set(MIN(throttle_now + throttle_inc, pct_max));
>      }
>  }
>  
> @@ -919,7 +933,8 @@ static void migration_trigger_throttle(RAMState *rs)
>              (++rs->dirty_rate_high_cnt >= 2)) {
>              trace_migration_throttle();
>              rs->dirty_rate_high_cnt = 0;
> -            mig_throttle_guest_down();
> +            mig_throttle_guest_down(bytes_dirty_period,
> +                                    bytes_dirty_threshold);
>          }
>      }
>  }
> diff --git a/monitor/hmp-cmds.c b/monitor/hmp-cmds.c
> index 9b94e67879..acd7539273 100644
> --- a/monitor/hmp-cmds.c
> +++ b/monitor/hmp-cmds.c
> @@ -419,6 +419,10 @@ void hmp_info_migrate_parameters(Monitor *mon, const 
> QDict *qdict)
>          monitor_printf(mon, "%s: %u\n",
>              
> MigrationParameter_str(MIGRATION_PARAMETER_CPU_THROTTLE_INCREMENT),
>              params->cpu_throttle_increment);
> +        assert(params->has_cpu_throttle_tailslow);
> +        monitor_printf(mon, "%s: %s\n",
> +            
> MigrationParameter_str(MIGRATION_PARAMETER_CPU_THROTTLE_TAILSLOW),
> +            params->cpu_throttle_tailslow ? "on" : "off");
>          assert(params->has_max_cpu_throttle);
>          monitor_printf(mon, "%s: %u\n",
>              MigrationParameter_str(MIGRATION_PARAMETER_MAX_CPU_THROTTLE),
> @@ -1271,6 +1275,10 @@ void hmp_migrate_set_parameter(Monitor *mon, const 
> QDict *qdict)
>          p->has_cpu_throttle_increment = true;
>          visit_type_int(v, param, &p->cpu_throttle_increment, &err);
>          break;
> +    case MIGRATION_PARAMETER_CPU_THROTTLE_TAILSLOW:
> +        p->has_cpu_throttle_tailslow = true;
> +        visit_type_bool(v, param, &p->cpu_throttle_tailslow, &err);
> +        break;
>      case MIGRATION_PARAMETER_MAX_CPU_THROTTLE:
>          p->has_max_cpu_throttle = true;
>          visit_type_int(v, param, &p->max_cpu_throttle, &err);
> diff --git a/qapi/migration.json b/qapi/migration.json
> index eca2981d0a..ee6c5a0cae 100644
> --- a/qapi/migration.json
> +++ b/qapi/migration.json
> @@ -552,6 +552,21 @@
>  #                          auto-converge detects that migration is not making
>  #                          progress. The default value is 10. (Since 2.7)
>  #
> +# @cpu-throttle-tailslow: Make CPU throttling slower at tail stage
> +#                         At the tail stage of throttling, the Guest is very
> +#                         sensitive to CPU percentage while the @cpu-throttle
> +#                         -increment is excessive usually at tail stage.
> +#                         If this parameter is true, we will compute the 
> ideal
> +#                         CPU percentage used by the Guest, which may 
> exactly make
> +#                         the dirty rate match the dirty rate threshold. 
> Then we
> +#                         will choose a smaller throttle increment between 
> the
> +#                         one specified by @cpu-throttle-increment and the 
> one
> +#                         generated by ideal CPU percentage.
> +#                         Therefore, it is compatible to traditional 
> throttling,
> +#                         meanwhile the throttle increment won't be excessive
> +#                         at tail stage.
> +#                         The default value is false. (Since 5.1)
> +#
>  # @tls-creds: ID of the 'tls-creds' object that provides credentials for
>  #             establishing a TLS connection over the migration data channel.
>  #             On the outgoing side of the migration, the credentials must
> @@ -631,6 +646,7 @@
>             'compress-level', 'compress-threads', 'decompress-threads',
>             'compress-wait-thread', 'throttle-trigger-threshold',
>             'cpu-throttle-initial', 'cpu-throttle-increment',
> +           'cpu-throttle-tailslow',
>             'tls-creds', 'tls-hostname', 'tls-authz', 'max-bandwidth',
>             'downtime-limit', 'x-checkpoint-delay', 'block-incremental',
>             'multifd-channels',
> @@ -676,6 +692,21 @@
>  #                          auto-converge detects that migration is not making
>  #                          progress. The default value is 10. (Since 2.7)
>  #
> +# @cpu-throttle-tailslow: Make CPU throttling slower at tail stage
> +#                         At the tail stage of throttling, the Guest is very
> +#                         sensitive to CPU percentage while the @cpu-throttle
> +#                         -increment is excessive usually at tail stage.
> +#                         If this parameter is true, we will compute the 
> ideal
> +#                         CPU percentage used by the Guest, which may 
> exactly make
> +#                         the dirty rate match the dirty rate threshold. 
> Then we
> +#                         will choose a smaller throttle increment between 
> the
> +#                         one specified by @cpu-throttle-increment and the 
> one
> +#                         generated by ideal CPU percentage.
> +#                         Therefore, it is compatible to traditional 
> throttling,
> +#                         meanwhile the throttle increment won't be excessive
> +#                         at tail stage.
> +#                         The default value is false. (Since 5.1)
> +#
>  # @tls-creds: ID of the 'tls-creds' object that provides credentials
>  #             for establishing a TLS connection over the migration data
>  #             channel. On the outgoing side of the migration, the credentials
> @@ -763,6 +794,7 @@
>              '*throttle-trigger-threshold': 'int',
>              '*cpu-throttle-initial': 'int',
>              '*cpu-throttle-increment': 'int',
> +            '*cpu-throttle-tailslow': 'bool',
>              '*tls-creds': 'StrOrNull',
>              '*tls-hostname': 'StrOrNull',
>              '*tls-authz': 'StrOrNull',
> @@ -834,6 +866,21 @@
>  #                          auto-converge detects that migration is not making
>  #                          progress. (Since 2.7)
>  #
> +# @cpu-throttle-tailslow: Make CPU throttling slower at tail stage
> +#                         At the tail stage of throttling, the Guest is very
> +#                         sensitive to CPU percentage while the @cpu-throttle
> +#                         -increment is excessive usually at tail stage.
> +#                         If this parameter is true, we will compute the 
> ideal
> +#                         CPU percentage used by the Guest, which may 
> exactly make
> +#                         the dirty rate match the dirty rate threshold. 
> Then we
> +#                         will choose a smaller throttle increment between 
> the
> +#                         one specified by @cpu-throttle-increment and the 
> one
> +#                         generated by ideal CPU percentage.
> +#                         Therefore, it is compatible to traditional 
> throttling,
> +#                         meanwhile the throttle increment won't be excessive
> +#                         at tail stage.
> +#                         The default value is false. (Since 5.1)
> +#
>  # @tls-creds: ID of the 'tls-creds' object that provides credentials
>  #             for establishing a TLS connection over the migration data
>  #             channel. On the outgoing side of the migration, the credentials
> @@ -921,6 +968,7 @@
>              '*throttle-trigger-threshold': 'uint8',
>              '*cpu-throttle-initial': 'uint8',
>              '*cpu-throttle-increment': 'uint8',
> +            '*cpu-throttle-tailslow': 'bool',
>              '*tls-creds': 'str',
>              '*tls-hostname': 'str',
>              '*tls-authz': 'str',
> -- 
> 2.19.1
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK