mac80211: use AQL airtime for expected throughput.
Subsystems (3)
RFC PATCH 1/1 mac80211: use AQL airtime for expected throughput.
Since the integration of AQL, packet TX airtime estimation is calculated and counted to be used for the dequeue limit.
Use this estimated airtime to compute expected throughput for each station.
It will be a generic mac80211 implementation. If the driver has get_expected_throughput implementation, it will be used instead.
Useful for L2 routing protocols, like B.A.T.M.A.N.
--- net/mac80211/driver-ops.h | 2 ++ net/mac80211/sta_info.h | 2 ++ net/mac80211/status.c | 22 ++++++++++++++++++++++ net/mac80211/tx.c | 3 ++- 4 files changed, 28 insertions(+), 1 deletion(-)
net/mac80211/driver-ops.h +2 -0
@@ -1142,6 +1142,8 @@ static inline u32 drv_get_expected_throughput(struct ieee80211_local *local,trace_drv_get_expected_throughput (&sta->sta);if (local->ops->get_expected_throughput && sta->uploaded) ret = local->ops->get_expected_throughput (&local->hw, &sta->sta); +else + ret =ewma_avg_est_tp_read (&sta->status_stats.avg_est_tp);trace_drv_return_u32 (local, ret);return ret;
net/mac80211/sta_info.h +2 -0
@@ -123,6 +123,7 @@ enum ieee80211_sta_info_flags {#define HT_AGG_STATE_STOP_CB 7#define HT_AGG_STATE_SENT_ADDBA 8 +DECLARE_EWMA(avg_est_tp, 8, 16) DECLARE_EWMA(avg_signal, 10, 8)enum ieee80211_agg_stop_reason {AGG_STOP_DECLINED ,
@@ -641,6 +642,7 @@ struct sta_info {s8 last_ack_signal;bool ack_signal_filled;struct ewma_avg_signal avg_ack_signal; +struct ewma_avg_est_tp avg_est_tp; } status_stats;/* Updated from TX path only, no locking requirements */
net/mac80211/status.c +22 -0
@@ -1145,6 +1145,28 @@ void ieee80211_tx_status_ext(struct ieee80211_hw *hw, sta->status_stats.retry_failed++; sta->status_stats.retry_count += retry_count; +if (skb && tx_time_est) { +/* max average packet size */ +size_t pkt_size = skb->len >1024 ?1024 : skb->len; + +if (acked) { +/* ACK packet size */ + pkt_size +=14 ; +/* SIFS x 2 */ + tx_time_est +=2 *2 ; + } + +/* Backoff average x retries */ + tx_time_est += retry_count ? retry_count *2 :2 ; + +/* failed tx */ +if (!acked && !noack_success) + pkt_size =0 ; + +ewma_avg_est_tp_add (&sta->status_stats.avg_est_tp, + ((pkt_size *8 ) *1000 ) / tx_time_est); + } +if (ieee80211_hw_check (&local->hw,REPORTS_TX_ACK_STATUS )) {if (sdata->vif.type ==NL80211_IFTYPE_STATION && skb && !(info->flags &IEEE80211_TX_CTL_HW_80211_ENCAP ))
net/mac80211/tx.c +2 -1
@@ -3793,7 +3793,8 @@ struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,IEEE80211_SKB_CB (skb)->control.vif = vif;if (vif && -wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL )) { + (!local->ops->get_expected_throughput || +wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL ))) {bool ampdu = txq->ac !=IEEE80211_AC_VO ;u32 airtime;
Signature · 1 line
2.36.1
Quoted 57 lines · Baligh Gasmi
Since the integration of AQL, packet TX airtime estimation is calculated and counted to be used for the dequeue limit.
Use this estimated airtime to compute expected throughput for each station.
It will be a generic mac80211 implementation. If the driver has
get_expected_throughputimplementation, it will be used instead.Useful for L2 routing protocols, like B.A.T.M.A.N.
Signed-off-by Baligh Gasmi <[email protected]>--- net/mac80211/driver-ops.h | 2 ++ net/mac80211/sta_info.h | 2 ++ net/mac80211/status.c | 22 ++++++++++++++++++++++ net/mac80211/tx.c | 3 ++- 4 files changed, 28 insertions(+), 1 deletion(-)net/mac80211/driver-ops.h +2
-0@@ -1142,6 +1142,8 @@ static inline u32 drv_get_expected_throughput(struct ieee80211_local *local,trace_drv_get_expected_throughput (&sta->sta);if (local->ops->get_expected_throughput && sta->uploaded) ret = local->ops->get_expected_throughput (&local->hw, &sta->sta); +else + ret =ewma_avg_est_tp_read (&sta->status_stats.avg_est_tp);trace_drv_return_u32 (local, ret);return ret;net/mac80211/sta_info.h +2
-0@@ -123,6 +123,7 @@ enum ieee80211_sta_info_flags {#define HT_AGG_STATE_STOP_CB 7#define HT_AGG_STATE_SENT_ADDBA 8 +DECLARE_EWMA(avg_est_tp, 8, 16) DECLARE_EWMA(avg_signal, 10, 8)enum ieee80211_agg_stop_reason {AGG_STOP_DECLINED ,@@ -641,6 +642,7 @@ struct sta_info {s8 last_ack_signal;bool ack_signal_filled;struct ewma_avg_signal avg_ack_signal; +struct ewma_avg_est_tp avg_est_tp; } status_stats;/* Updated from TX path only, no locking requirements */ net/mac80211/status.c +0
-0
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1148
Quoted 4 lines
net/mac80211/status.c +1
-0@@ -1145,6 +1145,28 @@ void ieee80211_tx_status_ext(struct ieee80211_hw *hw, sta->status_stats.retry_failed++; sta->status_stats.retry_count += retry_count; +if (skb && tx_time_est) {
Shouldn't this be conditioned on actually being used (i.e., existence of get_expected_throughput op? Also maybe pull it out into its own function to make it clear what it's doing...
Quoted 9 lines
net/mac80211/status.c +9
-0+/* max average packet size */ +size_t pkt_size = skb->len >1024 ?1024 : skb->len; + +if (acked) { +/* ACK packet size */ + pkt_size +=14 ; +/* SIFS x 2 */ + tx_time_est +=2 *2 ; + }
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1167
Quoted 10 lines
net/mac80211/status.c +10
-0+ +/* Backoff average x retries */ + tx_time_est += retry_count ? retry_count *2 :2 ; + +/* failed tx */ +if (!acked && !noack_success) + pkt_size =0 ; + +ewma_avg_est_tp_add (&sta->status_stats.avg_est_tp, + ((pkt_size *8 ) *1000 ) / tx_time_est);
Could we avoid adding this division in the fast path?
Quoted 10 lines
net/mac80211/status.c +2
-0+ } +if (ieee80211_hw_check (&local->hw,REPORTS_TX_ACK_STATUS )) {if (sdata->vif.type ==NL80211_IFTYPE_STATION && skb && !(info->flags &IEEE80211_TX_CTL_HW_80211_ENCAP ))net/mac80211/tx.c +0
-0
net/mac80211/tx.c struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw, · line 3797
Quoted 6 lines
net/mac80211/tx.c +2
-1@@ -3793,7 +3793,8 @@ struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,IEEE80211_SKB_CB (skb)->control.vif = vif;if (vif && -wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL )) { + (!local->ops->get_expected_throughput || +wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL ))) {
This implicitly enables AQL for every driver that doesn't set get_expected_throughput, no? That is probably not a good idea...
-Toke
Quoted 60 lines
Quoted 60 lines
Since the integration of AQL, packet TX airtime estimation is calculated and counted to be used for the dequeue limit.
Use this estimated airtime to compute expected throughput for each station.
It will be a generic mac80211 implementation. If the driver has
get_expected_throughputimplementation, it will be used instead.Useful for L2 routing protocols, like B.A.T.M.A.N.
Signed-off-by Baligh Gasmi <[email protected]>--- net/mac80211/driver-ops.h | 2 ++ net/mac80211/sta_info.h | 2 ++ net/mac80211/status.c | 22 ++++++++++++++++++++++ net/mac80211/tx.c | 3 ++- 4 files changed, 28 insertions(+), 1 deletion(-)net/mac80211/driver-ops.h +2
-0@@ -1142,6 +1142,8 @@ static inline u32 drv_get_expected_throughput(struct ieee80211_local *local,trace_drv_get_expected_throughput (&sta->sta);if (local->ops->get_expected_throughput && sta->uploaded) ret = local->ops->get_expected_throughput (&local->hw, &sta->sta); +else + ret =ewma_avg_est_tp_read (&sta->status_stats.avg_est_tp);trace_drv_return_u32 (local, ret);return ret;net/mac80211/sta_info.h +2
-0@@ -123,6 +123,7 @@ enum ieee80211_sta_info_flags {#define HT_AGG_STATE_STOP_CB 7#define HT_AGG_STATE_SENT_ADDBA 8 +DECLARE_EWMA(avg_est_tp, 8, 16) DECLARE_EWMA(avg_signal, 10, 8)enum ieee80211_agg_stop_reason {AGG_STOP_DECLINED ,@@ -641,6 +642,7 @@ struct sta_info {s8 last_ack_signal;bool ack_signal_filled;struct ewma_avg_signal avg_ack_signal; +struct ewma_avg_est_tp avg_est_tp; } status_stats;/* Updated from TX path only, no locking requirements */ net/mac80211/status.c +0
-0@@ -1145,6 +1145,28 @@ void ieee80211_tx_status_ext(struct ieee80211_hw *hw, sta->status_stats.retry_failed++; sta->status_stats.retry_count += retry_count;
Quoted 5 lines
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1148
Quoted 1 line
net/mac80211/status.c +1
-0+if (skb && tx_time_est) {Shouldn't this be conditioned on actually being used (i.e., existence of
get_expected_throughputop? Also maybe pull it out into its own function to make it clear what it's doing...
It's already the case I think, since the tx_time_est is not-zero only when actually an estimated time is set.
A dedicated function seems good for me, for clarity, yes.
Quoted 14 lines
Quoted 14 lines
net/mac80211/status.c +13
-0+/* max average packet size */ +size_t pkt_size = skb->len >1024 ?1024 : skb->len; + +if (acked) { +/* ACK packet size */ + pkt_size +=14 ; +/* SIFS x 2 */ + tx_time_est +=2 *2 ; + } + +/* Backoff average x retries */ + tx_time_est += retry_count ? retry_count *2 :2 ; +
Quoted 8 lines
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1167
Quoted 6 lines
net/mac80211/status.c +6
-0+/* failed tx */ +if (!acked && !noack_success) + pkt_size =0 ; + +ewma_avg_est_tp_add (&sta->status_stats.avg_est_tp, + ((pkt_size *8 ) *1000 ) / tx_time_est);Could we avoid adding this division in the fast path?
Maybe we can use the do_div() macro for optimization, I don't see how we can avoid it.
Quoted 13 lines
Quoted 13 lines
net/mac80211/status.c +2
-0+ } +if (ieee80211_hw_check (&local->hw,REPORTS_TX_ACK_STATUS )) {if (sdata->vif.type ==NL80211_IFTYPE_STATION && skb && !(info->flags &IEEE80211_TX_CTL_HW_80211_ENCAP ))net/mac80211/tx.c +0
-0@@ -3793,7 +3793,8 @@ struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw,IEEE80211_SKB_CB (skb)->control.vif = vif;
Quoted 7 lines
net/mac80211/tx.c struct sk_buff *ieee80211_tx_dequeue(struct ieee80211_hw *hw, · line 3797
Quoted 4 lines
net/mac80211/tx.c +2
-1if (vif && -wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL )) { + (!local->ops->get_expected_throughput || +wiphy_ext_feature_isset (local->hw.wiphy,NL80211_EXT_FEATURE_AQL ))) {This implicitly enables AQL for every driver that doesn't set
get_expected_throughput, no? That is probably not a good idea...
No, AQL will be disabled if it's already the case, according to the NL80211_EXT_FEATURE_AQL, only airtime estimation will be activated with this diff. The function ieee80211_sta_update_pending_airtime() is already checking for NL80211_EXT_FEATURE_AQL.
Quoted 3 lines
-Toke
Back to the base idea, what do you think about it, can it be a good road to take ?
Other factors can be added progressively to better estimate the throughput...
Quoted 93 lines · Baligh GASMI
Quoted 60 lines
Quoted 60 lines
Since the integration of AQL, packet TX airtime estimation is calculated and counted to be used for the dequeue limit.
Use this estimated airtime to compute expected throughput for each station.
It will be a generic mac80211 implementation. If the driver has
get_expected_throughputimplementation, it will be used instead.Useful for L2 routing protocols, like B.A.T.M.A.N.
Signed-off-by Baligh Gasmi <[email protected]>--- net/mac80211/driver-ops.h | 2 ++ net/mac80211/sta_info.h | 2 ++ net/mac80211/status.c | 22 ++++++++++++++++++++++ net/mac80211/tx.c | 3 ++- 4 files changed, 28 insertions(+), 1 deletion(-)net/mac80211/driver-ops.h +2
-0@@ -1142,6 +1142,8 @@ static inline u32 drv_get_expected_throughput(struct ieee80211_local *local,trace_drv_get_expected_throughput (&sta->sta);if (local->ops->get_expected_throughput && sta->uploaded) ret = local->ops->get_expected_throughput (&local->hw, &sta->sta); +else + ret =ewma_avg_est_tp_read (&sta->status_stats.avg_est_tp);trace_drv_return_u32 (local, ret);return ret;net/mac80211/sta_info.h +2
-0@@ -123,6 +123,7 @@ enum ieee80211_sta_info_flags {#define HT_AGG_STATE_STOP_CB 7#define HT_AGG_STATE_SENT_ADDBA 8 +DECLARE_EWMA(avg_est_tp, 8, 16) DECLARE_EWMA(avg_signal, 10, 8)enum ieee80211_agg_stop_reason {AGG_STOP_DECLINED ,@@ -641,6 +642,7 @@ struct sta_info {s8 last_ack_signal;bool ack_signal_filled;struct ewma_avg_signal avg_ack_signal; +struct ewma_avg_est_tp avg_est_tp; } status_stats;/* Updated from TX path only, no locking requirements */ net/mac80211/status.c +0
-0@@ -1145,6 +1145,28 @@ void ieee80211_tx_status_ext(struct ieee80211_hw *hw, sta->status_stats.retry_failed++; sta->status_stats.retry_count += retry_count;Quoted 5 lines
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1148
Quoted 1 line
net/mac80211/status.c +1
-0+if (skb && tx_time_est) {Shouldn't this be conditioned on actually being used (i.e., existence of
get_expected_throughputop? Also maybe pull it out into its own function to make it clear what it's doing...It's already the case I think, since the
tx_time_estis not-zero only when actually an estimated time is set.
A dedicated function seems good for me, for clarity, yes.Quoted 14 lines
Quoted 14 lines
net/mac80211/status.c +13
-0+/* max average packet size */ +size_t pkt_size = skb->len >1024 ?1024 : skb->len; + +if (acked) { +/* ACK packet size */ + pkt_size +=14 ; +/* SIFS x 2 */ + tx_time_est +=2 *2 ; + } + +/* Backoff average x retries */ + tx_time_est += retry_count ? retry_count *2 :2 ; +Quoted 8 lines
net/mac80211/status.c void ieee80211_tx_status_ext(struct ieee80211_hw *hw, · line 1167
Quoted 6 lines
net/mac80211/status.c +6
-0+/* failed tx */ +if (!acked && !noack_success) + pkt_size =0 ; + +ewma_avg_est_tp_add (&sta->status_stats.avg_est_tp, + ((pkt_size *8 ) *1000 ) / tx_time_est);Could we avoid adding this division in the fast path?
Quoted 2 lines
Maybe we can use the
do_div()macro for optimization, I don't see how we can avoid it.
Here's one idea on how to make it less expensive and improve the average of the throughput estimation at the same time:
In sta->status_stats add these fields:
u32 last_tp_update;
u32 pkt_size;
u32 tx_time_est;
And in this part of the code you do something like this:
u32 diff = ((u32)jiffies) - sta->status_stats.last_tp_update; sta->status_stats.pkt_size += (pkt_size * 8) * 1000; sta->status_stats.tx_time_est += tx_time_est;
if (diff > HZ / 10) {
ewma_avg_est_tp_add(&sta->status_stats.avg_est_tp,
sta->status_stats.pkt_size /
sta->status_stats.tx_time_est);
sta->status_stats.pkt_size = 0;
sta->status_stats.tx_time_est = 0;
}This ensures that you will only get max. 1 update per 100ms interval.
Does that make sense?
- Felix
Indeed it's less expensive.
I'll try to make something in this direction to see what it looks like.
- Baligh
Quoted 3 lines · Baligh GASMI
Indeed it's less expensive.
I'll try to make something in this direction to see what it looks like.
Thanks. Please also make sure that all of this extra work is not performed for drivers that don't need it, because they either implement .get_expected_throughput, or use minstrel, which also implements it in a better way.
- Felix
A message that is not in the archive (sent off-list or not yet received): CAA93jw4FnaCupm4uf4KOHXdLPntKKLq=5ncLLZOiiUvbmrSR-Q@mail.gmail.com
Quoted 1 line · Dave Taht
Sigh. Ideally we are trying to keep latency low by deeply understanding current conditions. Batching up anything on this large interval feels wrong. Powersave and beacons become "beating" frequencies.
Shorter intervals (or even doing it without batching at all) will likely lead to much more fluctuation, which could make the average less useful. I think batching it up will likely be beneficial for protocols like batman-adv.
- Felix