2015-03-13 09:54:48

by Felix Fietkau

[permalink] [raw]
Subject: [PATCH] mac80211: minstrel_ht: fix rounding issue in MCS duration calculation

On very high MCS bitrates, the calculated duration of rates that are
next to each other can be very imprecise, due to the small packet size
used as reference (1200 bytes).
This is most visible in VHT80 nss=2 MCS8/9, for which minstrel shows the
same throughput when the probability is also the same. This leads to a
bad rate selection for such rates.

Fix this issue by introducing an average A-MPDU size factor into the
calculation.

Signed-off-by: Felix Fietkau <[email protected]>
---
net/mac80211/rc80211_minstrel_ht.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/mac80211/rc80211_minstrel_ht.c b/net/mac80211/rc80211_minstrel_ht.c
index 80452cf..60698fc 100644
--- a/net/mac80211/rc80211_minstrel_ht.c
+++ b/net/mac80211/rc80211_minstrel_ht.c
@@ -17,10 +17,11 @@
#include "rc80211_minstrel.h"
#include "rc80211_minstrel_ht.h"

+#define AVG_AMPDU_SIZE 16
#define AVG_PKT_SIZE 1200

/* Number of bits for an average sized packet */
-#define MCS_NBITS (AVG_PKT_SIZE << 3)
+#define MCS_NBITS ((AVG_PKT_SIZE * AVG_AMPDU_SIZE) << 3)

/* Number of symbols for a packet with (bps) bits per symbol */
#define MCS_NSYMS(bps) DIV_ROUND_UP(MCS_NBITS, (bps))
@@ -33,7 +34,8 @@
)

/* Transmit duration for the raw data part of an average sized packet */
-#define MCS_DURATION(streams, sgi, bps) MCS_SYMBOL_TIME(sgi, MCS_NSYMS((streams) * (bps)))
+#define MCS_DURATION(streams, sgi, bps) \
+ (MCS_SYMBOL_TIME(sgi, MCS_NSYMS((streams) * (bps))) / AVG_AMPDU_SIZE)

#define BW_20 0
#define BW_40 1
--
2.2.2



2015-03-17 10:03:55

by Johannes Berg

[permalink] [raw]
Subject: Re: [PATCH] mac80211: minstrel_ht: fix rounding issue in MCS duration calculation

On Fri, 2015-03-13 at 10:54 +0100, Felix Fietkau wrote:
> On very high MCS bitrates, the calculated duration of rates that are
> next to each other can be very imprecise, due to the small packet size
> used as reference (1200 bytes).
> This is most visible in VHT80 nss=2 MCS8/9, for which minstrel shows the
> same throughput when the probability is also the same. This leads to a
> bad rate selection for such rates.
>
> Fix this issue by introducing an average A-MPDU size factor into the
> calculation.

Applied.

johannes