Hi Bob,
On 01.03.2013, at 19:41, Bob Copeland <me@xxxxxxxxxxxxxxx> wrote:
> On Fri, Mar 01, 2013 at 06:37:36PM +0100, Thomas Huehn wrote:
>> + //printk(KERN_ERR "before sort: max_tp_rate[0]= %i, [1]= %i, [2]= %i, [3]= %i,", mi->max_tp_rate[0], mi->max_tp_rate[1], mi->max_tp_rate[2], mi->max_tp_rate[3]);
>
> Some leftover debugging there, and elsewhere.
>
I should have removed the debugging, thx for finding, I will fix this in v2.
> I take it the changes to minstrel_update_stats fix the problem where
> a rate might get used more than once? A while ago I wrote the following
> patch for that, but never found the time to test it. It would be great
> to know it's fixed.
This patch brings minstrel-ht's way of electing the max probability rate to minstrel.
It relaxes the max probability criteria in such a way, that in cases where max probability of rates its above 95%, they are treated as if they have equally high success probability and now just the throughput matters as decision criteria. Lets make a concrete example of a minstrel statistic table snap shot:
Rate | Success | Throughput
| Prob. [%] | [MBit/s]
-----------------------------------------
6 | 98.5 | 6.0
9 | 99.9 | 8.7
12 | 95.3 | 11.2
18 | 99.8 | 17.3
24 | 95.3 | 21.6
36 | 90.0 | 29.4
48 | 90.2 | 37.6
54 | 84.3 | 39.3
In case we determine the maximal probability rate just by finding highest success prob. it would give us rate 9. With this patch and the "above 95% success everything is just equally successful" we would choose rate 24 as max. probability rate, as it provides the highest throughout among the rates with success prob. > 95%.
If I understand you right, your idea is to avoid equal rates for max_throughput and max_probability. The patch above would use equal rates for max_thr & max_prob. if the channel conditions allow this, so if they are pretty good. In other cases the max. probability rate can be totally different from max_thr. rate. So I do not
understand the reasoning behind enforcing different rates for max_tp and max_prob. Minstrel will just apply its rule of max_prob_rate regardless if this rate also provides best or second best or other throughput values. Would you mind to this this patch with your hwsim ?
Greetings Thomas
> From: Bob Copeland <me@xxxxxxxxxxxxxxx>
> Date: Fri, 18 May 2012 22:57:49 -0400
> Subject: [PATCH] minstrel: avoid reusing max tp/max prob rates if they are the same
>
> Some testing I did a while ago with mac80211_hwsim showed that
> the MRR chain would frequently have the same rate included more
> than once because it would be both the "max throughput" rate and
> the "max success probability" rate.
>
> If we truly cannot deliver packets with that rate due to changes
> in radio conditions, we can waste a lot of airtime with unsuccessful
> retransmissions. So, compute both of the max throughput rates up
> front, and then pick the max prob. rate from the leftovers.
>
> Signed-off-by: Bob Copeland <me@xxxxxxxxxxxxxxx>
> ---
> net/mac80211/rc80211_minstrel.c | 21 +++++++++++----------
> 1 files changed, 11 insertions(+), 10 deletions(-)
>
> diff --git a/net/mac80211/rc80211_minstrel.c b/net/mac80211/rc80211_minstrel.c
> index 79633ae..36255f1 100644
> --- a/net/mac80211/rc80211_minstrel.c
> +++ b/net/mac80211/rc80211_minstrel.c
> @@ -73,7 +73,7 @@ rix_to_ndx(struct minstrel_sta_info *mi, int rix)
> static void
> minstrel_update_stats(struct minstrel_priv *mp, struct minstrel_sta_info *mi)
> {
> - u32 max_tp = 0, index_max_tp = 0, index_max_tp2 = 0;
> + u32 max_tp = 0, index_max_tp = 0, max_tp2 = 0, index_max_tp2 = 0;
> u32 max_prob = 0, index_max_prob = 0;
> u32 usecs;
> u32 p;
> @@ -123,25 +123,26 @@ minstrel_update_stats(struct minstrel_priv *mp, struct minstrel_sta_info *mi)
> for (i = 0; i < mi->n_rates; i++) {
> struct minstrel_rate *mr = &mi->r[i];
> if (max_tp < mr->cur_tp) {
> + max_tp2 = max_tp;
> + index_max_tp2 = index_max_tp;
> index_max_tp = i;
> max_tp = mr->cur_tp;
> - }
> - if (max_prob < mr->probability) {
> - index_max_prob = i;
> - max_prob = mr->probability;
> + } else if (max_tp2 < mr->cur_tp) {
> + index_max_tp2 = i;
> + max_tp2 = mr->cur_tp;
> }
> }
>
> - max_tp = 0;
> + max_prob = 0;
> for (i = 0; i < mi->n_rates; i++) {
> struct minstrel_rate *mr = &mi->r[i];
>
> - if (i == index_max_tp)
> + if (i == index_max_tp || i == index_max_tp2)
> continue;
>
> - if (max_tp < mr->cur_tp) {
> - index_max_tp2 = i;
> - max_tp = mr->cur_tp;
> + if (max_prob < mr->probability) {
> + index_max_prob = i;
> + max_prob = mr->probability;
> }
> }
> mi->max_tp_rate = index_max_tp;
> --
> 1.7.2.5
>
> --
> Bob Copeland %% www.bobcopeland.com
> --
> To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
