Hi everybody,
About the metric computation from the tx_rate:
We made a patch (against compat-wireless 2014-06-03) that uses an API added
some time ago by Antonio Quartulli
The patch finds the expected throughput from minstrel_ht when possible
Don’t know if it will solve you issue, but anyway here it is:
==================
--- a/net/mac80211/mesh_hwmp.c
+++ b/net/mac80211/mesh_hwmp.c
@@ -12,6 +12,7 @@
#include <asm/unaligned.h>
#include "wme.h"
#include "mesh.h"
+#include "rate.h"
#define TEST_FRAME_LEN 8192
#define MAX_METRIC 0xffffffff
@@ -325,6 +326,8 @@ static u32 airtime_link_metric_get(struc
struct sta_info *sta)
{
struct rate_info rinfo;
+ /* Added for retreive rate */
+ struct rate_control_ref *ref = local->rate_ctrl;
/* This should be adjusted for each device */
int device_constant = 1 << ARITH_SHIFT;
int test_frame_len = TEST_FRAME_LEN << ARITH_SHIFT;
@@ -336,17 +339,27 @@ static u32 airtime_link_metric_get(struc
if (sta->fail_avg >= 100)
return MAX_METRIC;
- sta_set_rate_info_tx(sta, &sta->last_tx_rate, &rinfo);
- rate = cfg80211_calculate_bitrate(&rinfo);
+ /* CD le 01/04/2014 recupere les informations du minstrel[_ht] si
+ * disponible cf patch de AQ */
+ if (ref && ref->ops->get_expected_throughput) {
+ rate = ref->ops->get_expected_throughput(sta->rate_ctrl_priv);
+ } else {
+ sta_set_rate_info_tx(sta, &sta->last_tx_rate, &rinfo);
+ /* Rate is given in units of 10 Kbps by get_expected_throughput
+ * so times this rate by 10 to keep the same unit and not
+ * lose precision */
+ rate = cfg80211_calculate_bitrate(&rinfo)*100;
+ }
+ /* End of cd */
if (WARN_ON(!rate))
return MAX_METRIC;
err = (sta->fail_avg << ARITH_SHIFT) / 100;
- /* bitrate is in units of 100 Kbps, while we need rate in units of
+ /* bitrate is in units of 10 Kbps, while we need rate in units of
* 1Mbps. This will be corrected on tx_time computation.
*/
- tx_time = (device_constant + 10 * test_frame_len / rate);
+ tx_time = (device_constant + 1000 * test_frame_len / rate);
estimated_retx = ((1 << (2 * ARITH_SHIFT)) / (s_unit - err));
result = (tx_time * estimated_retx) >> (2 * ARITH_SHIFT) ;
return (u32)result;
==================
Regards
Jean-Pierre Tosoni
>
> -----Message d'origine-----
> De : Devel [mailto:[email protected]] De la part de Bob
> Copeland via Devel Envoyé : mercredi 10 juin 2015 16:11 À : Yeoh Chun-
> Yeow; [email protected] Objet : Re: Mesh Path - HWMP Problem
>
> On Wed, Jun 10, 2015 at 09:39:14PM +0800, Yeoh Chun-Yeow via Devel wrote:
> > > However when there is high traffic between two nodes A and B,
> > > sometimes the next hop of A and B changes to another address for a
> > > short time and the metric gets worse. After 2 – 3 seconds the original
> direct next hop returns.
> > >
> >
> > There are two possibilities that I think of:
> > 1. Node A does not receive direct PREP from node B.
> > 2. Node B does not receive direct PREQ from node A 3. Node A received
> > PREP from Node B via another hop which has better metric than direct
> > PREP from node B.
> >
> > For 1 and 2, it may due to the link is terrible and packet are dropped.
> >
> > For 3, you may look on reverting this patch and see how.
>
> I hope reverting this is not a solution as it fixed some path instability
> problems for me.
>
> Packet capture from a 3rd party station during the path switch would be
> helpful, it will tell us if some path selection frames got lost. Would be
> nice if you can grab a bunch of mpath dumps along with that.
>
> The path metric should prefer single hops over multiple hops due to the
> constant term; that said, there are some bogus things about the path
> metric as currently implemented:
>
> - the constant term is just made up and may not penalize multiple hops
> correctly
> - last_tx_rate is often the wrong thing (might be a probe rate, e.g.)
>
> We should probably add the metrics to the debug statements...
>
> --
> Bob Copeland %% http://bobcopeland.com/
> _______________________________________________
> Devel mailing list
> [email protected]
> http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel
_______________________________________________
Devel mailing list
[email protected]
http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel
