Hi Javier,
even using 100ms still causes problems.
Maybe I should give an idea of my setup. I have 6 mesh nodes within
range of each other. They all transmit TCP traffic at a constant bit
rate (500Kbps in my test) to an external laptop through a mesh portal.
With the current implementation and the default values, to refresh a
path a burst of preq are queued and transmitted during one second. Also
each PREQ is a broadcast to the all the mesh nodes. This puts a lot of
unnecessary load on the mesh.
I also noticed that in mesh_queue_preq() ifmsh->last_preq is not updated
until the preq is sent out, which will cause more preqs to be queued if
tx rate < rx rate.
The spec does specify minpreqinterval to be 100ms, it also specify
dot11MeshHWMPmaintenanceInterval to be 2000. "This attribute specifies
the minimum interval of time (in TUs) during which a mesh STA can send
only one Action frame containing PREQ element for path maintenance."
With that in mind I think a path should be refreshed at that rate
regardless of the path expiration time.
Even in nl8022.h the refresh time is specified as follows: "*
@NL80211_MESHCONF_PATH_REFRESH_TIME: how frequently to refresh mesh
paths * (in milliseconds)"
The core of the change would look like this: (note: an extra variable:
last_refresh_time is added to the path struct)
mpath = mesh_path_lookup(target_addr, sdata);
@@ -931,15 +941,20 @@
}
if (mpath->flags & MESH_PATH_ACTIVE) {
- if (time_after(jiffies,
- mpath->exp_time -
-
msecs_to_jiffies(sdata->u.mesh.mshcfg.path_refresh_time)) &&
+ if (time_after_eq(jiffies,
+ mpath->last_refresh_time +
+
msecs_to_jiffies(sdata->u.mesh.mshcfg.path_refresh_time)) &&
!memcmp(sdata->vif.addr, hdr->addr4, ETH_ALEN) &&
!(mpath->flags & MESH_PATH_RESOLVING) &&
!(mpath->flags & MESH_PATH_FIXED)) {
mesh_queue_preq(mpath,
PREQ_Q_F_START | PREQ_Q_F_REFRESH);
+ mpath->last_refresh_time = jiffies;
}
What do you think?
--Fabrice
_______________________________________________
Devel mailing list
Devel@lists.open80211s.org
http://open80211s.com/mailman/listinfo/devel
