#10397 NORM Not Tri: XOs may fail to communicate in noisy 802.11 environment

[Zarro Boogs per Child]

#10397: XOs may fail to communicate in noisy 802.11 environment
------------------------------------+---------------------------------------
           Reporter:  greenfeld     |       Owner:  dsaxena      
               Type:  defect        |      Status:  new          
           Priority:  normal        |   Milestone:  Not Triaged  
          Component:  connectivity  |     Version:  not specified
         Resolution:                |    Keywords:               
        Next_action:  never set     |    Verified:  0            
Deployment_affected:                |   Blockedby:               
           Blocking:                |  
------------------------------------+---------------------------------------

Comment(by martin.langhoff):

 It is a truism that XOs may fail to communicate if the RF environment is
 noisy enough :-)

 I would like to highlight to tidbits from Quozl and Wad that I think get
 to the heart of the matter.

  * mDNS uses multicast, APs will transmit those at their slowest rate and
 it is well known that slow frames are more (rather than less) vulnerable
 to interference due to their "time on air". XOs with sensitive antennas
 make this worse, as they'll pick up faraway signals, increasing the
 probability of interference.
  * If a registered XO fails to connect using Gabble (XMPP, which is TCP),
 it will fall back to Salut, which is mDNS. This means that (a) it will
 consume lots of airtime and (b) is very unlikely to work, so that airtime
 is wasted.

 There may be other concomitant bugs / behaviours that are affecting this
 scenario, but I'm fairly certain that the points above explain 90% of what
 you are seeing.

 "Urban" noisy environments (many un-coordinated APs) are not our primary
 goal, and improving this is hard. I wonder if this is surfacing any
 problem that we would see in a large/dense school in the field.

 A simple improvement would be to ask Gabble to try several times before
 attempting Salut and to use a longer timeout.

 On a related topic, we need to find a quieter testing spot near the MIA
 office. I think I have a good candidate (the carpark at my building).

-- 
Ticket URL: <http://dev.laptop.org/ticket/10397#comment:6>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system