Conference netcad::hub_mgnt

261.0. "DEChub90 fibre query." by KERNEL::HOGGAND () Fri Jun 04 1993 03:43

    I have a customer who is experiencing an unusual problem with his
    DEChub 90's. 
    
    At the end of a thinwire segment they have a DEChub90 housing a fibre
    repeater and a couple of DS90TL's. The repeater has three fibre
    connections each to a fibre bridge in one of three other DEChub90's.
    The diagram below shows one such fibre link.
             
     	  +-------+
    	  | 3100  |                           	
          +---+---+                                 +--------+-+
              |            +-+--------+             |        | |
              |            |H|Fibre   +-------------+ Fibre  |H|
    +-+-------+------------|U|Repeater|             | Bridge |U|
      |                    |B|        |             |        |B|
     Bridge		   | |        |             +--------+ |
       to		   |A+--------|             |DS90TL  |B|
      main		   | |DS90TL  |		    +--------+ |
     network               +-+--------+             |DS90TL  | |
    						    +--------+-+
    
    Each hub has two or more DECserver 90TL's in it and all users access a
    service offered by the 3100 on the local segment. All works well under
    normal conditions.
    
    However, if the bridge link to the main network (a Translan) fails,
    (either because the bridge ethernet port is unplugged, the NTU is
    unplugged or the bridge powered off) then some users accessing the 3100 are
    affected; HUB A users are not affected, but users on the other hubs are
    kicked off the system. Replacing the fibre bridges with fibre repeaters
    (90F's) cures this problem.
    
    Does the DEChub90 fibre bridge have any known problems that would
    explain this.
    
    Any help appreciated.
    
    Dave
    62870
     
    

    First, the users on Hub A are not affected as they are directly,
    (no bridge) connected to the same segment as the 3100.
    
    As per the "bridged" part of the network, the Vitalink bridge is
    the root bridge and every DEWGF in the 3 hubs knows it.  If the
    Vitalink bridge goes away then the DEWGF's must select another root
    bridge and this is where your problem starts.  It takes 10's of
    seconds (up to about 30) for spanning tree to select another root
    bridge.  While this is happening no traffic flows.  DECNET can
    withstand this outage as the timers are long enough.  LAT is another
    story.  LAT has a fixed 1 second retransmit timer.  There is also
    a retransmit limit count usually set at 8.  You could try uping
    this number until the sessions usually survive a spanning tree
    reconfig.  By replacing the bridges with repeaters, there is no
    spanning tree reconfig if the Vitalink bridge goes away and thus
    no LAT sessions are lost.
    
    How often does the Vitalink bridge go away?  If it unnecessarily
    high maybe something could be done to correct that problem and thus
    prevent excessive lost LAT sessions.

    By the way, all terminal servers work this way so third party servers
    will be no better.
    
    dave