Conference 7.286::fddi

1373.0. "DECnet errors when used with FDDI and DEC NIS" by 31302::HERNAN (Peter Hernan Western RNC) Tue Jun 14 1994 05:16

  This has been posted in the DECnet/OSI and the DEC NIS notes file as well.



                                                Both of the two FDDI
                               +-------+        controllers are on for IP
                          A*---|VMS/AXP|---*B   and LAVC. DECnet and Multi-
                               |server |        cast on the Ethernet.
                               /-------\
             ------   /=======/         \=======\   ------
           /        \/                           \/        \
          |   FDDI   |         +-------+         |   FDDI   |
           \        /==========|VMS/AXP|==========\        /
             --||--            |server |            --||--
               ||              +/-----\+              ||
            +--||---+          /       \          +---||--+
            |DEC NIS|    |/---/         \---\|    |DEC NIS|    DECnet and
            |  DAS  |    /                   \    |  DAS  |    IP routing,
            +-----\-+    |                   |    +-/-----+    LAVC bridging
                   \---\ |                   | /---/
                        \|                   |/
                         \                   /
                         |                   |
                 A*------|     +-------+     |--------*B
                         |     |VMS/AXP|     |
                         |-----+  WS   +-----|
                         |     +-------+     |
                         |                   |
                         |                   |
                         |     +-------+     |
                         |     |VMS/AXP|     |
                         |-----+  WS   +-----|
                         |     +-------+     |



  The DECnet and the Multicast applications, when run over the FDDI and 
through the DEC NIS, generate large numbers of errors when communicating 
between the servers and the workstations.

  The DECnet application, from the workstation to the servers, after 
running for a while messages seem to be delayed in the network. This delay 
eventually causes the application to log an error and hang. When the 
massages seem to arrive late the NSP Port "Roundtrip Delay Estimate" 
increase by several orders of magnitude (8-12 --> 3000-6000).

  The multicast application, server to workstation, generate many 
Unavailable User Buffer errors on the workstation CSMA-CD Station,
eventually the application fails. The "Unavailable User Buffer" counter
can increase at thousands per second, the ESA devise logs a fatal hardware 
error and goes off line.

  If, as the diagram indicates, the servers are connected directly to the 
Ethernets, the DECnet and Multicast applications run without any errors.
The DECnet is routed through the DEC NIS and the Multicast is bridged.



  What is the unit of measure for the "Roundtrip Delay Estimate"?

  How often is the "Roundtrip Delay Estimate" measured?

  What is the maximum value for this parameter?

  Will the loss of a packet cause this value to increase? rapidly?


  This system is about to go into final acceptance test, if this doesn't 
happen by the beginning of July this could cost DEC some $.
  
  The preferred configuration is for the servers to be connected to the FDDI 
without the additional Ethernet controllers.

  Any ideas would help. Anyone know of any problems with FDDI or the DEC 
NIS that might cause high packet loss in heavy traffic?

  Thanks