Conference 7.286::fddi

    There is no project manager for the VMS drivers.
    
    What would you like to know?

    Dave,
    
    we are currently working on a customer configuration (2 * VAX 6510 via
    DEMFA on FDDI, 40 Satellites on the Ethernet) which reports a error
    (port has closed virtual circuits, always between the 2 VAX 6510 - ok
    after ~15..20 seconds) as soon as the FDDI ring is connected. The two 6510
    have an alternate way (DEBNI) to Ethernet. No problems with the
    Ethernet only configuration.
    
    The last day I stepped through the PEDRIVER to find out what could lead
    to the VC close message. I believe that some kind of status code is
    passed through from the FXDRIVER.
    The FXDRIVER maintaines a lot of counters which could be usefull for
    troubleshooting. I'm looking for an engeneer which can pass me the
    listings of the version which was shipped with VMS 5.5-2HW.
    I would like to send him a mail if think I need it. At the moment I try
    to get as much informations as possible before we open a CLD and push it
    through our escalation channels.
    
    Regards,
       Thomas
    
    
    If you are intersted in our problems:
    
       SDA> SHOW LAN/FULL	ok
       SDA> SHOW PORT		ok
       SDA> SHOW CONN		ok
       Counters of bridges
          and concentrators	ok
       Counters of DECservers	ok
       FDDI analyses		ok

Could this simply be the VC closing and re-forming to take advantage of the
~4500 byte packet size possible on FDDI vs. ~1500 on Ethernet?  What is the
value for LRPSIZE?

    After installing FDDI LRPSIZE was set to 4500 byte. During
    troubleshooting we discussed the size and decreased the value to 1500.
    It changed nothing. We increased LRPSIZE again to 4500.
    During our FDDI analyses we didn't saw a packet with a size
    of ~4000 byte. It makes sense in my eyes bekause of the DECbridge
    and all the Satellites handling packets of ~1500 byte on the Ethernet.
    
    As far as I know there is no need for SCS to deal with big packets
    on the FDDI except for file serving...
    

Basically, PEDRIVER causes the VC to be closed when it doesn't receive SCS
traffic from another node for a period of time between 8 and 9 seconds.  Each
cluster node sends out a broadcast Hello packet every 3 seconds on each
adapter, so there should always be at least that much traffic coming through to
reset the listen timer. 

It sounds like you're losing packets somewhere along the way.  See
SPEZKO::CLUSTER 3316.2 for a way to check the counters for SCS packets which
had to be retransmitted. 

You may also find the LAVC$FAILURE_ANALYSIS tools helpful (see the V5.5
VAXcluster manual). 

I'd also check to see if an excessive DECnet executor pipeline quota might be a
contributing factor (see CLUSTER note 3166).  Over-running a node's Ethernet
adapter with DECnet packets could conceivably cause SCS traffic to be dropped
as well. 

    During the time we run the FDDI configuration (4 hours) ReXmt counts
    ~100 on the boot node and ~50 on the Satellites.
    The ReRcv count was ~50 on the boot node and ~30 on the Satellites.
    I remember these values from my last visit onsite.
    I try to arrange a FDDI configuration again and will have a closer
    look at these counters.
    
    What can be done if I get a lot of retransmitts ?
    
    When I started troubleshooting weeks ago I turned the MFA-0 DECnet lines
    and circuits permanently off. They are still turned off.
    
    I will have a look at the LAVC$FAILURE_ANALYSIS.
    

re: .0-< Who designs the FXDRIVER ? >-

Common Name:   DAVID GAGNE
Search Surname:  GAGNE  Search Given Name:  DAVID,  DAVID A  DTN:  381-1467
Intrnl Mail Addr:  ZKO3-4/U14  Location:  ZKO  Node:  STAR  Username:
GAGNE
Org Unit:  HARDWARE

Hope he doesn't shoot me.- Verell