Conference npss::gigaswitch

921.0. "Performance / Spanning tree problem" by NETRIX::"[email protected]" (Ryan Price) Thu Feb 06 1997 02:59

Hi,

I intend to get much more details about this problem later this 
month, but I thought in the mean-time, I'd bounce it off the brains 
in this forum just in case someone had seen something like this 
before.

Basically, there is a dual-attached pair of GIGAswitches, with 4 
OpenVMS systems each. One of them also has some other equipment 
hooked onto it (not enough details to give clues to the cause of 
the problem).

There are 2 problems, the symptoms of which are improved by 
disabling spanning tree on the switches:

1)	Performance is shocking (< 10Mb/s between 2 systems)
2)	Forwarding tables seem confused, with Physical and 
        DECnet addresses for the same system being seen on 
        different bridge ports (although the UNA on each GS 
        port is fine).

Now I happen to know that there were other changes made on 
the network (I just don't know WHAT changes) which could 
possibly have had the same affect. Later this month I hope 
to get my own time to analyse the network and make meaningful 
conclusions.

I have managed to convince them that what they are seeing 
is perhaps not a problem with spanning tree itself, but
rather a symptom of some other problem on their network,
which is how I convinced them to let me get a hold of the
network again for testing. What I'd now like some information 
to back up my theory.

BTW, the switches are running the firmware that came with 
clearVISN 1.0 (SCP 3.01, FGL 3.01 & CLOCK 3.0).

Any ideas at this stage would be most welcome.

Ryan Price
NPBU, South Africa
[Posted by WWW Notes gateway]

        The forwarding information appears confused because it is confused.
    
        Almost certainly, you have either an unresolved loop in the
    network, or duplicate addresses.  This explains all the symptoms.
    Fix the problem, not the symptoms - I.E. Identify the cause of the
    problem and don't disable spanning tree on the GIGAswitch/FDDI system. 
    This is the equivalent of overriding the safety cutoff on a machine
    gone haywire.
    
        Create an *accurate* topology map.  Don't believe them, examine all
    the bridges yourself.
    
        Switches don't make up random ports for forwarding.  ...well,
    they're not supposed to at least...   ...only kidding.  Start off by
    assuming that the switch is operating properly.  If an address is
    bouncing back and forth between ports, then that means frames are
    coming in over that port.  Figure out how that is happening.  You will
    most likely require an analyzer.
    
        BL3.1 is the current release (although it won't address their
    problem).
    
    MDL