Conference netcad::hub_mgnt

    
    The counters are not accurate. This is from experience. I also posted a
    note in this conference which was confirmed.

    Thanks for the speedy reply.
    
    Has anyone any other ideas on how to get a rough idea of FDDI ring
    performance? 
    Is the only option to use a sniffer?
    Would HUBwatch be able to give me any more accurate info, or does it
    use MIB data?
    
    Ta
    
    Bernie  

    The MIB-2 counters are specific to that particular concentrator.
    I don't how you could extrapolate from these numbers the networks
    utilization.  For example, two workstations could be utilizing
    80% of the network with unicast packets.  The concentrators on the
    ring would not "see" any of this traffic.
    
    re .2: Can you send me mail on how you tested the accuracy of the
           concentrator's MIB-2 counters?  Or, tell me what event led you
           to believe the counters are in accurate?
    
    
    Dennis

    
    re: .2
    
    I was at a customer site who had a W&G FDDI monitor on the ring. The
    numbers reported by HUBwatch were 15-20 points higher than the W&G was
    reporting.
    
    Does that answer the question?

    The discrepancy you noted is due to the sampling algorithm employed by
    hubwatch - not due to counter inaccuracy.  We put a great deal of
    effort into assuring the accuracy of the counters.  The algorithm is
    currently under discussion.
     

    
	The following is mail I sent to Bernie off-line that I thought was	
        relevant to this discussion.

    	
    	Dennis


Hi Bernie.  The problem with using MIB-2 counters to compute ring
utilization is that these counters relate to the concentrator's IP stack.
So, if a frame is not processed by some level of the IP stack (e.g., if),
it has no impact on the counters.

A concentrator only "sees" broadcast frames (e.g., ARP requests) and
uni-cast frames directed at it.  All other frames pass through the 
concentrator as though the concentrator is a piece of wire and therefore
don't get processed by the IP stack.  However, the MAC chip in the
concentrator does keep a couple of counts that can be used to compute
the ring's utilization.


Ring utilization can be computed knowing how many times the token
was seen (Ms) and the maximum number of times the token could have been
seen (Mm) over a time period (T).

	utilization = 1 - Ms/Mm

This works because the token is delayed only when a station is transmitting
a frame.

Mm is computed by dividing the length of the time period by the ring's
latency (D).

        Mm = T/D

Which gives,
	
	utilization = 1 - Ms/(T/D) = 1 - MsD/T

Normally written,
		
	U = 1 - MD/T

The needed values are available in the following MIBs:

	D = eMACRingLatency (Vendor MIB)
	M = fddimibMACTokenCts (FDDI MIB; RFC 1512)
	T = sysUpTime (MIB-2)


Read these MIB objects (in one SNMP Get request), wait X seconds and read
the MIB objects again.  T is the difference between the two sysUpTime values
and M is the difference between the two fddimibMACTokenCts values.  Note,
that this is an average.  Also, don't use X as T in you computation; there is
just too much variance in network performance for this to be reliable.


You may also want to check if there was a ring initialization during
your sample period.  If there was, take another sample.  The ring init
check can be done by reading eMACRingInitializationsInitiated and
eMACRingInitializationsReceived.  Both are in the Vendor MIB.  If either
of these counters change, there was a ring init.

Hope this helps.


Dennis




-- 
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Dennis Argo                         226-6603 (DTN)
                                508-486-6603 (office)
[email protected]
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++