Conference azur::mcc

5312.0. "Statistics Returned from Incorrect Data" by GLDOA::WITHERELL () Wed Jul 07 1993 13:49

    I am recording all counters in SNMP interfaces every 30 minutes. Each
    night I create a report using the following show command:
    
    show snmp test interface 1 total utilization, in domain test,
    for start = yesterday every 00:30:00 until today duration = 00:30:00
    
    Everything seems to work fine unless there have been missed recordings
    during the day. In that case, the show command returns the same
    total utilization for each 30 minute period until the recording
    resumes.
    
    The performance fm must be using the last counters  recorded before the
    missing period and the first counters recorded after the missing period.
    The time on the output of the show command is the time of interest and
    not the actual time the counters were collected. This is misleading.
    
    
	For example, suppose recording was stopped from 3:00 A.M. until
    9:00 P.M. the output of the show command might look like this:
    
    00:00:00 50 percent
    00:30:00 45.09 percent
    01:00:00 65.02 percent
    01:30:00 56.8 percent
    02:00:00 57 percent
    02:30:00 66 percent
    03:00:00 75.01 percent
    03:30:00 75.01 precent
    04:00:00 75.01 percent
    .
    .
    .
    09:00:00 75.01 percent
    09:30:00 64.02 percent
    .
    .
    .
    
    
    Is there a way to get statistics for only those period when data was
    actually collected?
    
    Thanks,
    
    Liz

Liz ...

There was no data recorded between 3am and 9am, so you figured the
statistics output timestamps would jump from 3am to 9am .. not
trick you into thinking there was actually data recorded .. Right?

Probably what the problem is, the Performance Analyzer (PA) asks for
historical data at 3:30am (in your example) .. There is no data for
this time, so the Historian returns data BEFORE this time: 3:00am

Then PA asks for data at 4:00am, but again there is no data at this
time, so data BEFORE is returned again .. I would think for 3:00am
again, but if both timestamps were the same, the data would be the
same and therefore not have changed, and no stats would be expected.
This is where I'm not sure whats going on inside PA.

The bottom line is, the PA should notice that the timestamps keep
repeating themselves and not calculate or return repeated statistics.
PA should keep looking until the timestamps change, THEN calculate
statistics .. your output would jump from 3:00am to 9:00am, skipping
where no data existed.

Now - where to go from here.  Can you QAR this (see note #7).  Myself
and another engineer are currently looking in to PA QAR/CLD's .. which
all have to do with Historical data.  Maybe we can figure this one out
too.

Thanks /keith

    Thanks for your help.
    
    I have QARed the problem. QAR number 00284.