[nsp] 7200, 12.0(21)S4 and "show int" counters?

[Gert Doering]

Hi,

is it just me, or has Cisco engineering reached a new all-time high in
"show int" counter brokenness?

I just upgraded one of our 7200s to 12.0(21)S4 to get some of the S2/S3
bugfixes plus the "E3/T3 adapters don't work" bugfix S3->S4.

The good news: the T3 PA works fine (PA-2T3).

The bad news: the "show int" output counters are finally and completely
broken - they don't hang eventually (as in earlier releases), but they 
don't count *at all* anymore.  Or to be precise, they count "something"
but that has no relation to reality.

Look at this:

Cisco-M-XII#sh int p4/0
POS4/0 is up, line protocol is up 
  Hardware is Packet over Sonet
[..]
  Last input 00:00:00, output 00:00:00, output hang never
  Last clearing of "show interface" counters never
[..]
  5 minute input rate 8321000 bits/sec, 1341 packets/sec
  5 minute output rate 2838000 bits/sec, 638 packets/sec
     513382 packets input, 396467799 bytes, 0 no buffer
     Received 0 broadcasts, 0 runts, 0 giants, 0 throttles
              0 parity
     18 input errors, 18 CRC, 0 frame, 0 overrun, 0 ignored, 0 abort
     28 packets output, 2658 bytes, 0 underruns
     0 output errors, 0 applique, 2 interface resets
     0 output buffer failures, 0 output buffers swapped out
     0 carrier transitions
Cisco-M-XII#sh int p4/0 acc
                Protocol    Pkts In   Chars In   Pkts Out  Chars Out
                      IP     518490  400148237     237198  136223697
                     CDP          7       2457          6       2130

I'm amazed.  How difficult can it be to implement interface counters
that actually count what's being sent out of the interface, and do so 
*correctly* and *all the time*?  This is broken since 11.2P, and I
tend to give up hope that it will *EVER* be fixed.  

(Yes, some sort of bug fix went into 12.0(22)S, but I'm not going to
touch *that* can of worms)

It would be so easy - just return the SNMP counters minus a correction 
value that is saved when you type "clear counter int ..." - as the SNMP 
counters are (mostly) correct.

Did I mention that the SNMP values for "5 minute ... rate" differ up to
20% from the "show int" values for the same rate?  I have no idea how
they managed *that*. 

thoroughly annoyed,

gert

-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert@greenie.muc.de
fax: +49-89-35655025                        gert.doering@physik.tu-muenchen.de