[BBLISA] statistics-based zero config network management: why doesnt this exist?

[John P. Rouillard]

In message <51FEAB39.7070303 at oinc.net>,
"Brian O'Neill" writes:
>On 8/4/2013 3:25 PM, Daniel Feenberg wrote:
>> On Sun, 4 Aug 2013, Edward Ned Harvey (bblisa4) wrote:
>>> The thing is:  Very rarely is SNMP sufficient.  For most devices, it
>>> counts no more than a ping monitor.  If you want reliable statistics of
>>> cpu, disk, network, memory usage, you have to install an agent.

That's difficult to do on switches, routers, nfs applinces etc.

>>> I emphasize reliable.  Because although SNMP technically supports all
>>> that, I've never seen it usable for that purpose.

Snmp may be the only game in town for some devices. Cisco is pretty
good, but Foundary gear had ifLastChange tracking sysUptime in quite a
few releses. Made trying to figure out which interface ports were
unstable quite difficult.

>> Should a switch record dropped packets somewhere? When we have network
>> problems pinging across the network to printers and other reliable devices
>> shows many dropped packets, but the (Dell) switches don't ever seem to
>> show a significant number of errors. Is a dropped packet not an error?
>> Could packets be lost some other way?
>>
>
>I'm actually in the process of trying to track down if the number of 
>RX-DROPs I'm seeing on the Linux side (seen in SNMP and netstat), but 
>not seeing on the switches (netgear managed switches), is a sign of a 
>switch issue or something else. It's not like the typical duplex issue, 
>where you see issues on each side.

I would expect that to be recorded in the etherlike mib under:

            dot3StatsIndex                      INTEGER,
            dot3StatsAlignmentErrors            Counter,
            dot3StatsFCSErrors                  Counter,
            dot3StatsSingleCollisionFrames      Counter,
            dot3StatsMultipleCollisionFrames    Counter,
            dot3StatsSQETestErrors              Counter,
            dot3StatsDeferredTransmissions      Counter,
            dot3StatsLateCollisions             Counter,
            dot3StatsExcessiveCollisions        Counter,
            dot3StatsInternalMacTransmitErrors  Counter,
            dot3StatsCarrierSenseErrors         Counter,
            dot3StatsFrameTooLongs              Counter,
            dot3StatsInternalMacReceiveErrors   Counter,
            dot3StatsEtherChipSet               OBJECT IDENTIFIER

and possibly dot3CollTable for more detiled info. Also check
RMON-MIB. IIRC we use: RMON-MIB::etherStatsDropEvents on our netgear
switches to look for packets being dropped due to queue
exhaustion/overload.

--
				-- rouilj
John Rouillard
===========================================================================
My employers don't acknowledge my existence much less my opinions.