[Cerowrt-devel] measuring bandwidth and latency with snmp

[Cerowrt-devel] measuring bandwidth and latency with snmp

[Dave Taht]

having just sat down to implement smokeping and mrtg again on
my lab network, I was struck by this pointless division of tasks and tools,
and how difficult it is to get back useful, combined, results.

There's no need to do ping.You can just measure the
time it takes to issue and get the snmp query response.

Am I missing something obvious?

nothing does this that I've found yet. Anyone know of anything?

The other big numbers that I wanted to pull out was the packet backlog
and drop rate from tc.

-- 
Dave Täht

Re: [Cerowrt-devel] measuring bandwidth and latency with snmp

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1146 bytes --]

On Fri, 30 May 2014 23:05:24 -0700, Dave Taht said:

> There's no need to do ping.You can just measure the
> time it takes to issue and get the snmp query response.
>
> Am I missing something obvious?

For devices with separate control and data planes, it's quite possible
to have 3 different numbers:

1) The time it takes for the data plane to receive and forward a packet.
2) The time it takes for the data or control plane to generate an ICMP response
3) The time it takes for the control plane to do slowpath handling of a
relatively heavy SNMP packet.

It's why traceroute often gives you wonky timing numbers and reports RTTs
to a middle hop are higher than to the destination.  It's because the
router was able to forward packets much faster than the slowpath to generate
the ICMP TTL Exceeded reply....

For longer hauls, there's the added fun that the source and destination may
not be in the same administrative domain - I have smokeping running on
a Raspberry Pi to measure from local Comcast to work.  My personal device
and the work box in a different group are *never* going to speak SNMP to each other. :)

[-- Attachment #2: Type: application/pgp-signature, Size: 848 bytes --]