[c-nsp] IPSLAs with OpenNMS or Other?

[Peter Rathlev]

I'll give it a shot, even though we don't really have a planned
professional setup. :-) We only use the data for internal purposes.
We're an enterprise and our SLAs (OLAs) currently do not require us to
present any data on latency/jitter/loss.

We use Cacti to graph the results. For voice circuits we typically
measure "standard" 20 packets, 192 bytes payload, 20 ms interval with a
jitter probe. We always use both an EF-marked and a DSCP0-marked.

For data circuit quality we use N x 1000 packets, 384 bytes payload, 50
ms interval, also a jitter probe. This means they run for longer time
and thus will catch the extreme jitter. OTOH it gives a good picture of
the base line.

Regarding the questions:

> 1) Is it best to locate the IPSLA monitor on the switch near the phone
> system or on the WAN edge router (right now we even have anything
> resembling congestion is on our WAN links)?

I'd locate the measuring unit as close to the phone system as possible,
unless of course your part of the responsibility only goes to the edge
router.

> 2) Any gotchas that I need to look out for? (False positives on bad
> performance, etc - for a start I plan on marking the test traffic with
> the same ToS bit that our VoIP will be marked so it gets the same
> priority)

We often use 3560 and 3750 switches as responders (they happen to be in
the right place) and see very varying quality compared to a 2800 in the
same place. I guess that's because of a slow processor or something.

> 3) This should be simple but whats the minimum IOS flavor required to
> configure the IPSLA monitor (2811 router if I decide to make the WAN
> router the IPSLA monitor or 3560 switch if I decide to locate monitor
> to the switch the PBX is on) (I cant figure out the IOS feature
> browser to save my life - sorry I am a N00b)

I seem to remember that the 2800 requires an Enterprise Base license to
run IP SLA probes. I don't know about 3560 since we only use those as
responders.

I think almost all currently availably IOS versions support either "rtr"
or "ip sla monitor" and a jitter probe.

> 4) Suggestion on other tools other than OpenNMS to monitor IPSLA
> stats?

Cacti works very well for us.

> 5) Suggested intervals, packet sizes, anything else of each test?

We largely went for the defaults believing (possibly naively) that this
would follow some "industry standard".

Regards,
Peter