Re: [lmap] Which performance metrics for LMAP?

["MORTON JR., ALFRED (AL)" <[e-mail]>]

Hi Benoit,

You've asked four good questions today, I'll take a stab at this one.

IMO, the metrics chosen need to serve two purposes, at least.

One set of metrics describe the packet transfer service (at the service layers).
Another (somewhat overlapping) set help subscribers understand
how well their applications will perform on the service
(inferred from measurements at the service layer).

These two points of view (POV) and their implications
on metric selection/parameter settings/reporting for IP packet transfer
were investigated in http://tools.ietf.org/html/rfc6703

IPPM's scope has always included transport layer, so your 5-tuple with
DSCP is covered there: these have all been part of the notion "packets of Type-P"
since the beginning, see http://tools.ietf.org/html/rfc2330#section-13
and these aspects are more important now than ever.

As a limit on "layer 3 and up", I would suggest that limited exceptions for
metrics at +transport layers would be useful, such as the DNS response time
metric constructed in
http://tools.ietf.org/html/draft-bagnulo-ippm-new-registry-independent-00#section-8

So IMO, LMAP infrastructure would support measurement at IP and transport layers
with a few notable exceptions, and look to others (IPPM WG) to select the best
metrics (and develop new ones, as necessary) for each POV that LMAP serves.

my 2cents,
Al


From: [e-mail] [mailto:[e-mail]] On Behalf Of Benoit Claise
Sent: Thursday, March 07, 2013 6:28 AM
To: [e-mail]
Subject: [lmap] Which performance metrics for LMAP?

Dear all,

Another clarifying question: I'm not too sure which performance metric types LMAP targets.
draft-linsner-lmap-use-cases-00 mentions:

   It is assumed that measurement

   tests run by OTT providers would only include the metrics associated

   with layer 3 and up

The different drafts use different terms: service, end user experience, quality of experience.
Does Quality of Experience imply performance metrics per application, per 5 tuple (IP addresses, ports, protocol), per DCSP?
Or do we want to stay at layer 3 performance metrics, typically IPPM type of metrics?
If the latter, then LMAP is about large scale deployment of IPPM, right?

While I fully understand the business needs for "and up" in "include the metrics associated with layer 3 and up", let's not boil the ocean.
Performance Metrics at Other Layer (PMOL<http://datatracker.ietf.org/wg/pmol/charter/>), a concluded WG, tried and it proved to be difficult.

Some more discussions, on the mailing list or during the BoF, on this topic would be appreciated.

Regards, Benoit