Minutes for LMAP at IETF-93
minutes-93-lmap-1

LMAP WG IETF 93 Minutes
Dan Romascanu and Jason Weil chairing

Based on notes taken by Philip Eardley

July 23, 2015

1. Note Well, Note Takers, Jabber Scribes, Agenda Bashing - Chairs

Dan: Intro
Agenda bash: Mobile LMAP use cases move first as Antonio is remote and has
issues? ok.

2. WG Status - Chairs

WG Status:
as per slides

9. Mobile LMAP Use Cases - Antonio Bovo

Antonio:
no slides - see i-d
dan: who has read it? No-one
Dan; result of study group in ieee 802.15. not planning stds work, but release
part of content to other organizations. Dan: Antonio, please look at our use
case and framework docs to see if we need to re-visit them. Could also
influence future re-charter.

3. LMAP Information Model ? Juergen

[1] changes:-
Task can now calculate multiple metrics from same pkt train

schedules are triggered by events - timings are just one possibility. see
slides last time from ieee - could be eg if you enter a certain region. actions
feed into results, not other schedules . for simplicity, optimizing for the
common case removed idea of multiple task outputs - but can tag, so only look
at what you're interested in sequential parallel and pipeline of actions. [2]
instruction task. Juergen proposing to ditch the term. Tim Carey - was an
artificial construct (from BBF perspective) Juergen - trying to distinguish
tasks that do measurement and infrastr task ok, get rid of it [check during
wglc] [3] schedule gets disabled [not tasks etc] Tim - got to rev tr69 model
for latest version of info model. fundamental problem is operational status vs
admin state. Don?t accurately reflect this in info model. Juergen - yes want to
know state of actions and taskas as well as schedules. [4] execution mode what
should it be? sequential, parallel or pipeline? Tim & Juergen favour seq or
pip. both favour pipeline as default.

Tim
concerned making changes on the draft that haven't been discussed. Thought we
settled on task outputs. Don?t know about the data models that are out there.
Same for changes of naming. not just one data model. Dan - perhaps we should
have had an interim. Tim - please get on list first, as organizations doing
data models other than the yang one. Tim - why multiple tasks? juergen - metric
defines a metric you can send a traffic pattern and calculate multiple metrics
from it. refer to multiple metrics in the registry. multiple registry entries?
do you do one ? task can declare it calculates multiple metrics Al - what i
want. no registry today. spec for stream of pkts in active, measure metrics on
stream, usually several on same stream. Tim- how does the CPE know? Al - have
to be unionable. if reg entry references different stream characteristics
should be in different tasks. Phil - CPE may be simple and not be able to
calculate the tasks are unionable. Tim - 2 registry entries for 2 metrics to
ping task. incumbent on registry to define relationship between tasks. phil -
registry going to WGLC on Friday.

4. LMAP YANG Data Model ? Juergen

similar changes to info model.
[1] granularity of periodic timer. millisec or sec?
Tim - BBF uses msec for diagnostics. so can do repetitive things like ping.
Juergen - are you triggering separate tasks at high freq?
Tim - implementation detail.
Joachim Fabini ? time slotted network want millisec to avoid conflicts.
various nits
Open issues - main one:
Config and reporting data models.
written to use notifications for reporting. needs open connex or call home ie
push data over hanging request, event stream over the response. not pretty
though do-able. typically implementations do rest call. we should do this
Juergen asked the YANG doctors. They say model as RPC. Same on wire - post to
collector but yang mechanics are simpler. Tim - we've replied on other
transport protocols like ftp, http more efficeint as can compres for large data
sets. profiles for flexibility. Juergen - ietf prefers one answer.

Dan - netconf discussions
Jason - do you expect DM to trigger any changes to DM?
Juergen - not from the reporting discussion.
Dan - what's left?
J- reporting, as discussed. Some clean-ups.
Dan - keeping reporting and control in same doc?
Juergen - next version in sept.
Dan - maybe interim in mid sept. virtual. BBF meeting in early sept. Seems too
early.

5. LMAP Protocol? Juergen

protocol comparison section can be removed.
editorials.

Dan - is H1 ok for updates?
Juergen - H2 sept for updates.
Oct 19 is cut-off for I-D for IETF-94

6. Use Cases for Collaborative LMAP - Rachel Huang

multiple autonomous measurement systems collaborate for measurement. multiple
controllers is outside charter at moment. e.g. content provider collaborate
with isp to guide design or better e2e performance troubleshooting. derived
some requirements. would like feedback and wg to include in next stage? Dan -
is there any (proprietary) deployment today for collaborative measurements?
Don?t know. Marcelo Bgnulo - do ISPs in room that will deploy lmap want this?
Kamala Subramaniam Microsoft - see next presentation on ISP use case Frode  -
Euro regs studying feasibility of collab. may conclude end of this year. Robert
Kistekeki- mPlane - building protocol do this. Jason  - is MA associated with 2
controllers, or is it inter-controller interface?

7. Router Buffer Sizes in WAN
- Kamala

In the WAN, ~~100 msecs
mine data from operator?s network in order to quantify buffer size (which
costs). Literature doesn't reflect reality that well. Issues are lack of std
way to mine empirical data & lack of concise way to present mined data

Google's geo-replication data - smaller of the networks doing this. least, max
& average latency on the slide. need to stdised these methods would be useful.

Wes George - how to get reasons for the discards?
ans - Use IPV6, slfow, snmp. no single way.
Wes TWC - we've seen whilst router on paper has buffer size but can be much
lower because of way carved up - for qos policy. rounds to 1% - so ~1msec of
buffer ONLY. Dan - discussion is starting to get out of lmap scope. Please be
more specific in your next version of i-d. Matt Mathis - ISPS are doing
optimize of buffer cap vs performance. valuable to them. so telemetry is to
help isp do this optimization. Dan - please send PDF that is OK with in
confidence removed from footer.

8. Internet Measurement System - Motoyuki Ooki

we're measuring e2e performance - sharing status. also propose lmap extensions.
set up is similar but not same as lmap.
controller supplies schedule with high priority
making ~20k measurements per week
Extend in future to measurement between different AS

Dan - we should open thread on inter-domain since several people have mentioned.
concerns about authentication on border etc. security considerations are needed
Dan - please show what's changed rather than all stuff.

Matt M - big opportunity is you want to take on
Dan - future work, so worth so
phil - out of scope is MA with multiple independent controllers. Measuring
across domains is ok

10. Next steps and open mic

Alissa - important to finish pending work. some low energy on some work items.
make sure get reviews from outside authors etc.

Vic Liu - protocol draft. our draft is based on yang protocol, so can merge our
draft with it. Dan - WG has made decision about protocol. it is not open. 
please make your comments in context of this protocol