IP Performance Metrics WG (ippm)
Monday, November 7, 2005 - 09:00--11:30
=======================================
The meeting was chaired by Henk Uijterwaal and Matt Zekauskas. Benoit Claise
and Al Morton took notes, which were edited into these minutes by the
chairs.
AGENDA:
1. Administrivia: Agenda, Status of Drafts and Milestones
2. Temporal Aggregation (and Composition Framework)
draft-svdberg-ippm-temporal-00.txt
3. Spatial Composition Draft and an overall Framework Proposal
draft-morton-ippm-composition-00.txt
4. Spatial and Multicast Metrics
draft-stephan-ippm-multimetrics-02.txt
5. Jitter Definitions discussion
6. Traceroute Metrics and Data Model
draft-niccolini-ippm-storetraceroutes-02.txt
7. TWAMP Draft
draft-ippm-two-way-active-measurement-protocol-00.txt
8. Recent ITU Liaison statements
Loki Jorgenson was unable to attend, thus he did not present his
opinions on the ITU's IP Performance documents (specifically, Y.1541).
1. Administrivia: Agenda, Status of Drafts and Milestones
--Henk Uijterwaal for both chairs
Henk opened the meeting with the usual working group information
boilerplate. He then talked about the state of the existing drafts,
in particular, that OWAMP has been stuck in a security review (we're
going to talk with Russ Housley; unfortunately, Stanislav Shalunov
couldn't make the Vancouver meeting); and also we're going to need
advice from the ADs as to what (if anything) to do with the
implementation reports we've gathered for the extant IPPM metrics.
Henk did another revision of the implementation report, taking comments
into account. There is work going on in the background on the capacity
draft, but there is no report today.
[After the meeting Russ and Stanislav had a productive email email
exchange solving most of the issues; Stanislav is to work on the last two
after the conference he is attending is over. Henk also sent out
email on the implementation reports after the meeting.]
Henk then discussed the state of the reordering drafts. There are
two drafts, a group draft and a individual submission from a group
at Colorado State University. The latter has been presented but
hasn't drawn much interest from the working group. The group draft
has been reviewed seriously by a number of people, and it has been
stable for about a year.
He also discussed the claim from the CSU group that the byte-offset
definition in the WG draft is derived from reorder-buffer-density.
Henk noted that both he and Matt had reviewed this claim and found
that the editor was reflecting consensus from the working group. It
appears that the two efforts appeared to be working in parallel
towards the same idea from different directions. Also, the ideas had
been publicly discussed at IPPM meetings and elsewhere, and that
therefore it did not appear that byte-offset was directly derived from
reorder-buffer-density (and in any case it is hard to analyze years
after the fact).
Henk proposed that a stable reference to the Colorado work be added
to the draft discussing reordering approaches, that we ensure there
is an acknowledgement to the individuals that participated in meetings
and email discussions (we believe this exists already), and that
we will go to WGLC on the group's draft. Unless the reorder buffer
density draft gets any outside support we will not pursue making
the Colorado draft a working group document. This decision will be
posted to the mailing list shortly. Those present in the room agreed
with this approach.
The dates on the milestones are being revised. Authors should give
Henk good estimates for their milestones.
2. Temporal Aggregation (and Composition Framework)
--Steven Van den Berghe
Steven Van den Berghe (who has been doing some work for GEANT2's JRA1 group)
presented a short draft on thoughts on a unified framework for thinking
about temporal and spatial composition of IPPM metrics, and then some
examples of temporal aggregation, and concerns that need to be addressed.
One example of temporal aggregation is taking a measurement every five seconds,
and then creating a meaningful summary of an hour. [See slides.]
Kaynam Hedayat asked if Steven had thought about jitter. Steven says
that they had, and have some ideas, but haven't worked them out.
Another participant asked about how precedence, or treatment of packets
would fit in. The underlying measurements take that into account,
as part of Type-P. You can say something about one forwarding behavior
directly; aggregating packets with different forwarding behaviors would
take more work, and may not be possible.
Emile Stephan asked about error propagation, and what error should
be reported. Steven said that error propagation needs to be worked
out as well.
Matt asked the group if they thought this would be useful; there
were a few (4-5) nodding heads, including some new participants.
Matt stated that personally, he felt the useful work was the work
that wasn't completed -- error propagation, jitter propagation,
considering values that are not averages.
3. Spatial Composition Draft and an overall Framework Proposal
-- Al Morton
Al Morton presented spatial composition, and a more formal proposal for
a framework for composition (that included discussion with Steven).
This draft's example just talks about composition of finite one-way delays.
The framework is such that you could also include the multimetrics draft
as a different approach within the framework. [See slides.] Al would
like comment on the overall framework, and then the individual drafts.
Roman Krzanowski stated that from his perspective the composition and
decomposition of metrics has been on his mind for a while, and that
providers have no guidance, and he thought we should work on drafts
like this to get some consensus.
4. Spatial and Multicast Metrics
-- Emile Stephan
Emile Stephan then gave an update on the multimetrics draft. This is looking
at measurements that are taken on the same stream at multiple points.
The latest version adds a new metric that more clearly defines a two point
one way delay -- one where you passively measure (possibly a synthetic stream)
at two points. He also discussed where this draft fit with the other two,
and in a general composition framework. He was looking for feedback as
to whether a purely passive metric was the right approach, compared to
something like an applicability statement for the current one-way delay
metric. [See slides.]
The result of the discussion on these three drafts (Van den Berghe, Morton,
Stefan) is that there will be four drafts: A framework draft with common
text, and the three discussing temporal, spatial, and multi-point composition.
5. Jitter Definitions discussion
--Roman Krzanowski
Roman Krzanowski then presented a problem he saw working on an MIT
study of metrics used by multiple operators, both in the US and
abroad. [See slides.] The basic issue was with jitter, and that US
operators tend to use inter-packet delay variation, where the European
operators tend to use delay variation from a single reference point.
The issue is not one of definition -- the IPPM IPDV metric can capture
either, but one of "best practice" or recommendations. He is going to
see if there is enough interest (among US and European operators)
generate an applicability statement -- especially with respect to
sharing measurement results.
Matt said that we'd like to see some interest, and if so set a
deadline to see progress or declare failure since previous efforts to
get providers to agree on sharing results had not been fruitful within
the IETF. Roman said he would summarize the discussion, send it to
the group, and a few select people working for service providers, and
see if we can generate some enthusiasm.
6. Traceroute Metrics and Data Model
--Martin Stiemerling
Martin Stiemerling presented the individual traceroute metric and data model
draft. They have been talking to folks in the GGF Network Measurement
Working Group. This draft shows a complete traceroute schema in XML,
and an alternative one based on the GGF work, and a preliminary sketch
of how to combine approaches. Issues to be resolved include (1) seeing
if the GGF intends to create a standard (2) see if a merged approach with
the GGF structure and IETF information model makes sense (3) if a merged
effort succeeds, should the work be published in the IETF, the GGF, or
both? There is a naming issue - the GGF has it's own way of naming things,
this draft tries to use existing DISMAN names; a common namespace is desirable.
The authors were going to compare the GGF document series with the IETF one
to see if the GGF is even a possible alternative, but the authors strongly
prefer something that uses IETF naming.
There were about 10 bobbing heads when asked if people thought this
was important; it seems clear that the work would be welcome, the
question is if anything should be done within IPPM.
7. TWAMP Draft [no slides]
--Keynam Heyadat
Next, Keynam Heyadat presented the TWAMP draft, and current status.
There have been some changes based on implementation experience, mainly
relating to time stamps, and some text cleanup. The timestamps and
sequence numbers are now in fixed locations so that a hardware implementation
is possible. There was also a correction to the draft; TWAMP packets
fit in ATM cells only in unauthenticated mode.
Henk wanted to make sure the authors are aware of the OWAMP security
considerations so that TWAMP does not run into the same issues (TWAMP
is based on OWAMP).
Keynam stated that he knows of four implementations that are underway;
three of which are substantially done. He feels that after there is
some interoperability experience with them, the draft may be close to
being done. He thought doing a report on implementation experience
would be a good idea.
8. Recent ITU Liaison statements
--Al Morton
Finally Al pointed out the recent Liaison Statements from the ITU to the
group. He showed how to find them, and gave a synopsis of what they were.
Start at the IETF home page, select liaison activities, and then
liaison statements.
The two of interest were one to NSIS, which includes Y1541 (so we can
get it "for free") and the most recent one which talks about a potential
project, G.chirp, which uses packet chirps to characterize available bandwidth
(and eventually the download time for web pages). It's an experimental
thing, but a stable draft is expected about January of 2007. People should
look at them, and let the group know if they have any comments. SG12 doesn't
meet until June of next year, so there is plenty of time to comment.
The NSIS statement: 'Communication of Recommendation Y.1541, "Network
Performance Objectives for IP-based Services"', September 2005.
From ITU-T Study Group 12, Question 17 to NSIS.
https://datatracker.ietf.org/documents/LIAISON/file183.doc
"Question 17/12 has been following the development of the Next Steps In
Signalling (NSIS) framework, requirements and protocols, and notes
with particular interest the work on QSpec Templates and QoS
Models. We understand your desire to reference Recommendation Y.1541
(published 05/02), and so communicate it to you such that a copy can
be referenced and accessed by the IETF community in a persistent way."
NSIS is working on signalling protocols that could communicate
requests for QOS, beyond IP stuff in Y.1541: bandwidth, priority, etc.
This liaison is sent to them, so that they would have a persistent
URL in order to access the recommendation easily for all members
of the IETF.
There are two copies available in the liaison. There is a version
that shows some revisions recently added; classes with additional
performance objectives, for high bandwidth and high sensitivity to
loss applications, such as "big TCPs", and emulation of TDM services
over IP networks. A third kind of thing that these classes support is
transport of IPTV.
The G.chirp statement: "Liaison statement on the Development of New
Recommendation on measuring file download and session time", November 2005.
From ITU-T SG12, Questions 13 and 17, to IPPM.
https://datatracker.ietf.org/public/liaison_detail.cgi?detail_id=148
"Q.13/12 would like to inform IETF IPPM Working Group that we have
started developing a new Recommendation G.chirp, which provides a
means for measuring file download and session time in a simplified
way. The technical content can be found in the attached document
(COM12-C11). We expect to produce a stable draft in January 2007. It
is very much appreciated if your group gives any suggestions or
comments on this topic."
G.chirp is active sending sequence which intends to characterize available
bandwidth, and eventually file download time. It is an
experimental thing that was proposed by TNO of The Netherlands.
The method is described in the contribution. It sends packets in
closer spacing using an algorithm for spacing, and attempts to assess
the download times from that.
A formal reply could be prepared back to SG 12 and the questions.
If folks would like to try out, TNO probably has some intellectual
property covering it. However, trying it out experimentally and
publishing results would probably be reasonable.
The most powerful way to communicate back to the ITU is via additional
liaison statements. If the group looks and thinks it is great idea,
or bad idea, it is best if a response comes back formally.
However, given that certain people participate in both the IETF and
ITU, any response is useful, even if it is informal.
The meeting was then closed.
|