Benchmarking Methodology WG (bmwg)
TUESDAY, March 8, 2005, 0900-1130
CHAIRS: Kevin Dubray
About 30 people attended the session in person, with several more
joining the jabber group and others simply listening to the audio stream.
The Chairs supplied a URL for all slides provided in advance of the meeting,
to foster remote participation.
The following minutes were prepared by Kevin Dubray and Al Morton,
based on the detailed account supplied by Jeff Dunn as official note-taker,
and the audio archive itself.
0. Agenda bashing
Tom Alexander proposed to share status slides on
the progress of IEEE 802.11 T on benchmarking wireless subnets.
His offer was accepted (time permitting) and we deferred the Chairs'
WLAN Status slide to the end of the meeting to introduce the topic.
1. Working Group Status (Chairs)
Here's the state of WG I-D's and activities at IETF 62:
WG Last Call
, ended 3/14 w/comments
, New Editors, revised
Revised on input
, Revised, new editor
Revised, LC imminent
Revised, LC imminent
Expired BMWG I-Ds
, Pending term. progress
Pending term prog.
New Work proposals.
LDP Convergence - need I-D revision.
Protection benchmarking - I-D pending?
RFC Editor Queue --
* required NORMATIVE reference resolution - resolved after meeting
See the BMWG mail archive for recent revision status comments:
With a revised version for this meeting, the IPsec terminology draft
is thought complete by its editors, and will be tested in WGLC.
The benchres term draft underwent WGLC recently, and the editors have
some comments to address. Another WGLC is expected shortly.
All editors were encouraged to consult the "tools" prototype
web pages to see the results of idnits checking:
For example, all drafts are required to have an IANA section,
and nits must be resolved before AD-level submission.
2. Revised Milestones. (Dubray)
We added a new milestone on Hash and Stuffing draft, and one
milestone is overdue (IPsec terminology draft to AD review).
Many other milestone dates are drawing near.
3. IGP Data plane convergence benchmark I-Ds. (S.Poretsky)
The 05 versions reflect final comments from IETF 61.
Scott presented a version history and the summarized the
changes in all drafts. He asked to test readiness for another
(possibly final) WG Last Call. There were several comments and
questions from the floor:
- the factor of 5 in equation 3 was suggested in earlier comments
- there may be some interesting test cases when using different
interface media types, but the current draft requires the same types.
This point should not impede progress.
- There is an IESG mandate to handle IPv6 on par with IPv4
- There should be a test case with a smaller number of routes (to vet config)
- Multiple ingress interfaces must share the Forwarding Capacity,
otherwise congestion will result
- Establish the Maximum Offered Load with a test up-front to set the baseline
4. Tech. for Benchmarking Core Router Accelerated Life Testing (S.Poretsky)
The Terminology & Methodology I-Ds were revised on comments.
Scott summarized the changes in these drafts. The Terminology was
clarified with a figure, and there are six new test cases in
the Methodology (where some additional work is planned).
He asked if the group was ready to begin the WGLC process on the Terms,
and this seems to be plausible. However, there were comments on the
Whitebox terms in Appendix I, where a uniform standard for describing
memory is needed (only CPU memory was intended). Regarding the
Methodology, there were several comments and questions from the floor:
- Suggestions for new test cases: Security, Authentication,
Encryption, IPv6 Tunnel, Policy Case with filters and ACLs.
- With so many test cases, the memo needs to organize them into groups
(e.g., all BGP cases). Tests could be grouped by purpose and split into
different drafts. It was suggested to start with the generic test cases
for simplicity and follow-up with more specific cases in separate memos.
- Section 4.6, method with Route Flap Dampening specifies DEFAULT
parameters, but this needs clarification, perhaps a single set of values
will lead to simpler comparison. RFC 2439 should be referenced.
- Each benchmark collected needs a method of measurement specified,
beyond the procedures given in the memo. For example, there is no method
to collect "Recovered Sessions Lost". The exact methods to measure
time intervals and rates are not defined.
5. Terminology for Benchmarking Network-layer Traffic Control Mechanisms
After a series of active Last Calls, a list of editorial issues have
been addressed in version 10, and the new Editors believed the work complete.
The idnits checker has identified some additional work, and if these
edits are completed in version 11, this memo will be ready for (possibly
the final) WG Last Call.
6. FIB Methodology (J.Dunn)
Jeff described the approach to bringing this memo to consensus, including
the addition of ISO 9646 formalism (test groups). It was clarified that
Verdicts only apply to the qualifications on a benchmark, and not the
benchmark itself (because BMWG does not establish pass/fail criteria).
There must be some quantitative criterion, such as "how long does it take?"
This draft will reuse many of the Dataplane convergence definitions.
Readership is encouraged here, there has been no comment on the list.
One of the next steps is to modify RFC 3222, FIB terminology.
Kevin asked that proposals for this be taken to the list.
Another future activity might be a RIB to FIB convergence draft.
This topic spawned considerable interest, but it is seen to be rather
difficult to observe externally. Methods could include observation of SNMP and
routing updates. Again, Kevin asked that participants post their questions
and suggestions for this topic on the list.
7. IPv6 Benchmarking Introduction (C. Martin)
Cynthia led a short discussion of IPv6 benchmarking terminology work.
A link to her draft was posted to BMWG mailing list, and the draft will
be resubmitted as an individual submission shortly after the meeting.
Cynthia planned to treat both Dual-Environment and IPv6-only aspects,
but a suggestion to split this work into separate drafts was supported
by the attendees and chairs. There are many transition mechanisms, and
they are expected to evolve with experience, while IPv6-only configs are
a stable target. Also, smaller drafts require less time to review,
allow easier course corrections, and tighter scopes may lead to
quicker consensus, so the split was agreed. Once the draft is posted
in the I-D directories, we can have discussions on the list and likely
follow-up with a new work proposal.
8. LDP Convergence (T. Eriksson)
Thomas presented the changes to the current version (from October), new terms
may be needed when the methodology is more mature. He then presented
some equipment configurations for the methodology in order to get feedback.
Tests of the Egress LSR will not see labeled traffic when PHP (Penultimate
Hop Popping) is used, and this is the norm for deployment today. There
could be two test cases, one with PHP enabled and one with labeled traffic.
There was a suggestion to measure LSP flapping, but the authors felt this
was better included under the Accelerated Stress testing work, since a
single convergence time benchmark is sought. There was a comment that the
considerations identified in draft-white-network-benchmark-01 that should
be taken account here. There has been commentary that these considerations
are useful. Unfortunately, this individual draft had a wider scope
than either BMWG or IPPM charters, and there was no new benchmark defined
in the text.
9. Automatic Protection Switching Proposal
Al gave a brief status of this item: At IETF-61, we discussed this item and there was
reasonable support to take-up this work. But there were comments identified that must
be dealt with before we proceeded with a work proposal.
Then we went into "suspended animation", and when we see the revisions to
the drafts we'll take this up at the proposal stage. Scott Poretsky noted
that there have been some comments off-list, so the work is progressing,
but off-list commentary is not part of our record unless it's brought-in
by re-posting the comments with permission, etc.
10. Status of Unofficial Liaison with 802.11"T", and their progress to date.
Following discussions in BMWG at IETF-61 (where there was considerable
interest expressed in this work), IEEE 802.11 Task Group T met and decided
that all of the work proposed in draft-alexander-wlan-meth-00 was within their
scope. This was communicated to the BMWG chairs by the IEEE 802 liaison
Bernard Aboba, who offered that the work could still be reviewed in BMWG
and that an informal liaison could be established if someone volunteered
to perform the role as go-between. Fortunately, Tom Alexander (co-author
of the methodology draft with Scott Bradner), volunteered and joined the
BMWG session in person. Tom presented his perspective on the status of
802.11 Task Group T (see slides). The output of this group will be a
Recommended Practice (SHOULD, not mandatory). The membership is broad, and
meetings are frequent, with weekly teleconferences, ad hocs, and 10-16 hours
scheduled during plenary weeks. There have been many presentations of
the methods that vendors are using/selling today, a wide range of approaches.
There has been progress on the ground rules for voting, but the first draft
terminology will not be ready for next week's meeting as scheduled. The group
will first define terminology and then work methodologies.
There are two processes that BMWG participants can follow to review and comment
on this topic. There is no draft yet available in the members-only
section of the www.802wirelessworld.com site, but when there is, the BMWG
chairs can request a password and communicate it to interested individuals
who will be expected to provide comments in return. Also, contributions
can be viewed now by anyone who registers their e-mail on the wirelessworld
site. Dorothy Stanley noted that the TG T Chairman's Agenda slides cover site
access. If BMWG input is sought early-on, then Tom Alexander will notify
BMWG list of the contribution and the topic or issue. For example, Task
Group T is beginning their work by defining terminology and has begun to
borrow terms from BMWG RFCs, so clarifications on terminology may be sought
from BMWG practicioners. Noting that permission is needed to use IETF RFCs
in this way, Scott Bradner gave his permission (as editor) to use the text
of RFC 2544.
There was a question about OAM tool definitions, and Tom answered that this
work is in progress under 802.11k. Also, Dorothy Stanley agreed to give a
walk-through of the wireless world site immediately following the meeting.
11. Wrap-up (Chairs)
Quite a few drafts are headed for WGLC, and folks should expect that we
will use the Active Review Templates developed last year, and that
*volunteers* will be tapped to provide detailed review. The chairs will
investigate an issue-tracker, or bug-tracker for our drafts.
They will also consider a charter re-fresh with the ADs. We might be able
define a "benchmark" more crisply. There have been some other forms of
testing that seem to be creeping into the group, and we need to clarify
BMWG's role in this area.