Benchmarking Methodology Working Group (bmwg) - IETF 119 Brisbane

Date and Time: 21/03/2024 Thursday 13:00-15:00 AEST
Location: M1, Brisbane Convention & Exhibition Center
Chairs: Sarah Banks, Giuseppe Fioccola
AD: Warren Kumari
Zulip Stream: https://zulip.ietf.org/login/#narrow/stream/bmwg

Note taker(s), Jabber, IPR

Paolo Volpato volunteered.

Administrative and WG Status (Chairs)

WG Drafts

Multiple Loss Ratio Search + Test results, Maciek Konstantynowicz and Vratko Polak

draft-ietf-bmwg-mlrsearch

Draft update and test results shared. Recap on MLRsearch to improve
RFC2544 search, especially for software platforms. Test results also
discussed.
Draft considered ready for review by the authors that would like to
progress towards Working Group Last Call.

Giuseppe Fioccola: thanks for the presentation and for sharing the
results. In the architecture there is a controller. Are you considering
any protocols or YANG model for tester configuration?
Vratko Polak: it is doable but manager-controller is based on libraries
dependent on implementation and options. So, integration between the two
can be formalized.
Giuseppe Fioccola: please add these consideration in the section about
the controller.
Vratko Polak: not sure if it is in scope, but another draft can specify
it.
Giuseppe Fioccola: you can simply add a statement and specify that it is
out of scope so that the reader is aware of that.
Giuseppe Fioccola: WGLC Request to be discussed on the list based on
further review.

A YANG Data Model for Network Tester Management + Hackathon results, Vladimir Vassilev

draft-ietf-bmwg-network-tester-cfg

Network tester management YANG Data Model discussed. Current tools: CLI
(SCPI), CiscoTRex, Keysight Open API.
The code is available for the Clients and Server side (GitHub).
YANG trees presented. Two YANG models: analyzer and generator. Hackathon
with hardware simulator: 1.5ms for simplified tests (10KB traffic).

Vratko Polak: you mentioned you did lot of tooling not using Netconf but
Rest, but in the other I-D allows data to be formatted in JSON instead
XML. Both are ok to users.
Vladimir Vassilev: different servers, the good thing is that the can all
support the module. XML difficult to read, provided an example.
Giuseppe Fioccola: clarification. Looking at your YANG model is not very
detailed. Can it be augmented by future models?
Vladimir Vassilev: we provided the freedom to the user to specify the
frame representation. Details should be specified by testers with
representation is hexadecimal. You understood correctly but we don’t
have limitations to generate the solution.
Giuseppe Fioccola: I see. It is possible to generate any kind of frame,
but, eventually, augmentation is still possible in the future.

Considerations for Benchmarking Network Performance in Containerized Infrastructures, Minh-Ngoc Tran

draft-ietf-bmwg-containerized-infra

Scope of I-D shared, applicable to containerized VMs. Networking models,
resources configuration discussed. Additions on Kernel, user space
acceleration, eBPF, Smart-NIC acceleration.

Luis Contreras: are you considering recommendation about the size of the
packets? The size of the packets can have impacts on the performance.
Minh-Ngoc Tran: it was addressed in previous I-D.

Proposals

Recommendations for using Multiple IP addresses in Benchmarking Tests, Gabor Lencse

draft-lencse-bmwg-multiple-ip-addresses

Problem description for using Multiple IP addresses in Benchmarking
Tests introduced. The importance of address and port randomization is
proved by measurements.

Vratko Polak: there is a 3rd type of RSS related to IPv6 and is the flow
label.
Gabor Lencse: it is interesting, please send me the reference on the
list.
Vratko Polak: it cannot be important for the tester, but between 2 DUTs
when using same encapsulation.

Benchmarking methodology for Segment Routing, Luis M. Contreras

draft-vfv-bmwg-sr-bench-meth

Draft overview introduced after the merge of the two previous I-Ds.
History discussed and main changes shown.

Gabor Lencse: I was ok with separate drafts and with merged draft. I
have read the draft and I believe it is ready for the adoption call.
Giuseppe Fioccola: Thank you for the feedback. I will discuss with Sarah
how to progress it.

Characterization and Benchmarking Methodology for Power in Networking Devices, Carlos Pignataro or Qin Wu

draft-cprjgf-bmwg-powerbench

Background discussed. Terminology and benchmarking defined and
methodology discussed. Initial benchmarking tests listed and discussed
during the E-IMPACT interim meeting.

Giuseppe Fioccola: it could be good to cross-post the discussion when it
is on the e-impact mailing list.
Qin Wu: we also setup a GitHub repo.

SRv6 Service Benchmarking Guideline, Xuesong Geng

draft-geng-bmwg-srv6-service-guideline

Background shown. A very long list of services is classified to BE,
Policy, and Policy with Compression.
It is a guideline document. Seek feedback and comments from WG.

Luis Contreras: is the draft intended for interoperability?
Xuesong Geng: it is for a single vendor.
Luis Contreras: to reproduce tests IP and topology are included?
Xuesong Geng: we are thinking of a separate document to discuss it.

AOB