I commented at the WG meeting, and would like to add a few comments here:
On 07/02/2019 13:27, Zaheduzzaman Sarker wrote:
Hi Stewart,
Thanks for a good review.
For the security consideration section, we can use stronger words if that is
required. This document merely specifies test cases when people are testing
their algorithm in a controlled environment and does not specify protocol
usage. I was wondering if using normative language is an overkill here. For
those reasons we are actually thinking of taking out 2119 usage completely. I
have a modified text proposal below.
Please see inline below for more.
BR
Zahed
On 2019-02-04, 20:29, "Stewart Bryant" <stewart.bry...@gmail.com> wrote:
Reviewer: Stewart Bryant
Review result: Almost Ready
I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please treat these comments just
like any other last call comments.
For more information, please see the FAQ at
<https://trac.ietf.org/trac/gen/wiki/GenArtfaq>.
Document: draft-ietf-rmcat-eval-test-08
Reviewer: Stewart Bryant
Review Date: 2019-02-04
IETF LC End Date: 2019-02-11
IESG Telechat date: Not scheduled for a telechat
Summary:
A well written documents an close to being ready for publication.
I am concerned that the Security section is weak on use outside a
controlled environment.
There are a fair number of minor issues and nits that need attention,
but most of them are simple to fix.
One concern that I have that I doubt is readily fixable is that long
multi-nested lists do not work well in paginated ASCII with line spaces
and sometimes it is difficult to be sure of the context of a test element
note.
[ZS] I share the concern here, but I don’t think I have a better alternative
now.
Major issues:
8. Security Considerations
The security considerations in [I-D.ietf-rmcat-eval-criteria] and the
relevant congestion control algorithms apply. The principles for
congestion control are described in [RFC2914], and in particular any
new method MUST implement safeguards to avoid congestion collapse of
the Internet.
The evaluation of the test cases are intended to be run in a
controlled lab environment.
SB> I wonder if there shouldn't me a MUST in that sentence?
SB> There have been issues on SP networks with users running unsuitable
SB> performance benchmarks on live networks, including complaints to the
SB> operators concerning the results achieved.
I'm not sure I agree:
To me, the MUST is appropriate for an IETF-defined method that
could be deployed on the general Internet - by design, or by generating
traffic flowing over a virtual network interface. I'd be really concerned
if an IETF-defined mechanism didn't protect from congestion collapse.
Whether that could mean implementing a dynamic CC algorithm, or a
circuit-breaker like function.
If the work targets use in constrained environment, as it might in
this case then other mechanisms are needed to prevent this
traffic from appearing in the Internet.
Hence, the applications, simulators and
network nodes ought to be well-behaved and should not impact the
desired results. It is important to take appropriate caution to
avoid leaking non-responsive traffic from unproven congestion
avoidance techniques onto the open Internet.
SB> Again I am surprised this is not much stronger in prohibiting
SB> use on the Internet.
[ZS] what about :
" The evaluation of the test cases are intended to be run in a
controlled lab environment. Hence, the applications, simulators and
network nodes must be well-behaved and should not impact the
desired results. Moreover, proper measures must be taken to
avoid leaking non-responsive traffic from unproven congestion
avoidance techniques onto the open Internet. "
I'm not sure that's quite there - see above.
========
Minor issues:
This memo describes a set of test cases for evaluating congestion
control algorithm proposals for real-time interactive media.
SB> It would be useful to add here the statement in the abstract that
SB> these tests should be done in a controlled environment.
[ZS] The abstract mentions the this "This document describes
the test cases to be used in the performance evaluation of such
congestion control algorithms in a controlled environment."
We can repeat that in the intro as well.
===========
Expected behavior: depending on the convergence observed in test case
5.1 and 5.2, the candidate algorithm may be able to avoid congestion
collapse. In the worst case, the media stream will fall to the
minimum media bit rate.
SB> Do you need to specify the variant of TCP? You do state it later, but some
comment here would be useful.
[ZS] Not sure what would be useful to describe here more, the expected
behaviour is not really coupled to what TCP congestion control is used, the
general demand is to avoid congestion collapse here.
SB> What behaviour do you expect the TCP to show.
It would be bad if SB> an aggressive media application kill the TCP
completely.
[ZS] I don’t think we should say anything about TCP behaviour here. The idea is
to test the new congestion control behaviour with available TCP behaviours not
vice versa. But TCP can certainly improve its performance from the test results
(.
maybe reference to RFC2914 would help?
============
the first flow
(S1) MUST arrive at a steady-state rate approximately twice of that
of the other two flows (S2 and S3).
SB> I am not sure what you mean by priority I assume that you mean
SB> QoS ranking in the routing system. In which case I don't see
SB> how you can expect the result you specify.
[ZS] no this is not about routing priority on the network nodes, this is at the
media sender. When a media sender has multiple flows that shares the same
bottleneck then the media sender can use techniques to distribute the available
bandwidth to the multiple flows that it is sending. The point here is the media
flows should get their share of the available bandwidth as per the priority set
by the application.
Perhaps: /bandwidth/capacity/ - or at least be very careful in cases
where bandwidth is used.
============
Expected behavior: the candidate algorithm is expected to achieve
full utilization at both bottleneck links without starving any of the
three congestion controlled media flows.
SB> I am not sure what you mean by this. Do you mean that the bottlenecks
SB> will saturate, but make no comment about how much of the bottleneck
SB> capacity each flow gets for itself?
[ZS] bottlenecks will saturate -- yes. The success criteria --- the existence
of multiple bottleneck should not result in flow starvation to any flows that
is sharing those bottlenecks. Yes, there is no comment on fairness here
explicitly. Not sure we need one here, do we?
============
Nits/editorial comments:
[ZS] thanks for those nits. will take care of them.
Checking nits according to https://www.ietf.org/id-info/checklist :
----------------------------------------------------------------------------
** There are 9 instances of too long lines in the document, the longest one
being 4 characters in excess of 72.
== Outdated reference: A later version (-06) exists of
draft-ietf-rmcat-wireless-tests-05
===========
3. Structure of Test cases
SB> In the text below it was sometimes hard to get the context right in the
SB> triple (or more) nested list. Please consider using subsections or some
other SB> demarcation.
===========
+ Bottleneck queue type: for example, Droptail, FQ-CoDel, or
PIE.
SB> There need references, and by convention expansion on first use.
==========
+ Path loss ratio: characterizes the non-congested, additive,
losses to be generated on the end-to-end path. MUST
SB> s/MUST/This MUST/ ?
==========
B. Variation in sending bit rate and goodput. Mainly observing
the frequency and magnitude of oscillations.
SB> goodput needs a reference or a definition. I don't think it is a
universally known term.
===========
Expected behavior: the candidate algorithm is expected to detect the
path capacity constraint, converges to the bottleneck link's capacity
SB> s/converges/converge/
===========
Due to asymmetric nature of the link
SB> s/Due to/Due to the/
===========
SB> Is there a diagram error in the figure above?
Figure 6: Testbed Topology for TCP vs congestion controlled media
Flows
===========
have the same bandwidth share on the link. It has to make it's way
SB>s/it's/its/
===========
The candidate algorithm MUST reflect the relative priorities
assigned to each media flow. In the previous example,
SB> An explicit reference to the test would help the reader
==========
_______________________________________________
Gen-art mailing list
Gen-art@ietf.org
https://www.ietf.org/mailman/listinfo/gen-art
