<div dir="ltr"><div dir="ltr"><br></div><div>I have asked a relatively simple question that Jake's got right, so I'm not alone in my own bubble.</div><div>I've asked this to Greg White months ago and his answer was that unresponsive traffic x is assumed to be small.</div><div>That the queue protection mechanism will ensure that.</div><div><br></div><div>You Bob send me a reference, that I checked again, and considers the other extreme case where x is very large.</div><div><br></div><div>I have not found anything that covers the most realistic cases which is case c in my previous message, when x </div><div>just varies normally when you have unresponsive traffic that varies just like in today networks.</div><div><br></div><div>The normal case is the one I'm interested the most. This is something that in LDD for Cable may</div><div>enter the access network of many cable subscribers, surfing the web, using WebEx working form home, doing normal things.</div><div>The document "Destruction testing" :) is not very useful to my use case and specific question.</div><div><br></div><div>We have been promised below millisecond latency up to the 99-th quantile and I have simple technical questions.</div><div>LLD is a new initiative in the public IETF forum so it is normal that people start asking questions. </div><div>No need to get blood pressure above the threshold.</div><div><br></div><div>As long as LLD is a cable industry only thing using PHB and private marking, all this discussion may be irrelevant in this forum,</div><div>but this is not the case.</div><div><br></div><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Jul 4, 2019 at 1:55 PM Bob Briscoe <<a href="mailto:ietf@bobbriscoe.net">ietf@bobbriscoe.net</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex">
<div bgcolor="#FFFFFF">
Luca,<br>
<br>
<br>
<div class="gmail-m_3469002528728889916moz-cite-prefix">On 19/06/2019 14:02, Luca Muscariello
wrote:<br>
</div>
<blockquote type="cite">
<div dir="ltr">Jake,
<div><br>
</div>
<div>Yes, that is one scenario that I had in mind. </div>
<div>Your response comforts me that I my message was not totally
unreadable. </div>
<div><br>
</div>
<div>My understanding was</div>
<div>- There are incentives to mark packets if they get
privileged treatment because of that marking. This is similar
to the diffserv model with all the consequences in terms of
trust.</div>
</div>
</blockquote>
[BB] I'm afraid this is a common misunderstanding. We have gone to
great lengths to ensure that the coupled dualQ does not give any
privilege, by separating out latency from throughput, so:<br>
<ul>
<li>It solely isolates traffic that gives /itself/ low latency
from traffic that doesn't.</li>
<li>It is very hard to get any throughput advantage from the
mechanism, relative to a FIFO (see further down this email).</li>
</ul>
The phrase "relative to a FIFO" is important. In a FIFO, it is of
course possible for flows to take more throughput than others. We
see that as a feature of the Internet not a bug. But we accept that
some might disagree...<br>
<br>
So those that want equal flow rates can add per-flow bandwidth
policing, e.g. AFD, to the coupled dualQ. But that should be (and
now can be) a separate policy choice. <br>
<br>
An important advance of the coupled dualQ is to cut latency without
interfering with throughput.<br>
<br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div>- Unresponsive traffic in particular (gaming, voice, video
etc.) has incentives to mark. Assuming there is x% of
unresponsive traffic in the priority queue, it is non trivial
to guess how the system works.</div>
<div>- in particular it is easy to see the extreme cases, </div>
<div> (a) x is very small, assuming the system is
stable, the overall equilibrium will not change. </div>
<div> (b) x is very large so the dctcp like
sources fall back to cubic like and the systems behave almost
like a single FIFO.</div>
<div> (c) in all other cases x varies according to
the unresponsive sources' rates. </div>
<div> Several different equilibria may exist,
some of which may include oscillations. Including oscillations
of all fallback mechanisms.</div>
<div>The reason I'm asking is that these cases are not discussed
in the I-D documents or in the references, despite these are
very common use cases.</div>
</div>
</blockquote>
[BB] This has all already been explained and discussed at length
during various IETF meetings. I had an excellent student (Henrik
Steen) act as a "red-team" guy. His challenge was: Can you contrive
a mis-marking strategy with unresponsive traffic to cause any more
harm than in a FIFO? We wanted to make sure that introducing a
priority scheduler could not be exploited as a significant new
attack vector.<br>
<br>
Have you looked at his thesis - the [<a href="https://tools.ietf.org/html/draft-ietf-tsvwg-aqm-dualq-coupled-09#ref-DualQ-Test" title=""Destruction Testing: Ultra-Low Delay using Dual Queue
Coupled Active Queue Management"" target="_blank">DualQ-Test</a>] reference
at the end of this subsection of the Security Considerations in the
aqm-dualq-coupled draft:<br>
<a href="https://tools.ietf.org/html/draft-ietf-tsvwg-aqm-dualq-coupled-09#section-4.1.3" target="_blank">4.1.3.
Protecting against Unresponsive ECN-Capable Traffic</a> ?<br>
(we ruled evaluation results out of scope of this already over-long
draft - instead giving references).<br>
<br>
Firstly, when unresponsive traffic < link rate,
counter-intuitively it doesn't matter which queue it classifies
itself into. Any responsive traffic in either or both queues still
shares out the remaining capacity as if the unresponsive traffic had
subtracted from the overall capacity (like a FIFO). <br>
<br>
Beyond that, Henrik tested whether the persistent overload mechanism
that switches off any distinction between the queues (<a href="https://github.com/L4STeam/sch_dualpi2_upstream/blob/master/net/sched/sch_dualpi2.c" target="_blank">code
in the reference Linux implementation</a>, <a href="https://tools.ietf.org/html/draft-ietf-tsvwg-aqm-dualq-coupled-09#appendix-A.2" target="_blank">pseudocode
and explanation in Appendix A.2</a>) left any room for mis-marked
traffic to gain an advantage before the switch-over. There was a
narrow region in which unresponsive traffic mismarked as ECN could
strengthen its attack relative to the same attack on the Classic
queue without mismarking. <br>
<br>
I presented a one-slide summary of Henrik's experiment here <a href="https://datatracker.ietf.org/meeting/99/materials/slides-99-tcpm-ecn-adding-explicit-congestion-notification-ecn-to-tcp-control-packets-02#page=12" target="_blank">in
2017 in IETF tcpm</a>.<br>
I tried to make the legends self-explanatory as long as you work at
it, but shout if you need it explained.<br>
Each column of plots shows attack traffic at increasing fractions of
the link rate; from 70% to 200%.<br>
<br>
Try to spot the difference between the odd columns and the even
columns - they're just a little different in the narrow window
either side of 100% - a sharp kink instead of a smooth kink. <br>
I included log-scale plots of the bottom end of the range to magnify
the difference.<br>
<br>
Yes, the system oscillates around the switch-over point, but you can
see from the tcpm slide that the oscillations are also there in the
3rd column (which emulates the same switch-over in a FIFO). So we
haven't added a new problem.<br>
<br>
In summary, the advantage of mismarking was small and it was hard
for the attacker not to trip the dualQ into overload state when it
applies the same drop level in either queue. And that was when the
victim traffic was just a predictable long-running flow. With normal
less predictable victim traffic, I cannot think how to get this
attack to be effective.<br>
<br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div>If we add the queue protection mechanism, all unresponsive
flows that are caught cheating are registered in a blacklist
and always scheduled in the non-priority queue.</div>
</div>
</blockquote>
[BB] <br>
1/ Queue protection is an alternative to overload protection, not an
addition. <br>
<ul>
<li>The Linux implementation solely uses the overload mechanism,
which is sufficient to prevent the priority scheduler amplifying
a mismarking attack (whether ECN or DSCP).</li>
<li>The DOCSIS implementation use per-flow queue protection
instead.<br>
</li>
</ul>
2/ Aligned incentives<br>
<br>
The coupled dualQ with just overload protection ensures incentives
are aligned so that, normal developers won't intentionally mismark
traffic. As explained at the start of this email:<br>
<blockquote>the DualQ solely isolates traffic that gives /itself/
low latency from traffic that doesn't. Low latency solely depends
on the traffic's own behaviour. Traffic doesn't /get/ anything
from the low latency queue, so there's no point mismarking to get
into it.<br>
</blockquote>
However, incentives only address rational behaviour, not accidents
and malice. That's why DOCSIS operators asked for Q protection - to
protect against something accidentally or deliberately introducing
bursty or excessive traffic into the low latency queue.<br>
<br>
The Linux code is sufficient under normal circumstances though.
There are already other mechanisms that deal with the worms,
trojans, etc. that might launch these attacks. <br>
<br>
3/ DOCSIS Q protection does not black-list flows. <br>
<br>
It redirects certain /packets/ from those flows with the highest
queuing scores into the Classic queue, only if those packets would
otherwise risk a threshold delay for the low latency queue being
exceeded. <br>
<br>
If a flow has a temporary wobble, some of its packets get redirected
to protect the low latency queue, but if it gets back on track, then
there's just no further packet redirection. <br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div>It that happens unresponsive flows will get a service
quality that is worse than if using a single FIFO for all
flows.</div>
</div>
</blockquote>
4/ Slight punishment is a feature, not a bug<br>
<br>
If an unresponsive flow is well-paced and not contributing to
queuing, it will accumulate only a low queuing score, and experience
no redirected packets.<br>
<br>
If it is contributing to queuing and it is mismarking itself, then Q
Prot will redirect some of its packets, and the continual reordering
will (intentionally) give it worse service quality. This deliberate
slight punishment gives developers a slight incentive to mark their
flows correctly.<br>
<br>
I could explain more about the queuing score (I think I already did
for you on these lists), but it's all in Annex P of <a href="https://specification-search.cablelabs.com/CM-SP-MULPIv3.1" target="_blank">the
DOCSIS spec</a>. and I'm trying to write a stand-alone document
about it at the moment.<br>
<br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div><br>
</div>
<div>Using a flow blacklist brings back the complexity that
dualq is supposed to remove compared to flow-isolation by
flow-queueing.</div>
<div>It seems to me that the blacklist is actually necessary to
make dualq work under the assumption that x is small, </div>
</div>
</blockquote>
[BB] As above, the Linux implementation works and aligns incentives
without Q Prot, which is merely an optional additional protection
against accidents and malice.<br>
<br>
(and there's no flow black-list).<br>
<br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div>because in the other cases the behavior</div>
<div>of the dualq system is unspecified and likely subject to
instabilities, i.e. potentially different kind of
oscillations. <br>
</div>
</div>
</blockquote>
<br>
I do find the tone of these emails rather disheartening. We've done
all this work that we think is really cool. And all we get in return
is criticism in an authoritative tone as if it is backed by
experiments. But so far it is not. There seems to be a presumption
that we are not professional and we are somehow not to be trusted to
have done a sound job.<br>
<br>
Yes, I'm sure mistakes can be found in our work. But it would be
nice if the tone of these emails could become more constructive.
Possibly even some praise. There seems to be a presumption of
disrespect that I'm not used to, and I would rather it stopped.<br>
<br>
Sorry for going silent recently - had too much backlog. I'm working
my way backwards through this thread. Next I'll reply to Jake's
email, which is, as always, perfectly constructive.<br>
<br>
Cheers<br>
<br>
<br>
Bob<br>
<br>
<blockquote type="cite">
<div dir="ltr">
<div>Luca</div>
</div>
</blockquote>
<blockquote type="cite">
<div dir="ltr"><br>
<div><br>
</div>
<div><br>
</div>
</div>
<br>
<div class="gmail_quote">
<div dir="ltr" class="gmail_attr">On Tue, Jun 18, 2019 at 9:25
PM Holland, Jake <<a href="mailto:jholland@akamai.com" target="_blank">jholland@akamai.com</a>> wrote:<br>
</div>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex">Hi
Bob and Luca,<br>
<br>
Thank you both for this discussion, I think it helped
crystallize a<br>
comment I hadn't figured out how to make yet, but was
bothering me.<br>
<br>
I’m reading Luca’s question as asking about fixed-rate traffic
that does<br>
something like a cutoff or downshift if loss gets bad enough
for long<br>
enough, but is otherwise unresponsive.<br>
<br>
The dualq draft does discuss unresponsive traffic in 3 of the
sub-<br>
sections in section 4, but there's a point that seems sort of
swept<br>
aside without comment in the analysis to me.<br>
<br>
The referenced paper[1] from that section does examine the
question<br>
of sharing a link with unresponsive traffic in some detail,
but the<br>
analysis seems to bake in an assumption that there's a fixed
amount<br>
of unresponsive traffic, when in fact for a lot of the
real-life<br>
scenarios for unresponsive traffic (games, voice, and some of
the<br>
video conferencing) there's some app-level backpressure, in
that<br>
when the quality of experience goes low enough, the user (or a
qoe<br>
trigger in the app) will often change the traffic demand at a
higher<br>
layer than a congestion controller (by shutting off video, for<br>
instance).<br>
<br>
The reason I mention it is because it seems like unresponsive<br>
traffic has an incentive to mark L4S and get low latency. It
doesn't<br>
hurt, since it's a fixed rate and not bandwidth-seeking, so
it's<br>
perfectly happy to massively underutilize the link. And until
the<br>
link gets overloaded it will no longer suffer delay when using
the<br>
low latency queue, whereas in the classic queue queuing delay
provides<br>
a noticeable degradation in the presence of competing traffic.<br>
<br>
I didn't see anywhere in the paper that tried to check the
quality<br>
of experience for the UDP traffic as non-responsive traffic
approached<br>
saturation, except by inference that loss in the classic queue
will<br>
cause loss in the LL queue as well.<br>
<br>
But letting unresponsive flows get away with pushing out more
classic<br>
traffic and removing the penalty that classic flows would give
it seems<br>
like a risk that would result in more use of this kind of
unresponsive<br>
traffic marking itself for the LL queue, since it just would
get lower<br>
latency almost up until overload.<br>
<br>
Many of the apps that send unresponsive traffic would benefit
from low<br>
latency and isolation from the classic traffic, so it seems a
mistake<br>
to claim there's no benefit, and it furthermore seems like
there's<br>
systematic pressures that would often push unresponsive apps
into this<br>
domain.<br>
<br>
If that line of reasoning holds up, the "rather specific"
phrase in<br>
section 4.1.1 of the dualq draft might not turn out to be so
specific<br>
after all, and could be seen as downplaying the risks.<br>
<br>
Best regards,<br>
Jake<br>
<br>
[1] <a href="https://riteproject.files.wordpress.com/2018/07/thesis-henrste.pdf" rel="noreferrer" target="_blank">https://riteproject.files.wordpress.com/2018/07/thesis-henrste.pdf</a><br>
<br>
PS: This seems like a consequence of the lack of access
control on<br>
setting ECT(1), and maybe the queue protection function would
address<br>
it, so that's interesting to hear about.<br>
<br>
But I thought the whole point of dualq over fq was that fq
state couldn't<br>
scale properly in aggregating devices with enough expected
flows sharing<br>
a queue? If this protection feature turns out to be
necessary, would that<br>
advantage be gone? (Also: why would one want to turn this
protection off<br>
if it's available?)<br>
<br>
<br>
</blockquote>
</div>
<br>
<fieldset class="gmail-m_3469002528728889916mimeAttachmentHeader"></fieldset>
<pre class="gmail-m_3469002528728889916moz-quote-pre">_______________________________________________
Ecn-sane mailing list
<a class="gmail-m_3469002528728889916moz-txt-link-abbreviated" href="mailto:Ecn-sane@lists.bufferbloat.net" target="_blank">Ecn-sane@lists.bufferbloat.net</a>
<a class="gmail-m_3469002528728889916moz-txt-link-freetext" href="https://lists.bufferbloat.net/listinfo/ecn-sane" target="_blank">https://lists.bufferbloat.net/listinfo/ecn-sane</a>
</pre>
</blockquote>
<br>
<pre class="gmail-m_3469002528728889916moz-signature" cols="72">--
________________________________________________________________
Bob Briscoe <a class="gmail-m_3469002528728889916moz-txt-link-freetext" href="http://bobbriscoe.net/" target="_blank">http://bobbriscoe.net/</a></pre>
</div>
</blockquote></div></div>