Re: [ippm] I-D Action: draft-ietf-ippm-capacity-metric-method-08.txt

["MORTON, ALFRED C (AL)" <[e-mail]>]

Hi Magnus,

Thanks for your review of 08. You have some questions to answer in the thread below.

However, I read your review as saying you only need two points addressed now:

1. Warn that people must not use the Section 8.1 Load rate adjustment algorithm as a general Congestion Control Algorithm (CCA).  We will add that in Section 8.1. We didn't call the algorithm a CCA anywhere, and we already make our goal clear (again and again).

2. The w*FT Timeout needs a better description. Your discussion of this wandered into one-way delay jitter and buffer depth. We might be able to base the timeout on the setting of the high delay range threshold (default 90ms) with the same recurring timeout property for repeated lost messages as described now. Something like 

           HDRT + FT + (w*FT) = Lost Status Backoff timeout

where: 
HDRT = high delay range threshold (default 90ms)
FT   = feedback time interval (default 50ms)
w    = index for repeated timeouts (w=0 initially, w++ on each timeout)

So here's our status: 
We've prepared two new versions of the draft with a large amount of new material predominantly in response to your comments, which quite frankly, have expanded tremendously over the last month (starting with a few short paragraphs in your DISCUSS).  We need to hear that addressing 1. and 2. now will satisfy your needs completely, then take next steps.

Thanks,
Al  

> -----Original Message-----
> From: ippm [mailto:[e-mail]] On Behalf Of
> [e-mail]
> Sent: Tuesday, March 30, 2021 8:38 AM
> To: [e-mail]
> Cc: [e-mail]
> Subject: Re: [ippm] I-D Action: draft-ietf-ippm-capacity-metric-method-
> 08.txt
> 
> Hi Ignacio,
> 
> Thanks, I'm aware of these protocols. The capacity metric method design
> aim is measurement, not transport. That's not a license to ignore other
> purposes, of course. I'd consider some different constraints to apply,
> like a slower ramp-up and brief operation as close to the (congestion
> free) access rate as possible, if the feedback indicates stable
> performance. And then turn off and stay silent (and stay silent earlier if
> networking conditions aren't stable - the design aim isn't to annoy
> persons on either end of the measurement). It's certainly not designed to
> exchange information on a level of minutes or hours in a fair and reliable
> manner. I wouldn't expect anybody to run a speed test for minutes or
> hours, no matter whether the implementation is based on transport or on a
> measurement protocol (I'd also expect the performance of my other to decay
> while running a TCP based speed test on my access).
> 
> I understand that long RTTs are an issue. A transport getting rare
> feedback should act conservative. A measurement protocol shouldn't create
> persistent or annoying congestion either, if RTT is high. To me, reaching
> the design aim of accurate access speed measurement as independent from
> RTT as possible is important, as the performance of "speed tests" based on
> TCP is depending on RTT, and causes them to be less accurate.
> 
> Regards,
> 
> Rüdiger
> 
> -----Ursprüngliche Nachricht-----
> Von: J Ignacio Alvarez-Hamelin <[e-mail]>
> Gesendet: Dienstag, 30. März 2021 13:42
> An: Geib, Rüdiger <[e-mail]>
> Cc: magnus.westerlund=[e-mail]; [e-mail]
> Betreff: Re: [ippm] I-D Action: draft-ietf-ippm-capacity-metric-method-
> 08.txt
> 
> Hi Rüdiger,
> 
> There is a family of different congestion avoidance algorithms, but the
> goal is to achieve the maximum throughput. Nevertheless, it is clear that
> some part of the network along the path from sender and server is
> congested; any of these algorithms will start to work. In other words, you
> have to put your server close enough to your link under measurement. The
> popular M-Lab uses at least three TCP flows to determine the rate as a way
> to trying to avoid this kind of problem, but in that case, servers are far
> away.
> Among the possible algorithms to maximize the uses vs. congestion in a
> short time, there is BBR used by QUIC (if I remember well):
> 
> 
> https://urldefense.com/v3/__http://dl.ifip.org/db/conf/networking/networki
> ng2018/2B1-
> 1570416152.pdf__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6
> cxSqflIg18$
> 
> https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/
> pii/S0140366419303470?casa_token=ggSPL6OXQH0AAAAA:zXNHOPgQeydZo3XBJRbeAApY
> rGQ7iZPocVZHCu11P19Hs_KdwRO9sMEse-
> HhKgiRGHsJeR4_MK0__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCN
> jQ6cxSlQZad4t$
> 
> I hope this answer the question and give a better understanding, cheers,
> 
> 
> 	Ignacio
> 
> 
> _______________________________________________________________
> 
> Dr. Ing. José Ignacio Alvarez-Hamelin
> CONICET and Facultad de Ingeniería, Universidad de Buenos Aires Av. Paseo
> Colón 850 - C1063ACV - Buenos Aires - Argentina
> [phone] / 5285 0705
> e-mail: [e-mail]
> web: https://urldefense.com/v3/__http://cnet.fi.uba.ar/ignacio.alvarez-
> hamelin/__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSuun
> ueiD$
> _______________________________________________________________
> 
> 
> 
> > On 30 Mar 2021, at 07:48, <[e-mail]>
> <[e-mail]> wrote:
> >
> > Hi Magnus,
> >
> > [MW] To accurately measure the capacity of the path the load algorithm
> strive to load the path so that the one way latency increases to indicate
> that the bottleneck buffer is sufficiently filled to achieve high utility.
> This load is not intended to be fair against other applications, it is
> intended to provide as accurate measurement as possible, potentially
> pushing other applications out of the way.
> >
> > [RG] The sender rate adaption mechanism operates by congestion feedback,
> as do many standard transport protocols. The purpose of the algorithm is
> not to determine, whether a queue or a policer burst size "is sufficiently
> filled to achieve high utility", the purpose is to accurately determine
> the access speed which can be reached without queue-build up and/or packet
> loss. After feedback indicating a sender rate which has caused congestion
> has been received, the sender rate is adapted until the maximum access
> speed not causing congestion is determined. To decide on a sender rate
> fulfilling this this criterium, a sender rate causing a congestion which
> can be accurately identified as congestion by a measurement system is
> required. If you are aware of a method reaching this goal without causing
> limited congestion, please share.
> >
> > [RG] I don't understand whether your feedback is focused on improvements
> of parametrization or on the basic behaviour of the algorithm. The
> algorithm is not designed to be unfair.
> >
> > Regards,
> >
> > Ruediger
> >
> > -----Ursprüngliche Nachricht-----which
> > Von: ippm <[e-mail]> Im Auftrag von Magnus Westerlund
> > Gesendet: Dienstag, 30. März 2021 10:54
> > An: [e-mail]
> > Betreff: Re: [ippm] I-D Action:
> > draft-ietf-ippm-capacity-metric-method-08.txt
> >
> > Hi,
> >
> > I have reviewed the latest versions changes and have some comments.
> >
> > Section 8.1
> >
> >   If no status feedback messages arrive at the sender for the interval
> 
> >   greater than the Lost Status Backoff timeout = w*FT, beginning when
> 
> >   the last message (of any type) was successfully received at the
> sender:
> >
> > "w" is used here and paragraphs below. But it is not really defined.
> Based on what is written below it is an algorithm constant, so maybe it
> should be given a name.
> >
> > So the backoff happens after w*FT. And I agree that with a regular
> feedback interval that will be working independent of RTT, starting the
> time when the first feedback has been received. What I thin this paragraph
> could be clearer on is that w*FT need to be considered in relation to the
> one-way delay jitter. And the most likely cause here is buffer depth in
> the bottleneck. It must also be related to upper delay threshold. What
> occurs if upper delay threshold is higher than w*FT? Is there a potential
> that a step increases is sufficient to push the buffer occupancy more than
> w*FT longer, i.e. causing the feedback to be delayed sufficient so that it
> triggers the lost feedback rather than the upper threshold? I haven't
> calculated what traffic increases that are possible and if that excess
> traffic can cause this to happen without additional cross traffic over
> bottleneck. I would assume this is fine to happen in the initial fast
> ramp-up, but it should not really happen in the regular ramp-up case.
> >
> >   The RECOMMENDED initial value for w is 3, taking Round Trip Time
> >   (RTT) less than FT into account.  A test with RTT longer than FT is
> > a
> >
> >   valid reason to increase the initial value of w appropriately.
> >   Variable w SHALL be incremented by 1 whenever the Lost Status
> > Backoff
> >
> >   timeout is exceeded.  So with FT = 50ms, a status feedback message
> >   loss would be declared at 150ms following a successful message,
> > again
> >
> >   at 50ms after that (200ms total), and so on.
> >
> >
> > I still think that Section 8 could be more explicit about the goal of
> the load algorithm.
> >
> > To accurately measure the capacity of the path the load algorithm strive
> to load the path so that the one way latency increases to indicate that
> the bottleneck buffer is sufficiently filled to achieve high utility. This
> load is not intended to be fair against other applications, it is intended
> to provide as accurate measurement as possible, potentially pushing other
> applications out of the way.
> >
> > The goal of being explicit about this is so that there is no doubt to
> anyone what this algorithm does, and that it is not a generic congestion
> control algorithm.  So Section 14.1 discuss this, but I think the document
> can be more honest here and state it explicitly in Section 8.
> >
> > Now, I am no longer an AD, but I don't have a problem with a limited
> scope algorithm that just is open with its intention of being unfair. I
> think having an algorithm that people may copy without understanding what
> it does is more a problem.
> >
> > Cheers
> >
> > Magnus Westerlund
> >
> >
> >> -----Original Message-----
> >> From: I-D-Announce <[e-mail]> On Behalf Of
> >> [e-mail]
> >> Sent: den 29 mars 2021 23:31
> >> To: [e-mail]
> >> Cc: [e-mail]
> >> Subject: I-D Action: draft-ietf-ippm-capacity-metric-method-08.txt
> >>
> >>
> >> A New Internet-Draft is available from the on-line Internet-Drafts
> > directories.
> >> This draft is a work item of the IP Performance Measurement WG of the
> >> IETF.
> >>
> >>        Title           : Metrics and Methods for One-way IP Capacity
> >>        Authors         : Al Morton
> >>                          Ruediger Geib
> >>                          Len Ciavattone
> >> 	Filename        : draft-ietf-ippm-capacity-metric-method-08.txt
> >> 	Pages           : 37
> >> 	Date            : 2021-03-29
> >>
> >> Abstract:
> >>   This memo revisits the problem of Network Capacity metrics first
> >>   examined in RFC 5136.  The memo specifies a more practical Maximum
> >>   IP-layer Capacity metric definition catering for measurement
> >>   purposes, and outlines the corresponding methods of measurement.
> >>
> >>
> >> The IETF datatracker status page for this draft is:
> >> https://urldefense.com/v3/__https://datatracker.ietf.org/doc/draft-
> ietf-ippm-capacity-metric-
> meth__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSk8Syqb0
> $
> >> o
> >> d/
> >>
> >> There are also htmlized versions available at:
> >> https://urldefense.com/v3/__https://tools.ietf.org/html/draft-ietf-
> ippm-capacity-metric-method-
> 08__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSndfx2ub$
> >>
> https://urldefense.com/v3/__https://datatracker.ietf.org/doc/html/draft-
> ietf-ippm-capacity-
> metric__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSjV8vJ
> r1$
> >> -
> >> method-08
> >>
> >> A diff from the previous version is available at:
> >> https://urldefense.com/v3/__https://www.ietf.org/rfcdiff?url2=draft-
> ietf-ippm-capacity-metric-
> met__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSrJUzesA$
> >> h
> >> od-
> >> 08
> >>
> >>
> >> Please note that it may take a couple of minutes from the time of
> > submission
> >> until the htmlized version and diff are available at tools.ietf.org.
> >>
> >> Internet-Drafts are also available by anonymous FTP at:
> >> https://urldefense.com/v3/__ftp://ftp.ietf.org/internet-
> drafts/__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSiqz4
> ahP$
> >>
> >>
> >> _______________________________________________
> >> I-D-Announce mailing list
> >> [e-mail]
> >> https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/i-d-
> announce__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSgqh
> X936$
> >> Internet-Draft directories:
> https://urldefense.com/v3/__http://www.ietf.org/shadow.html__;!!BhdT!yVLn0
> Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSk0bEN04$  or
> >> https://urldefense.com/v3/__ftp://ftp.ietf.org/ietf/1shadow-
> sites.txt__;!!BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSuY
> fdqPq$
> >
> > _______________________________________________
> > ippm mailing list
> > [e-mail]
> >
> https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/ippm__;!
> !BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSnKRhcmW$
> 
> _______________________________________________
> ippm mailing list
> [e-mail]
> https://urldefense.com/v3/__https://www.ietf.org/mailman/listinfo/ippm__;!
> !BhdT!yVLn0Ivu5MvPahTyoYRvk8VD3mI_hYOlYzLC0AGB1C50nCNjQ6cxSnKRhcmW$