RE: dsmterm-05 - Forwarding Congestion

Do you all agree w/ Jerry's assessment?  The short answer is
in the last paragraph 

-Kevin

-------- Original Message --------
Subject: RE: [bmwg] dsmterm-05 - Forwarding Congestion
Date: Mon, 3 Mar 2003 11:10:08 -0800
From: "Perser, Jerry" <jerry.perser <at> spirentcom.com>
To: "'Kevin Dubray'" <kdubray <at> juniper.net>
CC: "David Newman (E-mail)" <dnewman <at> networktest.com>,   "Scott Poretsky 
(E-mail)" <sporetsky <at> avici.com>,   Shobha Erramilli	 
<shobha <at> qnetworx.com>,   Sumit Khurana <sumit <at> research.telcordia.com>

 > But why the reticence to call the term "Lossful Forwarding
 > Congestion."  And why not call out the lossful attribute
 > out explicitly in the definition?
 >
 > See: http://www.ietf.org/proceedings/02nov/152.htm
 >
 >
 > -Kevin
 >

The focus on Congestion is centered around Traffic Control Mechanisms.  By
externally observing the DUT/SUT ability to forward packets, it can be
determined whether Traffic Control Mechanisms are active.

Active and Enabled are two different things.  At low OLOAD, the Traffic
Control Mechanisms may be enabled, but not effecting the forwarding of

WG Last Call: draft-ietf-bmwg-dsmterm-05

Metric Mavens:

A WG Last Call period for the Internet-Draft regarding
"Terminology for Benchmarking Network-layer Traffic Control
Mechanisms":
              <draft-ietf-bmwg-dsmterm-05.txt>

will be open from 4 March 2003 until 25 March 2003.

Please weigh in on whether or not you feel the Internet-Draft
should be given to the Area Directors for consideration in
progressing the draft to an Informational RFC.  Send your comments
to this list or kdubray <at> juniper.net.

A URL for the Internet-Drafts is:
  http://www.ietf.org/internet-drafts/draft-ietf-bmwg-dsmterm-05.txt

-Kevin

RE: dsmterm-05 - Forwarding Congestion

Mon, 11 Nov 2002 , Fred Baker stated that:
"A congested interface is one for which the queuing delay is measurably 
non-zero, and as a result causes issues for applications running over the
connected internet fabric. It is hard to meaningfully discuss without
bringing the effects on the applications into the discussion."

W/out the 'Lossful' modifier as suggested by Kevin, the term definition for
'Forwarding Congestion":
          "Packet Loss, not increased Delay, is the metric to indicate
          the condition of Forwarding Congestion.  Packet Loss is a
          deterministic indicator of Forwarding Congestion."

is, at best, inaccurate.

- Debby Stopp

> -----Original Message-----
> From: Kevin Dubray [mailto:kdubray <at> juniper.net]
> Sent: Monday, March 03, 2003 12:06 PM
> To: bmwg <at> ietf.org
> Subject: RE: [bmwg] dsmterm-05 - Forwarding Congestion
> 
> 
> Do you all agree w/ Jerry's assessment?  The short answer is
> in the last paragraph 
> 
> -Kevin
> 
> -------- Original Message --------
> Subject: RE: [bmwg] dsmterm-05 - Forwarding Congestion

RE: dsmterm-05 - Forwarding Congestion

> W/out the 'Lossful' modifier as suggested by Kevin, the term
> definition for
> 'Forwarding Congestion":
>           "Packet Loss, not increased Delay, is the metric to indicate
>           the condition of Forwarding Congestion.  Packet Loss is a
>           deterministic indicator of Forwarding Congestion."
>
> is, at best, inaccurate.

This is a curious assertion. Let's invert it and see if it still stands up.

Under what conditions would packet loss NOT be a deterministic indicator of
Forwarding Congestion?

dn

ps. The backstory here, for those new to this draft, is that we have opted
for loss as a congestion indicator rather than delay or other metrics
because it is deterministic.

As noted in the ID, increased latency or jitter also may indicate congestion
(in fact, we point to other references that discuss this). The problem with
applying these other metrics here is that they are not deterministic: We
can't say, for example, that a latency increase of X percent will always
indicates the presence of congestion in all cases. There's just no
one-size-fits-all answer, the way there is with loss.

RE: dsmterm-05 - Forwarding Congestion

Debby,

I was reading draft-ietf-bmwg-mcastm-10.txt and found several metrics (I
counted 6) that have the same 'Lossful' attribute as Forwarding Congestion.
For example, Aggregated Multicast Throughput requires loss to equal zero.

Do you believe that Mixed Class Throughput, Aggregated Multicast Throughput,
Encapsulation Throughput, etc. are also "at best, inaccurate."?

If not, why?

Jerry.

> -----Original Message-----
> From: Debby Stopp [mailto:dstopp <at> ixiacom.com]
> Sent: Tuesday, March 04, 2003 8:53 AM
> To: bmwg <at> ietf.org
> Subject: RE: [bmwg] dsmterm-05 - Forwarding Congestion
> 
> 
> Mon, 11 Nov 2002 , Fred Baker stated that:
> "A congested interface is one for which the queuing delay is 
> measurably 
> non-zero, and as a result causes issues for applications 
> running over the
> connected internet fabric. It is hard to meaningfully discuss without
> bringing the effects on the applications into the discussion."
> 
> W/out the 'Lossful' modifier as suggested by Kevin, the term 
> definition for

RE: dsmterm-05 - Forwarding Congestion

In section 3.1.3 Forwarding Congestion:

"At offered rates above throughput, the DUT/SUT is considered congested."

Can you explain this?

In my understanding of the classic definition of throughput and congestion,
the DUT/SUT could be congested long before any packets are dropped. 

thx,
/m

RE: dsmterm-05 - Forwarding Congestion

First off, which "classic definition" of throughput are you referring to?
The one in use here, from RFC 1242, defines tput as the highest rate without
loss.

Second, if we accept the 1242 definition what is it about the quoted
statement that requires explanation?

By definition, any rate above the tput level involves loss.

dn

> -----Original Message-----
> From: bmwg-admin <at> ietf.org [mailto:bmwg-admin <at> ietf.org]On Behalf Of
> Michele Bustos
> Sent: Tuesday, March 04, 2003 1:40 PM
> To: Jerry Perser (E-mail)
> Cc: bmwg <at> ietf.org
> Subject: RE: [bmwg] dsmterm-05 - Forwarding Congestion
>
>
> In section 3.1.3 Forwarding Congestion:
>
> "At offered rates above throughput, the DUT/SUT is considered congested."
>
> Can you explain this?
>
> In my understanding of the classic definition of throughput and
> congestion,
> the DUT/SUT could be congested long before any packets are dropped.

RE: dsmterm-05 - Forwarding Congestion

RE: dsmterm-05 - Forwarding Congestion

New BMWG co-chair.

BMWG folk,

I'm very pleased to announce that Al Morton (acmorton <at> att.com) has
agreed to contribute as co-chair of the BMWG.  Welcome, Al!

Kevin