[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ippm] questions of TCP metrics
Hi Loki,
You mentioned MOS scores of VoIP and that is a very good analogy to what providers face today.
In a VoIP environment, IP layer packet metrics can be correlated to VoIP quality, but network providers obviously find MOS scores a much more meaningful metric of the VoIP application running over the network.
Similarly, network providers can run IP layer tests (as they do today with RFC-2544) and infer TCP / application layer performance. But network providers seek a more meaningful and realistic measure of potential application performance and would benefit from TCP layer throughput measurements.
Thanks,
Barry
Principal Member of Technical Staff
JDSU Communication Test (formerly Acterna)
Emerging Markets and Technology Research
One Milestone Center Court
Germantown, MD 20876
(W) 240-404-2227
(C) 301-325-7069
-----Original Message-----
From: ippm-bounces at ietf.org [mailto:ippm-bounces at ietf.org] On Behalf Of Loki Jorgenson
Sent: Tuesday, October 13, 2009 1:38 PM
To: ippm at ietf.org
Subject: [ippm] questions of TCP metrics
Please pardon my relative naïveté and probable ignorance of past discussions on this topic - but there are two points that I am not clear on:
1) "TCP performance has always been in IPPM's charter" - does this statement derive entirely from historical precedence?
In light of the existence of PMOL (regardless of its current state of quiescence), it would seem obvious that TCP is part of the "other layers" scheme. Specifically, it would seem that TCP performs as consequence of an application's implementation and therefore that TCP performance would be non-specific to (although clearly a function of) IP-level characteristics, just as MOS would related but non-specific in the case of VoIP.
The IPPM charter comes as close as referencing "bulk transport capacity" but it is not trivial (for me) to conclude that a specific mechanism for enabling bulk transport is necessarily within scope. Prior to PMOL, IPPM would seem the likeliest group to have undertaken a TCP metric. And, certainly, if another group such as PMOL were undertaking a TCP metric, it would be obvious that IPPM would be involved.
This may well be simply a matter of philosophical leaning - but so that I am clear - on what basis do you say, Al, that TCP performance is unquestionably an IPPM matter?
2) There are at least two regimes of TCP performance that seem important to consider:
i) non-equilibrium behaviors such as during slow-start
ii) steady-state behaviors
Apart from the vagaries of different TCP implementations, does not the work of Padhye et al and Mathis et al (such as listed below) offer the basis for at least developing useful steady-state descriptions?
Matt Mathis himself has said he does not see it being viable developing metrics along these lines. While I can imagine they are not 100% accurate under all circumstances, they would at least seem like useful extrapolations of IP-level characteristics. What then are the most important shortcomings of these models?
Refs:
Modeling TCP Reno Performance: A Simple Model and its Empirical Validation
J. Padhye, V. Firoiu, D. Towsley and J. Kurose,
IEEE/ACM Transactions on Networking, April 2000.
[1] M. Mathis, J. Semske, J. Mahdavi and T. Ott, The Macroscopic Behavior of the TCP Congestion Avoidance Algorithm, Computer Communication Review, Vol 27, Number 3, July 1997
Al Morton writes:
>Message: 6
>Date: Tue, 13 Oct 2009 11:56:15 -0400
>From: Al Morton <acmorton at att.com>
>Subject: Re: [ippm] Testing TCP Throughput Capacity in Operator Networks
>To: Henk Uijterwaal <henk at ripe.net>
>Cc: ippm at ietf.org
>Message-ID: <200910131556.n9DFuPQu029929 at alpd052.aldc.att.com>
>Content-Type: text/plain; charset="us-ascii"; format=flowed
>
>At 09:26 AM 10/13/2009, Henk Uijterwaal wrote:
>>Al,
>>
>>>At 06:17 AM 10/13/2009, Henk Uijterwaal wrote:
>>>>...PMOL deals with higher layers, besides that (AFAIK) the PMOL
>>>>WG is in some form of hybernation.
>>>Expect to see more action in PMOL when the revised draft of the
>>>framework and guidelines is published (new editor).
>>
>>I wasn't aware of this activity. That said, do you agree that this
>>work belongs in IPPM rather than PMOL?
>>
>>Henk
>
>I think that TCP performance has always been in IPPM's charter.
>
>But like a few others, I have doubts that TCP-based throughput
>measurements will produce useful results, starting with the
>issue of what TCP version and other configuration options to use.
>However, Matt Mathis' suggestion deserves attention.
>
>I should also add that Ethernet-layer throughput testing has
>been proposed in ITU-T SG12 Q17, and contributions are expected on
>this topic at the November meeting (this is scheduled at the
>same time as IETF, so I won't be in Hiroshima in person).
>
>Al
---
Loki Jorgenson
Apparent Networks
t 604 433 2333 ext 105
m 604 250-4642
_______________________________________________
ippm mailing list
ippm at ietf.org
https://www.ietf.org/mailman/listinfo/ippm