In-Reply-To: <1081976344.43da888b9d93c@webmail.pi.se>
References: <1081976344.43da888b9d93c@webmail.pi.se>
Mime-Version: 1.0 (Apple Message framework v613)
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Message-Id: <FB711995-8EC2-11D8-9065-000A957FC5F2@csperkins.org>
Content-Transfer-Encoding: quoted-printable
Cc: toip@snowshore.com, avt@ietf.org
From: Colin Perkins <csp@csperkins.org>
Subject: Re: [avt] Use of redundancy in rfc2793bis text transmission -
 optimizing interoperability
Date: Thu, 15 Apr 2004 10:55:10 +0100
To: =?ISO-8859-1?Q?Gunnar_Hellstr=F6m?= <gunnar.hellstrom@omnitor.se>
Content-Transfer-Encoding: quoted-printable
Sender: avt-admin@ietf.org
Errors-To: avt-admin@ietf.org
Precedence: bulk
Content-Transfer-Encoding: quoted-printable

On 14 Apr 2004, at 21:59, Gunnar Hellstr=F6m wrote:
> I often get comments on the real time interactive text conversation=20
> transport
> RFC2793bis, that it must more clearly require the use of redundancy to=20=

> assure
> good success rate in text transmission even in bad network conditions.
>
> Next version of draft-ietf-avt-rfc2793bis is about to be published,=20
> and I would like to have  agreeable wording on this issue.
>
> A traditional requirement for basic text conversation quality is that=20=

> no more
> than 1% characters may be lost in conditions where voice=20
> communications is
> barely usable. Where characters are dropped marks for missing text=20
> should be
> inserted in the received text. A higher quality level, called good=20
> text quality requires no more than 0.2% characters to be dropped. This=20=

> is of course no exact scientific measure and many factors influence=20
> both the loss and the perception of usability of voice conversation,=20=

> but it gives a fair design goal.
>
> With voice coding and transmission schemes prevailing today, voice=20
> gets barely
> usable around 20% packet loss.
>
> Therefore, in order to assure proper interoperability with the=20
> required quality level for text at 20% packet loss, we need to require=20=

> one original and two redundant transmissions according to RFC2198.=20
> (lowering text loss to 0.8%)

Why do you think it acceptable to be sending at full rate when there is=20=

20% packet loss, rather than reducing the sending rate to reduce packet=20=

loss? Do you have some analysis to show that, with typical text=20
conversation packet rates and network RTTs, transmission with 20%=20
packet loss is TCP friendly (or otherwise)?

The requirements you list may be appropriate from a media quality=20
viewpoint, however you also need to demonstrate that the payload format=20=

is safe to deploy. This is particularly important for a format which=20
adds a large amount of FEC to allow it to operate in the presence of=20
large amounts of congestion, but doesn't specify a congestion control=20
response.

--=20
Colin Perkins
http://csperkins.org/=


_______________________________________________
Audio/Video Transport Working Group
avt@ietf.org
https://www1.ietf.org/mailman/listinfo/avt