Re: Going dual-stack, how do apps behave and what to do as an operator (Was: Apple Airport Extreme I

On 9/15/07, Jeroen Massar <jeroenunfix.org> wrote:
> [spam: Check http://www.sixxs.net/
misc/toys/ for an IPv6 Toy Gallery ]
>
> Somewhat long, hopefully useful content follows...
>
> Barrett Lyon wrote:
> [..]

[ clip ]

> Of course when there is only a A or AAAA only that
protocol will be
> used. All applications are supposed to use
getaddrinfo() which sorts
> these addresses per the above specification, the app
should then
> connect() to them in order, fail/timeout and try the
next one till it

Since when is a timeout on the Internet ok? Haven't we moved
beyond
that? This is a controllable timeout. We don't have to do
it, which is
the point. What's the right way to do this?

Thank you, and thank you Barret for starting the thread.


-M<

On 9/21/07, Mark Andrews <Mark_Andrewsisc.org> wrote:
>
> In article
<2d106eb50709202254q6f4ea4b7v6beda6deee5f7143mail.gmail.com> you write:
> >
> >On 9/15/07, Jeroen Massar <jeroenunfix.org> wrote:
> >> [spam: Check http://www.sixxs.net/
misc/toys/ for an IPv6 Toy Gallery ]
> >>
> >> Somewhat long, hopefully useful content
follows...
> >>
> >> Barrett Lyon wrote:
> >> [..]
> >
> >[ clip ]
> >
> >> Of course when there is only a A or AAAA only
that protocol will be
> >> used. All applications are supposed to use
getaddrinfo() which sorts
> >> these addresses per the above specification,
the app should then
> >> connect() to them in order, fail/timeout and
try the next one till it
> >
> >Since when is a timeout on the Internet ok? 
Haven't we moved beyond
> >that?
>
>         You mean to say you get 100% connectivity with
IPv4?

I mean to say that I don't willingly set out to deliver <
100%.

On 21-sep-2007, at 7:54, Martin Hannigan wrote:

>> All applications are supposed to use getaddrinfo()
which sorts
>> these addresses per the above specification, the
app should then
>> connect() to them in order, fail/timeout and try
the next one

> Since when is a timeout on the Internet ok? Haven't we
moved beyond
> that? This is a controllable timeout. We don't have to
do it, which is
> the point. What's the right way to do this?

I agree that it's not acceptable to engineer things such
that  
timeouts occur by design. However, things tend to break, and
in those  
situations it's important to recover as well as can be
expected. So  
the correct way to operate here is for the network designer
to make  
reasonably sure ("unreliable datagram" etc) that
everything works,  
for the stack designer to make sure that there is a good
algorithm  
for selecting the "best" combination of
destination and source  
addresses and for the application to cycle through all
addresses if  
the two former efforts weren't completely successful.