Re: failure detection

[Paul Jakma <paul@clubi.ie>]

On Sun, 14 Aug 2005, Iljitsch van Beijnum wrote:

> > - OS maintainers will be prepared fundamentally change their network
> >  stacks to support shim6
>
> Well, obviously they have to implement it or it won't work, no pain no gain. 
> :-)

I'd rather the bar for implementing shim6 were not that high.

> > Don't try get clever, cause the upper layer knows /more/ than 
> > shim6 does:
>
> Just now you questioned whether the upper layer knows enough to be unhappy!

I said it might not have a notion of unhappy/happy, different thing.

Anyway, such feedback typically goes /up/ the stack, not down. 
Ultimately such feedback is presented to the *user* (eg in case of a 
user trying to download something and their throughput progress meter 
is low, the user may want to try a different mirror site).

> > - It might have a list of addresses it wants to try, don't delay it
>
> It doesn't. The application may have a list of addresses, but TCP 
> or UDP only see a single one.

Uhmm.. SCTP? How do you even know what the transport has or doesn't 
have?

Further, when you're at/near the bottom of the pile, then looking up 
the only thing you will see of the application and user is the 
transport.

Stop even thinking about TCP. Forget TCP. You're specifying a 
data-layer essentially, it's job should be best-effort delivery with 
a minimal, bounded time in which to determine failure (if it's even 
possible to do so).

Leave the transport layer stuff to the "experts", ie TCP or whatever 
the hell is up the stack. There might even be some kind of 
QoS-measuring software or protocol running to make decisions about 
latency/throughput between different links and change routing between 
them.

Don't presuppose, don't add in stuff to Shim6 that doesn't need to be 
there.

> So basically you're saying that applications can multihome better 
> than the network stack? I disagree.

I didn't say that at all.

I'm raising my concerns about the scope of the problem shim6 is 
trying to solve, particularly the talk of creeping up into the 
transport. This working group is not:

- nominally an expert group on OS networking design
- chartered to investigate data-layer/transport interaction

Shim6 is here to provide a shim layer for IPv6 multihoming soltuion 
that can operate (from the charter) "... without upsetting transport 
protocols or applications."

I'd suggest the WG follow this advice. ;)

regards,
--
Paul Jakma	paul@clubi.ie	paul@jakma.org	Key ID: 64A2FF6A
Fortune:
... The prejudices people feel about each other disappear when then get
to know each other.
		-- Kirk, "Elaan of Troyius", stardate 4372.5