Re: shim6 @ NANOG (forwarded note from John Payne) (fwd)

[Kurt Erik Lindqvist <kurtis@kurtis.pp.se>]

On 3 mar 2006, at 19.01, John Payne wrote:

> On Mar 3, 2006, at 3:26 AM, Kurt Erik Lindqvist wrote:
>
> > This was something that was brought up in the famous L.A Nanog  
> > session, but I actually fail to see in what way you think shim6  
> > add delay to the initial hit? You will look up the destination in  
> > DNS, get multiple AAAA's back, perform RFC3483 and start  
> > communicating. That is the same, with or without shim6. If it is a  
> > short lived session and failure occurs, that is no worse than IPv4  
> > which will also have to timeout and retry.
>
> The issue is first hit under failed conditions.
>
> AIUI with the current plan, you'd get multiple AAAAs back vs one  
> multihomed A leading to a potentially long cycle of TCP timeouts  
> before getting a AAAA that actually worked.

Ok, so this is actually not a DNS timeout issue, and it isn't a TCP  
timeout issues really either. When shim6 "suspects" a path failure it  
will try locator pairs until one working is found. The time it takes  
to to pair wise trials is relative to the number of locators but  
should be fairly quick.

> Unless you want to try parallalising SYNs and racing them off...  
> which has other issues :)

There are no SYNs involved..

- kurtis -