Re: [RAM] The mapping problem: rendezvous points?

[Lars Eggert <[e-mail]>]

On 2007-5-9, at 15:58, ext David Conrad wrote:
>>> I'll ask again:  how does this ARP thing work again?
>> First off, all the ARP implementations I know of queue the packet  
>> during the loopkup, i.e., no loss.
>
> That's what I thought, but folks have been saying that routers  
> don't queue anymore.

I'm only familiar with end-host ARP code.

>> Second, the duration of an ARP lookup vs. the end-to-end RTT is  
>> typically miniscule. This is not so with LISP-like schemes.
>
> Yes, but memory for queues (for edge device I/O queues) is much  
> cheaper than it was...

I'm not worried about memory at all.

Whether a packet is queued while an ARP lookup is performed won't  
noticeably affect the end-to-end RTT that TCP measures, i.e., it  
won't affect TCP's RTT sampling. When LISP queues a packet while it  
does its magic, that is likely to take more time, because it's not a  
purely local operation, and this delay may influence TCP's RTT  
sampling, causing inefficiencies.

(When LISP forwards packets along a path with higher stretch during  
the setup phase, rather than queuing them, that may cause other  
inefficiencies, which I described in my first reply.)

Lars

_______________________________________________
RAM mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/ram