Re: [RRG] How to Incrementally Deploy APT

["William Herrin" <bill@herrin.us>]

On Fri, Feb 22, 2008 at 6:47 PM, Michael Meisel <meisel@cs.ucla.edu> wrote:
>  There is one important thing you're missing here -- the router doesn't
>  need to *wait* for a response from the default mapper, the default
>  mapper will forward the packet. It does still get a reply, though, and
>  adds the entry to its cache once it receives it. So, in the mean time,
>  the router can continue to deal with other traffic.

So the router doesn't ask the default mapper for a map; instead it
sends the packet whose destination was a cache-miss to the default
mapper. Then the default mapper encapsulates and forwards the packet.
The default mapper at that point can also choose to send the router a
new entry for its cache so that the next packet for that destination
can flow directly through the router.

Do I understand now?

>  I'm not sure if this
>  makes any difference to the issue you're describing, though.

Yes, a big difference. This means that it should be possible to
implement the router's part entirely in hardware. It's the difference
between working and collapsing under the load at the engineering
level.


>  >>  > A-(B-C-D)-E-F
>  > Eh? But C is an entire APT network that was formerly a BGP AS. It
>  > contains pron servers that I want to surf! How do I surf them from A?
>
>  Yes, but C is a transit network, by which we mean it's part of the
>  infrastructure, think Level3 or Sprint. There aren't any pr0n servers
>  for you in there (that we know of). If their customers have servers,
>  then those customers need to have their prefixes in the mapping table.

Okay, so C in my diagram is really two components: C1, the APT transit
network and C2 the endpoint network attached only to C1 which contains
my pr0n servers. I can surf the servers in C2 from A. Is that correct?

How do C2's prefixes get to A's BGP table so that A knows it should
send those destinations to B?


Stepping back, I think the single most important question you could
address at this point is: which of the currently deployed router
models can be software-upgraded to support APT in the fast (hardware)
path? This is make or break for your proposal. If you have to replace
every DFZ router to get APT in the fast path, deployment will cost
more than 4 Billion with a B dollars. If you only have to deploy
2*27000 default mappers, deployment could cost as little as 150
million dollars. I think you'll agree that's a -huge- difference.

Regards,
Bill Herrin



-- 
William D. Herrin                  herrin@dirtside.com  bill@herrin.us
3005 Crane Dr.                        Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg