[ppml] Policy to help the little guys

[Ted Mittelstaedt]

> -----Original Message-----
> From: ppml-bounces at arin.net [mailto:ppml-bounces at arin.net] On 
> Behalf Of Jo Rhett
> Sent: Wednesday, March 19, 2008 4:56 PM
> To: David Williamson
> Cc: Randy Bush; arin ppml
> Subject: Re: [ppml] Policy to help the little guys
> 
> 
> On Mar 18, 2008, at 11:59 PM, David Williamson wrote:
> > If my site has no useful interaction with somewhereistan,
> > I can simply block the short prefixes from them.  Or, at a higher 
> > level, I could just block all of the /8s associated with a specific 
> > RIR, based on my business model/needs.  On the flip side, I could 
> > accept a bunch of /28s or /29s from sites important to me.
> >
> > All it would take is some hard to maintain filters.
> 
> One assumes you are no longer using those purple boxes in  
> production?  Because the memory usage of a blocked route and the  
> memory usage of an accepted route are identical in that case.   
> Filtering the prefixes won't help you there.
> 
> It's not as bad, but nearly so with Cisco last time I checked.
> 
> So if we accept a /16 of /29 routes, you're looking at adding 8192k  
> new entries to the table.  Most of our peers are currently  
> advertising 244k routes to us.  That brings it to 252k routes, which  
> is very *very* near the max table size of most production equipment  
> out there.
> 

OK I think I'm missing something in this discussion - could someone
fill me in.

My understanding is that a "little guy" multihomed ISP accepts routes
fundamentally to allow himself to balance his OUTBOUND traffic.

  (his inbound traffic is a function of how other people are choosing to
send
traffic to him, based possibly on his advertisements and prepends he might
be sending out, which may or may not be paid attention to, thus have
nothing to do with the size of his route table)

Now, I would further assume that networks out there advertising /29's
don't have a lot of hosts in them - thus they are likely a teensy, tiny
speck of the "little guy's" -outbound- traffic.  Thus, the little guy
ISP doesen't need to bother with accepting their advertisements (or more
properly, having his upstreams modify their filters to send him longer
prefixes than /24) he can just pick one of his feeds and default route
whatever outbound traffic he is sending to these /29's.  Sure, you lose
redundancy for those /29s - but there's few people on 'em so what?

I mean, the scenario of thousands of hosts that can generate millions of
http or whatever requests into little guy's servers all behind a /29 seems
to
me to be extremely farfetched.  Does such a network even exist at all?

The scenario also seems unlikely that most "little guy" ISPs are even going
to
have a large amount of outbound traffic anyway.  "little guy" companies that
tend to generate a large amount of outbound traffic seem to me to be mainly
in
the hosting business - in which case their servers are very likely in a
colocate facility somewhere and their outbound connectivity consists of
a gigabit ethernet connection to their colocate host who has to bother with
concerning themselves with all this traffic balancing stuff.

I guess what I'm asking is in an Internet with the majority of users
mostly behind a handful of superbig AS's, all little guy really needs to
pay any attention to for effective load balancing is the large prefix
advertisements, since the small under /24 prefixes are not the ones with
the large numbers of users where the bulk of his traffic is flowing to.

So, is this policy discussion actually solving a real problem, or are
we all doing someone's theoretical CCIE homework question for them?

Ted