[arin-ppml] Rationale for /22

[Kevin Kargel]

> 
> Kevin Kargel wrote:
> > One quick point on the existance or emergence of small multi-homed
> > networks..
> >
> > In the past year and a half I have seen a very noticeable increase in
> the
> > number of small organizations that are going or attempting to go multi-
> homed
> > with DSL connections to multiple upstream providers.  They are doing
> this as
> > a means to try and do what they did before with expensive dedicated
> circuits
> > like T1.  Increased reliability of a $39.95 SDSL has led these customers
> > down the primrose path to think that if they have a couple of these then
> > they don't need an $800 ATM circuit.
> >
> > I am not saying I agree with this philosophy or even that I think it is
> > functional, but I am seeing it happen.
> >
> > As these customers get educated they figure out that they are
> technically
> > "multi-homed" and that they "can" get their own IP space,
> 
> No, they can't because they aren't really multihomed.

Please educate me, if they have more than one connection to the internet it
was my understanding they were multi-homed.  I do not recall any
specification on the type of connection as a requirement to be multi-homed. 

Technically there is nothing stopping an organization from being multi-homed
with a collection of 56K dialup accounts.

> 
> > and if they invest
> > in a router that can do BGP they can gain transparent (to the end user)
> > routing reliability using these DSL connections.
> 
> Very few ISP's provide BGP over DSL.  We are the only one in our city
> that does, as a matter of fact.

The Big Bells won't do it, and I suspect the larger ISP's won't do it, but
there are plenty of small ISP's around the country who will.  In fact their
admins might even think it would be a fun experiment.

> 
> > Bean counters tend to
> > prefer non-recurring expenditures to recurring expenditures.
> >
> > These organizations are small hospitals, clinics, small banks, farmers,
> > insurance agencies, non-chain manufacturers and distributers, et.al.
> They
> > don't need a lot of IP, but they do need routing failover.  They are
> > learning that it is theoretically (albeit not realistically) possible to
> do
> > what used to require a T1 on SDSL better for a fraction of the cost.
> >
> > I believe we will see a mushrooming incidence of these small multi-homed
> > organizations in the routing tables in the coming years.
> >
> > This will affect the issue of global routing table size and does have a
> > bearing on reducing the minimum allocation unit.
> >
> 
> I don't think so, Kevin.  The main reason for this is that all of the
> small "2 DSL multihomed" routers that are on the market operate on
> the same principle, they expect 2 DSL lines that are PPP-mode DSL.
> The router uses a single NAT translation table with 2 default routes
> and some fancy load-balancing programming that puts half of the
> translation sessions on one default and the other half on the other
> default.

It is easily done with a Cisco SoHo router (or with a SonicWall for that
matter). Any routing device (can you say Linux?) that will run BGP and allow
you to forward IP and customize the LAN network will do. Bridge two DSL
modems to act as media converters, configure your router WAN ports for
PPPoE, get your upstreams to assign static or persistant IP's to your PPPoE
authentication, and you are off and running.  

It is also eminently doable using DD-WRT.


> 
>  From the outside if the org is fielding servers internally, the
> org publishes both IP numbers for the server forward - with the
> caveat that if one of the DSL sessions goes down then connectivity
> to the internal server is intermittent.  Obviously, you would have
> severe problems fielding any kind of serious server behind such
> a thing - like a DNS server for example.

So long as the upstream providers accept and promulgate your BGP
advertisements I don't see the problem.

> 
> NAT is an essential and required component of these schemes because
> these routers are mainly load-balancing OUTBOUND requests for HTTP
> traffic and suchlike.

That is why the orgs that want to do this need to invest ($500 or more) in a
real router.  A residential grade router won't do it right, I agree.

> 
> These routers are cheap, generally under $200 or so.  The orgs that
> deploy these solutions are generally very satisfied with them because
> their primary purpose is failover, so that if one of their DSL lines
> goes down, they still can surf the web.  But their understanding of
> failover is pretty much limited to that - when you tell them that an
> upstream ISP of theirs could lose it's connectivity, yet their little
> router wouldn't detect this, you generally get blank stares.
> 
> Trying to convince an org like you have described to toss out their
> little sub $200 "load balancing" NAT/router and go to a real router that
> costs 20 times more that can speak BGP is an exercise in futility.
> They have a solution that gets them 90% of the way towards real
> multihoming, is cheap, and does that 90% really well.  And most
> importantly, they don't have to actually understand anything at all
> about networking, they just plug it in and go.

I am not saying everyone will want to do it, or even that a majority of
those that can will opt to do it, I am just saying that some fringe element
*is* doing it.  I will also predict that we will see more organizations
doing it in the future.

> 
> Ted
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 3224 bytes
Desc: not available
URL: <https://lists.arin.net/pipermail/arin-ppml/attachments/20090803/a3f5f452/attachment.bin>