Speeds and feeds

tqbf@pobox.com tqbf@pobox.com
Sat, 6 Jun 1998 17:40:31 -0500 (CDT)


> Keeping the second router as a spare is pretty expensive.  Especially if you
> want more than a DS-1 worth of bandwidth - which was the original point of

It's exactly as expensive as using two seperate routers to manage two
seperate DS1 connections. One solution (mine) is simply easier to manage
than the other.

> the post, I think.  Anyway, an internal routing protocol (other than
> something like iBGP) isn't going to work in this instance.

Exactly why do you think that an IGP won't solve the problem of balancing
traffic over two links to the same destination? Two equal-cost links to
the same place should be balanced automatically by routing software. In
case you're confused, static routes qualify as an IGP.

> You'd need to have an agreement with your upstream to balance inbound
> traffic over the two lines; running an IGP between your end and the ISP
> should solve this.

> You'd need to have an agreement with the whole internet!  A provider isn't

No, you don't. If you only have one provider, you have no need to
advertise your addresses to the Internet. What you do need is to establish
between your site and your ISPs that there are in fact two equal-cost (at
least by policy) connections into your site. Running a dynamic IGP with
your ISP (perhaps they do OSPF with their DS1 customers for this purpose)
solves that problem. So do static routes.

> experience, it is near impossible to get any of the larger providers to run
> something like OSPF with you either.  This discussion isn't relevant unless

Don't do business with providers that aren't flexible enough to do exactly
what you want them to. It's a buyer's market, and the large providers
certainly don't differentiate themselves in any way meaningful enough to
offset inflexibility.

However, you are basing this argument off the assumption that when I say
IGP, I mean "dynamic routing protocol". 

> If we are considering the case of an IGP with a single provider, there would
> be a limited win from multiple DS-1 connections IMHO.  In that case, I would

Limited win? You mean BESIDES increasing available bandwidth by a factor
of two? Multi-homing is a non-trivial task, and doing it for real has
prerequisites that many (most) organizations cannot meet.

> suggest that they would want to look at the original suggestion of a larger
> pipe - i.e. limited bandwidth DS-3.

This is a silly suggestion. A capped DS3 has exactly the same reliability
problems (more so, in fact) than two DS1 connections. You are arguing that
there is limited benefit in running two DS1s to a single provider, because
you don't win the stability of being able to fail-over to the second ISP.
Your solution is thus to buy a SINGLE pipe to one ISP, so if your link
fails, you don't even get fail-over to a second link to the same provider.

If you don't need more than 2 DS1s worth of bandwidth, 2 DS1s is (in my
opinion) a much better solution than a single DS3.

-----------------------------------------------------------------------------
Thomas H. Ptacek	  The Company Formerly Known As Secure Networks, Inc.
-----------------------------------------------------------------------------
http://www.pobox.com/~tqbf	 "If you're so special, why aren't you dead?"