Martin,
The issue is not with advertising the blocks out, but what you have not
taken into account for is how you plan to route out to the internet. The
whole point of running BGP is optimal routing and redundancy. If an ASP is
paying thousands of dollars per month for a redundant link then I promise
you it will get used, ASP's have not exactly proven themselves to be rolling
in cash, unless it's from a VC. Most ASP's go into Colo's anyway. These
colo's are the ones maintaining the BGP so its clients have all the benefits
of having redundant links to the Internet via a single ethernet cable to
their rack/cabinet/cage.
My 2-cents.
Patrick Greene
-----Original Message-----
From: Martin Picard
To: cisco-nsp@puck.nether.net
Sent: 2/22/2001 7:42 PM
Subject: [nsp] Dual homing without BGP
Hi,
From an ASP redundancy standpoint, dual homing to
two different providers is essential. Most of the times
a small block of IP addresses is required (from one of
the Upstream ISP). In order to be fully redundant, this
block has to be propagated through the internet by
both Upstream ISPs. I understand that this causes
some "non-summarized" routes to be advertised but
are there any other choices ?!
Assuming not, both providers will advertise the prefix.
The issue of wether or not one agrees to have a more
specific prefix to advertise and also allow others to
advertise it puzzles me but moreover is the way to do
it.
It seems that when we're talking BGP peering, there's
no problem, everybody can advertise whatever.
Again, from an ASP standpoint that only has a local
LAN of servers with redundant routers to get to the
internet, the full knowledge of all Internet routes is
useless. Granted that the ASP can filter out all routes
if so desired and that BGP can be used only to
provide the Internet the paths to the ASP. But does
it absolutely require that the ASP runs eBGP with
both providers. I mean, both providers can inject
the ASP prefix in the Internet and the ASP just
default to one or the other. If BGP is the only option,
I just can't imagine the number of ASNs required
to support all upcoming ASPs and I don't see any
technical reasons why it cannot be otherwise.
I am probably missing something...
Please enlight me !!!
tx
martin
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:30 EDT