RE: IXP hardware

["Barry Raveendran Greene" <bgreene at cisco.com>]

The value route-servers and route-reflectors add on an IXP in a emerging
market is cost scaling.  The more peers you got, the more BGP memory context
and CPU power you'll need. An IXP in Mozambique is not an IXP in the US or
Europe where you have 7XXXs, 12XXXs, M40s, or M20s. Those have plenty of
memory and horsepower to handle the load of +50 peers. A 3600, 261x, 265x,
and especially a 2500 will not have the power to scale to +50 on a direct
peer to peer model. Of course you'll not see 50 ISP on an IXP in Mozambique.
You might see 10 - which is right at the limit of a 2500 for direct peer to
peer. Get above 10 and the ISP will need to swap out their router for
something more powerful.

While we're in agreement that the big IXPs in US and Europe do not use
Router Servers any more (they never used Route Reflectors), I also put
myself in the shoes of a small ISP in Mozambique - where several thousand US
dollars for a 2600 is a _major_ capital expense.

Hence, it is wise to put all three options on the table. Let the local team
walk through the pros and cons. A RR or a RS server on an IXP does not mean
people "have to connect." They can still do peer to peer and have the RR/RS
on the IXP. This is the model HK went down. +80 peers from the RR and other
(i.e. UUNET & HKTel for example) who went peer to peer with each other. HKIX
still has small ISPs with their originally 2501s w/ T1s talking through the
RR to the likes of UUNET, HKTel, and other big boys in Asia. I'll go on the
record and state that if it wasn't for HKIX's RR, these small ISPs would
have been crushed by the big boys. The HKIX RR gave them a cost efficient
way to carve out a niche in the market and survive.

Barry

PS - Randy and I always do this. The technical dialog between two engineers
who have two ways of doing something is beneficial for the people lurking.
At the same time, we're always working together behind the scenes via NSRC
and ISOC to help IXPs get started.

PSS - Yes, I've been meaning to put one of these dialogs in the whitepaper.

> -----Original Message-----
> From: Randy Bush [mailto:randy at psg.com]
> Sent: Friday, March 08, 2002 4:05 AM
> To: Brian Candler
> Cc: Barry Raveendran Greene; antonio at nambu.uem.mz; afnog at afnog.org
> Subject: Re: IXP hardware
>
>
> ever wonder why the route server approach is no longer used at the
> big exchanges in the states, europs, ...?
>
> ever wonder why the route reflector approach is not used in the
> ...?
>
> they have bad failure modes and add no real value.
>
> do it the old-fashioned direct inter-isp bgp peering way.  it's
> simple.  it's stupid.  simple and stupid are reliable and scale
> well.
>
> randy
>


-----
This is the afnog mailing list, managed by Majordomo 1.94.5

To send a message to this list, e-mail afnog at afnog.org
To send a request to majordomo, e-mail majordomo at afnog.org and put
your request in the body of the message (i.e use "help" for help)

This list is maintained by owner-afnog at afnog.org