Re: Requirements for IP Multihoming Architectures

[Tony Li <tli@Procket.com>]

Path bloat is far less onerous than prefix bloat, as path bloat can be
addressed by bigger faster processors with bigger memories, both of which
don't look to be a limiting factor.

Prefix bloat inflates the forwarding table, which is more painful becuause
it's a large memory that has to be accessed at line rate.

Tony




Joe Abley writes:
 | On Wed, Mar 21, 2001 at 04:04:58AM +1100, Geoff Huston wrote:
 | > >  + path or prefix bloat (the prefixes announced by the multi-homing
 | > >    site can be aggregated by the single provider)
 | > 
 | > path bloat is bad becuase?
 | > 
 | > I'm not sure I understand why path bloat is evil in and of itsxelf.
 | 
 | My understanding was:
 | 
 |   The larger the number of paths per prefix, the more expensive it is
 |   to walk the tree and do path selection. This impacts convergence time
 |   when processing updates and withdrawals (or, conversely, places greater
 |   demands on router hardware to manage acceptable convergance).
 | 
 | This is based on the assumptions that:
 | 
 |  + walking the tree is expensive,
 | 
 |  + walking a larger tree is more expensive,
 | 
 |  + network-wide convergence time increases with the time taken
 |    for individual nodes to process changes, and
 | 
 |  + the time taken to converge following changes needs to be controlled
 |    (within the observed period of oscillation of changes received from
 |    other networks, perhaps).
 | 
 | 
 | Joe