[c-nsp] understanding switching

[Jon Lewis]

On Sun, 5 Jun 2005, Ryan O'Connell wrote:

> >ospf's auto-cost reference-bandwidth supports a range of 1-4294967 on the
> >various platforms I've looked at.  Is there a reason not to default to a
> >much larger value, or even the top end of the allowed range?
>
> When OSPF was invented, 64k and 2Mb/s circuits were common. 10Mb/s
> Ethernet was regarded as fast and 16Mb/s Token Ring had only just been
> released. Having a higher default wouldn't have allowed enough
> granularity on slow circuts.

Ah...the cost has to fit (or be truncated to fit) into a 16 bit uint, so
if you go too high with the reference bandwidth, all your "low speed"
links end up with an equal cost of 65535...just as with the default, all
your 100mbit or better links end up with a cost of 1.  I suppose that
could be a problem.

It looks like any modern network could probably get away with setting ospf
reference-bandwith to somewhere around 1000-10000 or more depending on
what their smallest links are that they'd care about differentiating.

----------------------------------------------------------------------
 Jon Lewis                   |  I route
 Senior Network Engineer     |  therefore you are
 Atlantic Net                |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________