MED is evil? (was: Re: I-D ACTION:draft-mcpherson-rfc3065bis-00.txt)

[Curtis Villamizar <[e-mail]>]

In message <[e-mail]>, Danny McPherson writes:
> 
> > Just the fact that so many knobs are available sheds light on the
> > fact that many ISPs think MEDs are just evil and should be stripped.
> 
> And I tend to agree...
> 
> -danny


Whether to strip MED depends on what you are trying to accomplish with
MED.

Consider the case where ISP A and ISP B peer at multiple places and A
is directly connected to destination X and B is directly connected to
destination Y.

Some ISPs consider it "fair" if both ISPs take the shortest path out
of their own netork to the other.  In this case both will strip MED
before advertising to the IGP.  They will usually accept MED for the
purpose of determining which of multiple routers to select at a
peering point.

If ISP A has a much faster or less congested network than ISP B, ISP A
may accept MED from ISP B and route based on MED.  If ISP B sets MED
according to IGP cost, then a longer path might be taken through the
ISP A network but the shortest path will be taken in ISP B's network
in both directions.  This is often the case if B is a customer of A.

Curtis