[HECnet] Circuit costs (again!)

[Paul_Koning at Dell.com]

> On Jan 25, 2016, at 12:08 PM, Robert Armstrong <bob at jfcl.com> wrote:
> 
>> Paul_Koning  (Paul_Koning at Dell.com) wrote:
>> just setting all costs to the same number would not be all that horrible,
> either
> 
>  The effect of this would be to make DECnet prefer the route with the least
> number of hops.  That makes a lot of sense to me and I would absolutely
> agree, BUT there's a problem...
> 
>  The problem is the HECnet bridge program.  The bridge effectively makes
> everything adjacent to everything else, even though the actual communication
> goes thru psilo.  Thus things look to DECnet like they're only one hop away
> when in reality they're at least two.

That's true.  So set a bridge circuit cost to 2N if everyone else is N, or if using delay-based costs, set it to 2 times the average delay based cost between bridge and its clients.

As more datalink support is being added to various places, it might be worth while moving away from bridge connections where that is reasonable.  Multinet over TCP is a good alternative.  GRE is ok, too.  DDCMP (as, for example, implemented in SIMH V4.0) is good.

	paul