🚀 go-pugleaf

In article <1992Nov17.191249.21391@den.mmc.com>, geger@phantom.den.mmc.com (George Eger (303) 971-6974) writes:
> 
> Is there a particular reason why SAS rings of more than 2 stations are not
> allowed by SMT?  Is this just a ploy to sell more hardware?  Seems to me
> there are good system configurations that could warrant SAS rings (like a
> bunch of equipment all in one rack/room that doesn't need the complication
> of DAS or concentrators).
> 
> Comments?


I've played with such variations of FDDI.  They are a royal pain to get
and keep working.  Without CMT signalling, all you know is that the
whole ring is dead.  You don't know which machine is powered off, or
which connector is loose, or whatever.  Or, worse, what combination of
such things is wrong.

The cost of CMT is only about 300 lines of C (ironically about the same
whether you do it all yourself in the host or "node processor" or if
you use one of the fancy PHY chips that claim to do it all for you).
You do need another set of ODL's, which cost <$100/pair.

Much more complication is involved in other parts of FDDI and SMT, with
much lower return on investment, than insisting on a dual ring.

One skary kludge that is effective and cheap is to have a very few
single attach stations lurk in the primary ring between two dual
attached stations.  The single attached stations simply pass symbols
except when the ring is idle or active, and then they look for tokens
and do normal MAC things.  I don't think this kludge is worth the
effort.

If you can afford the other over-designed parts of FDDI, then you can
afford the dual ring.  If you can't, then look into other media.  One
hack I've thought sounds neat and cheap is something I've called
"gatling ethernet."


I think that a the "one rack/room" situation calls for simple
dual-attached stations, without concentrators.


vjs