Conference 7.286::fddi

	DEMFA can easily be converted to full Duplex mode.
	We have done it in the lab. to demonstrate
	its feasibility. There is no need to chage the
	driver. The CNS code needs to be changed so that
	it recocnizes that there is only one node on the
	other side, and enters full duplex mode. I believe
	Henry Yang designed the protocol.

	In the lab. we found that RMC FIFO did not drop
	any packets when DEMFA was exercised for various
	full duplex traffic, including very small packets
        (30 bytes). System architects for Gigaswitch based
	systems were very concerned about packets dropped at
	RMC FIFO. They did extensive experiments with DEMFA
	in full duplex mode and were glad to note
	that NO packets were dropped at RMC interface. They
	exercised DEMFA extensively for many different traffic patterns
	of interest to them.

	Of course note DEMFA was not desigened to meet
	200Megits/sec. full duplex traffic and does drop packets
	at the memory interface which is acceptable to Gigaswitch
	based system designers for If you want more
	info. send me mail.

	/satish 

The two major issues for any adapter would be:

1. does it have the internal bandwidth needed (particularly: to local packet
   memory, and to the bus)
2. would it make any difference in system performance anyway

On (2) -- if your host software isn't capable of driving the adapter at
wire speed (i.e., 100 Mb/s with idle time left over) then you will get NO
gain from full duplex.

	paul

	Ref .-1

	 Your argument is quite right but remember there are at least
	two performance parameters - Bandwidth and Latency. Most people
	think of bandwidth and forget latency. Is full duplex useful
	from latency viewpoint?

	/satish

That's true.  But systems that have a hard time approaching FDDI throughput
usually have excessive latency as well.

	paul