On Tue, 27 May 2003, Jochen Kunz wrote:

> Well. I have Ethernet switches with FDDI uplink that do proper IP
> (de)fragmentation...

 Hmm, I'm not sure how joining of packets could be done this way as such a
switch cannot know if another packet is going to arrive anytime soon and
an Ethernet host surely cannot use an MTU above about 1500.  And the other
way it might not work anyway as packets are usually sent with the DF
option set to get the PMTU discovery working -- I don't know if such
switches respect it.

 Then any router passing packets through an interface having a smaller MTU
will invalidate an attempt to send them in full-size FDDI frames. 

 But an FDDI-based LAN or MAN (we have an FDDI ring as our academic MAN's
core -- originally 90 km long, now divided into three rings with ATM
interconnects) can still benefit from the bigger MTU. 

> Do I _have_ to handle the SMT in the driver, like the SysKonnect cards,
> or is this optional? I.e. does the DEFZA SMT in "hardware" like the
> DEF[TEPQ]A? 

 Yes, it does SMT in "hardware", but it doesn't receive nor send SMT
frames from/to the ring.  There are two addidional queues for SMT frames
provided by the firmware.  A driver is responsible for copying incoming
SMT frames to the firmware's receive queue and ones produced by the
firmware in the transmit queue to the ring.  As a side effect, you have
full control on what gets passed between the network and the firmware at
the SMT layer. 

> AFAIK the SMT code is not trivial. This is mainly the reason the
> SysKonnect (E)ISA / PCI or NPI SBus cards are not supported. This cards
> handle SMT in the driver. SysKonnect published a Linux driver that
> includes the SMT code, but this driver is GPLed... So no *BSD driver for
> them. 

 Ouch -- cannot they be convinced to re-release the code with a
BSD-compatible license?  At least the hardware-dependent part?  I suppose
they chose GPL to prevent the SMT handler to be reused commercially. 

  Maciej

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +