Any support for *ISA* bus Fast Ethernet

Any support for *ISA* bus Fast Ethernet

Post by Stephan Green » Thu, 17 Oct 1996 04:00:00



I have the opportunity to place a spare 486-66 system onto an internal
100 MBps (aka Fast Ethernet) LAN as a general purpose web, SMTP/POP, and
FTP server.  Unfortunately, the system only has ISA-bus and VESA slots.

I recall seeing on Intel's web site that they have a fast ethernet card
for ISA systems (I think it is actually a National design they bought
and uses a different chip set than the rest of their EtherExpress line);
does anyone have any information about the ability of 2.x kernels
(RedHat 4.0 system) to support this card?  Using a 10Mbit card is not an
option, and I doubt I can get a spare pentium with PCI bus instead
(though that would certainly make life much easier!)

TIA!

Steve Greene

 
 
 

Any support for *ISA* bus Fast Ethernet

Post by Henry A Wor » Thu, 17 Oct 1996 04:00:00


: I have the opportunity to place a spare 486-66 system onto an internal
: 100 MBps (aka Fast Ethernet) LAN as a general purpose web, SMTP/POP, and
: FTP server.  Unfortunately, the system only has ISA-bus and VESA slots.
:<...>

Most of the ISA 100Base-TX cards (and some EISA) use the SMC 91c100
chispet which will work with the smc9194 driver. However, there is
a gotcha, the driver doesn't know how to set the card into MII-PHY mode
(the i/f to the 100Mbs transciever), you have to set it up in the
EEPROM from DOS (and clear the AUI-select). But the configuration
utility that came with my card (made by Danflex, Cogent also makes one
and perhaps a T4 card as well -- they both have web sites) doesn't
have an option to set that mode. SMC's website has an upgraded
version which does have the MII option, but doesn't seem to work
(I need to reverify that). However, the SMC site also has a diagnostic
called ETX which has an EEPROM test option that will allow you to set
the MII-select bit and clear the AUI-select bit in a menu like the
configuration utility and write this out to EEPROM. You may also
have to set the MTU to no more than 1499 with the ifconfig command,
I'm not sure yet if that problem is the SMC chip/driver or a problem
with the tulip driver on the other end of the crossover link. If you
should get a board with one of these chips I could send you patches
for the smc9194 driver that add support for forcing the MII select
and full duplex from insmod parms.

Between the ISA bus and being a PIO chip, performance is not great
and sustained transfers can consume all available cpu/bus cycles.
I generally see it sustaining around 2.2MBps for ftp transmit
and a bit better for receive. Probably not worth building a
100Base-Tx net around, though the price increment over name-brand
10Base NICs is less then the performance gain and it does let
you connect to an existing 100Base-Tx network. It's too bad no
one seems to be making a VLB NIC with this chip, it has a VLB
compatible sync i/f mode and since it's PIO it wouldn't run into
the VLB busmastering bugs and limitations that would plague any
attempts to market a board based on the more popular 100Mbps
chips (SMC even has artwork for a VLB board on its website).

I'm starting to experiment on 91c100 enhancements to add transmit
buffer preallocation in an attempt to cut the transmit latencies,
but it will be a couple weeks before I can spend much time on it.
Currently the driver has to request a buffer allocation from the
chip and then wait for an interrupt before transfering the transmit
data (can you say S_L_O_W). The 91c94 has a 2KB buffer for each direction,
so there isn't much alternative unless you severely limit MTU. The
91c100, OTH, has a 128KB shared buffer so it begs for a better buffer
allocation policy. The shared buffer also creates a problem that,
with the current 91c94 driver, incoming packets can fill the shared
buffer faster then they can be processed and thus block transmits, it
should be possible to also set the EEPROM to reserve a portion
of the buffer for transmits, but it's not an option in the
configurtation utility.

--

No, I don't speak for Amdahl... I'm not even sure I speak for myself.