Performance: AsanteFAST versus Farallon EtherFast?

eharmon

6502
So, it's been benchmarked here before that the AsanteFAST 10/100 Ethernet card...just isn't too fast. Unsurprisingly older OT just can't push very quick anyway, and so the returns are really diminished.


Curiously though, looking at photos on Bitsavers, I realized there's a difference between that card and the Farallon "EtherFast":

The AsanteFAST uses an SMSC LAN91C100, while the Farallon uses a LAN91C100FD. What's that mean? The LAN91C100FD properly supports full duplex transfers.


Note that I'm not sure the Farallon is really an "EtherFast" at all...perhaps this was a Fast EtherTX 10/100? Farallon has it pegged at 20MBps "bus performance" which is kinda corny...guess they didn't even support NuBus 90!: https://web.archive.org/web/19980524081443/http://www.farallon.com/product/fen/fecards.html

Has anyone ever compared the two? I wonder if the Farallon wins...
 
What is the maximum theoretical transfer speed of NuBus? The bus alone presents a huge bottleneck. Add in that these cards are likely polled I/O driven and not bus mastering DMA (read: Cheap) attached to an OS not exactly known for its I/O throughput.
 
If you can do 11MBs on a scuzzy Nubus card, you should be able to get more then 10Mb speed with a network Nubus card.

Apples networking sucks in 68K and early PPC machines. I can get much faster transfers on a 486 with VLB or EISA 10Mb ethernet.
 
I got a maximum of 866KiB/sec on a AsanteFAST 10/100 with a Rocket under System 6 for a download and 606KiB/sec upload. Agree with Unknown_K that the networking performance is mediocre, but still very easy to set up compared to NFS...
 
Today I was doing some research and noticed in a picture the Farallon uses a full duplex chip and I was curious to learn more! So then I found my own thread here...oops.

Anyway, they do share pinouts, so I suspect it's a drop in replacement:

LAN91C100: https://datasheet4u.com/datasheet-pdf/SMSCCorporation/LAN91C100/pdf.php?id=582511
LAN91C100FD: https://rocelec.widen.net/view/pdf/logpzofyr8/SMSCS00288-1.pdf

Application Note: https://static5.arrow.com/pdfs/2004/0512/semi_a/2/smsc/application note/an82.pdf

So, theoretically it's a drop in replacement, except you need to activate full duplex mode by setting a bit (see page 23 of the FD datasheet - SWFDUP), so it won't have full duplex without modifying the ROM/driver.

Probably does nothing for ethernet performance either -- it's clearly bottlenecked by the OS. And on a switched network, half duplex isn't slowing down your other devices.

So overall, I'm still curious if the Farallon is faster (it may not set the duplex bit anyway), but I suspect modding a (rare) AsanteFAST doesn't have much value, unless the chip happens to be dead.
 
So overall, I'm still curious if the Farallon is faster (it may not set the duplex bit anyway), but I suspect modding a (rare) AsanteFAST doesn't have much value, unless the chip happens to be dead.
It does not, in fact, set the duplex bit:
The PN990 and PN994 Farallon cards do not support Full Duplex mode in Macintosh machines.
This is taken from the Farallon Fast Ethernet FAQ. PN990 is the FastEtherTX-10/100 NuBus Card. As a knock against it, it also claims not to support autonegotiation.

On the plus side, I have noted that the AsanteFAST has 128KiB of buffer, so it may interrupt less, or at least drop fewer packets when the machine is busy.

I wonder if this was, in part, to reduce talk time on 100Mbps hubs. Giving the product a purpose even if it's still dog slow on Classic.

EDIT: Well...okay nevermind on the first comment. There's a Farallon PN990 with a 3com chipset, and a PN990a with the SMSC! My guess is the a rev is quite a bit better, if you come across one: https://web.archive.org/web/2001012...com/support/software/download/fenubusupd.html
 
Last edited:
Back
Top