(Note the follow-up to tech-net...  I think this is more appropriate.)

[ On , February 17, 1999 at 00:23:52 (GMT), Matthias Scheler wrote: ]
> Subject: Re: 3c905b Ethernet Cards
>
> In article <199902162040.MAA05270@lestat.nas.nasa.gov>,
> 	Jason Thorpe <thorpej@nas.nasa.gov> writes:
> > If it's a NetBSD-current snapshot, it DOES support your 905b, BTW.
> 
> Where we are talking about the 3c905b... did anyone do a benchmark
> with "ttcp" using the new "ex" driver on a 100MBit/Sec network?
> 
> I would like to get rid of DEC 2114x cards due to the broken driver
> but didn't find a good replacement yet. Compaq ThunderLan based
> "tl" boards are too expensive, the "fxp" driver is slow as hell and
> the 3Com 3c905 has a slow hardware design.

I haven't actually had a chance to do any really good benchmarking with
anything TCP related, never mind ttcp, but plain old 'ping -f' suggests
to me that the "fxp" driver is a quite decent driver particularly if you
put some decent hardware behind it, such as the Intel EtherExpress Pro
10+/100B card (I believe the model number is PILA8460, but I'm not 100%
sure).  These cards are definitely not the cheapest, but seem to be one
of the best available without going all the way to Intel's new
server-class cards (which I don't expect are supported yet anyway).

The numbers below are from a pair of PII-300's running 1.3.3 with the
ping being run between two such cards running at 100baseT and connected
via an ethernet switch vs. the system's onboard le0 cards also running
at 100baseT and connected to separate ports and a separate VLAN on the
same switch.  The le0 ports connect routed network (NNN.NNN.NNN.*), and
the fxp0 cards connect the private network (192.168.1.*).  There may be
enough traffic on the routed network to account for the slightly lower
performance of the le0 intefaces, though I wouldn't want to bet on it.
The two machines in this test were effectively idle.

le0 at pci1 dev 5 function 0: PCnet-PCI Ethernet
le0: address 00:60:94:57:XX:XX
le0: 8 receive buffers, 2 transmit buffers
le0: interrupting at irq 11

fxp0 at pci0 dev 15 function 0: Intel EtherExpress Pro 10+/100B Ethernet
fxp0: interrupting at irq 15
fxp0: Ethernet address 00:90:27:32:XX:XX


03:45 [757] # ping -c 10000 -f NNN.NNN.NNN.251
PING public.my.domain (NNN.NNN.NNN.251): 56 data bytes
----public.my.domain PING Statistics----
10000 packets transmitted, 10000 packets received, 0% packet loss
round-trip min/avg/max = 0.204/0.218/0.481 ms
  4395.5 packets/sec sent,  4395.7 packets/sec received		[[ le0 @56 ]]

03:46 [758] # ping -c 10000 -f 192.168.1.251
PING 192.168.1.251 (192.168.1.251): 56 data bytes
----192.168.1.251 PING Statistics----
10000 packets transmitted, 10000 packets received, 0% packet loss
round-trip min/avg/max = 0.194/0.205/0.407 ms
  4652.7 packets/sec sent,  4653.1 packets/sec received		[[ fxp0 @56 ]]

03:46 [759] # ping -c 10000 -s 1400 -f NNN.NNN.NNN.251
PING public.my.domain (NNN.NNN.NNN.251): 1400 data bytes
----public.my.domain PING Statistics----
10000 packets transmitted, 10000 packets received, 0% packet loss
round-trip min/avg/max = 0.756/0.771/1.187 ms
  1249.3 packets/sec sent,  1249.3 packets/sec received		[[ le0 @1400 ]]

03:47 [760] # ping -c 10000 -s 1400 -f 192.168.1.251
PING 192.168.1.251 (192.168.1.251): 1400 data bytes
----192.168.1.251 PING Statistics----
10000 packets transmitted, 10000 packets received, 0% packet loss
round-trip min/avg/max = 0.667/0.681/0.873 ms
  1408.4 packets/sec sent,  1408.5 packets/sec received		[[ fxp0 @1400 ]]

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>