Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 23 Aug 95 18:46:20 +0000
From:      Andrew.Gordon@net-tel.co.uk
To:        freebsd-hardware@freebsd.org
Subject:   Performance problem with SMC 10/100Mb Ethernet cards
Message-ID:  <"MAC-950823194610-09D3*/G=Andrew/S=Gordon/O=Net-Tel Computer Systems Ltd/PRMD=Net-Tel/ADMD=Gold 400/C=GB/"@MHS>

Next in thread | Raw E-Mail | Index | Archive | Help
Since SMC are currently offering a cheap deal on an 'evaluation kit' containing
two EtherPower 10/100 adaptors and a crossover cable to connect them, I ordered
one to try out.  The results so far are not encouraging:

My test setup is:

                   +--------------+                          +----------------+
                   | P90 PCI      |                          | 486DX4/100 PCI |
                   +--------------+                          +----------------+
                        |   |   crossover cable (100baseTX)         |
main ethernet (10base2) |   -----------------------------------------
------------------------+-------

Both machines are running 2.0.5R(*).  The main ethernet and the two-station
network are addressed as two separate class C networks.  The second interface in
the Pentium box is an SMC Etherpower 10.  The Pentium box is a made-by-intel
Triton-based motherboard.  I have tried two different 486 machines, one
made-by-intel motherboard and SCSI discs, one Taiwanese (Gigabyte?) motherboard
with IDE disc, with the same results.

My basic test is FTP-ing a large file (eg. kernel.GENERIC) between the two
systems.


Problems:

1) When operating at 100Mb/sec, there seems to be substantial (random) packet
loss going from the Pentium to the 486.  When you fluke an error-free transfer
throughput of about 2Mbyte/sec is achieved, but more commonly there are pauses
of a second or so giving a net througput of only 300kbyte/sec typically.  If
configured for 10Mb/sec (ifconfig de0 -link2), a consistent throughput of
1Mbyte/sec is observed, with no pauses.  Similar pauses are also noticable using
telnet, and also if the 486 is re-booted to Windows, accessing a Samba server on
the Pentium box.

2) Regardless of the interface speed, an FTP transfer from the 486 to the
Pentium runs smoothly (fast) if the FTP connection was made to the address of
the 10/100 interface in the Pentium box, but if the address specified was that
of the other interface in the same machine, throughput starts slow but gradually
speeds up, reaching full speed after about 25Kbyte.


In an attempt to eliminate the one piece of hardware I could replace, I changed
the SMC-supplied cat5 UTP cable (with RJ45 connectors) by a homemade cable using
IBM type1 cable and DB9 connectors, using the other connector on the cards.  My
homemade cable appeared to work a bit better for problem 1, but by no means
eliminated it (so the difference may be experimental error).


Since others have reported good things with these cards, I must be doing
somthing wrong.  Do I have a hardware fault, FreeBSD driver problem, or is
100Mb/sec just too fast for 486DX4/100 boxes?

Any ideas appreciated.....

Andrew.


-----
(*) initially tested with pure 2.0.5R.  I subsequently checked the driver in
-current, which had changed very slightly, so built a 2.0.5R kernel with the de
driver out of -current.  No change.



Want to link to this message? Use this URL: <http://docs.FreeBSD.org/cgi/mid.cgi?"MAC-950823194610-09D3*/G=Andrew/S=Gordon/O=Net-Tel Computer Systems Ltd/PRMD=Net-Tel/ADMD=Gold 400/C=GB/">