[tulip] The REALLY problem of Transmit-Timout!

[Anonymous Sender]

Hi for all!

I have a little network with one Linux server (Pentium II 350MHz 256MB RAM 2x8GB IDE) and four Windows98 workstations (different CPU / RAM / HD / NIC / VGA). From last 2 years I have upgrade the Linux box with different distributions & kernels (RH 5.x/6.x, Mandrake 5.x/6.x, kernels 2.0.x/2.2.x). I use Samba (from 2.0.0 to 2.0.7) on Linux for emulate a NT server (very good).

The problem are the continuos TRANSMIT ERROR (timeouts) in the network. To test the network I make 'ping –t server' in any Windows client, and when 'transmit timeout' occurs I need to manually restart the HUB. With this the network restart, and we can continue out job. This problem occurs randomly: days no, days little, days heavy...

I have tested with TWO different HUBS (one 100MB x 8 ports / actually 10/100 x 8 ports); and TWO different NICs on Linux box: 3Com 905B PCI and Intel EEPRO 100+ PCI. I installed more than 20 drivers for two NICs (originals from kernel source, modified from official "Linux Network Drivers' at SCYLD (old on NASA server), modified from different people at mailing list on TUX, betas, and official drivers from 3Com and Intel), with different options (Multicast limit, max_interrupt, debug), and different compile variations (gcc/pgcc, -O/O2/O6). WITH ALL I have problems, but...

I have read most messages from last 8 moth on mailing list on TUX, for 3Com 59x/90x NIC, EEPRO100 NIC and Digital 21*4* NIC. On ALL lists there are people with TIMEOUTS PROBLEMS. You think that the problem is on hardware? I think NO !!!!! (Please test I said, the are messages for all different hardware: http://www.tux.org/hypermail/linux-vortex/ ; http://www.tux.org/hypermail/linux-tulip/ ; http://www.tux.org/hypermail/linux-eepro100/ )

If I restart eth0 with ifconfig, the network restart; if I power off-on the Hub, the network restart; if I unplug-plug any network wire of the network, the network... obviously... restart. FROM ALL THIS...

PLEASE!, make a modified driver, that when RESTART de NIC, restart ALL components (chip, transceiver, queues) and, THIS IS THE MOST IMPORTANT, "RESTART THE ETHERNET INTERFACE WITH THE KERNEL". I think the problem is on the queues with Ethernet interface (at O.S. level): at moment you don't flush queues after a network error, or no restart send-receive status after restart NIC. Is this possible? What you think?

Please, send any comment, or Transmit-Timeout problem continues... at eternum!

Thanks for all!

LARS18TH