collisions with 1.09t

[Stephen Williams]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Greetings …

I recently recompiled my kernel and after found that I was getting massive
collisions on the NIC under heavy load (which are enough to disconnect the
NIC if allowed to continue). I had not noticed these problems previously. I
tried retrogrades and recompiles but to no effect.

I’ve also tried swapping the NIC for an older model with a different chip.
No difference.

My hub is not one of the best. When I remove the hub and connect via a
crossover cable to an identical NIC the collision problem vanishes.

Incidentally, the same model NIC connected to the same hub making the same
transfers under Windows produces no collisions.

My long-term solution will be to replace the hub with a better quality unit
but I am isolated and cannot do this immediately. (I had planned to do this
anyway.)

So my question is, how might I detune the NIC so as to reduce the collision
problem? In Intel’s Windows drivers it is possible to do so. Is there any
way of doing this with eepro100?

Also, how might I check to be sure that the NIC is not erroneously
defaulting to full duplex?

My impression is that this problem does have something to do with eepro100.
It seems to show up only under certain circumstances and then only under the
heaviest of loads. I suspect that the recompile did do something as I am
reasonably sure that this problem didn’t exist previously (the only change I
made was to CONFIG_FILTER). Also it seems to occur only under the heaviest
of loads (for example, FTP isn’t a problem but transfers under SAMBA are).
It wouldn’t surprise me if the authors are not experiencing this problem
because they are using better equipment. But likely the common man will.

FYI, here is the technical info on my NIC (thought as I said, I’ve tried
other older models) and eepro100 version (which is now running under
2.2.15pre15):

eepro100.c:v1.09j-t 9/29/99 Donald Becker
http://cesdis.gsfc.nasa.gov/linux/drivers/eepro100.html
eepro100.c: $Revision: 1.20.2.3 $ 2000/03/02 Modified by Andrey V. Savochkin
<saw@saw.sw.com.sg> and others
eth0: Intel PCI EtherExpress Pro100 at 0xc2826000, 00:90:27:BB:18:3E, IRQ
11.
  Receiver lock-up bug exists -- enabling work-around.
  Board assembly 721383-007, Physical connectors present: RJ45
  Primary interface chip i82555 PHY #1.
  General self-test: passed.
  Serial sub-system self-test: passed.
  Internal registers self-test: passed.
  ROM checksum self-test: passed (0x04f4518b).

If I can provide any other info please ask.

Best,

S.
==========================================================
PGP keys: http://www.ucc.uconn.edu/~jewil/sjw.html
RSA: 4E11 E5BB 3F2D E856  AC10 2F49 5C82 9BBB
DH/DSS: DCED DE1E F46C 1081 B67E  B3A7 1EC1 AC37 2E69 04C4
==========================================================

-----BEGIN PGP SIGNATURE-----
Version: PGP 6.5.3

iQA/AwUBONLYmh7BrDcuaQTEEQKT1gCgqT3QLuhnhoy6qTmZU69qugvyacYAoJQr
G7djb4g7Jytr8VUXoy52Fl7b
=I0HD
-----END PGP SIGNATURE-----

-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-eepro100-request@beowulf.org