[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: tulip driver 2.4.x kernel



Since your last e-mail, I updated my firmware to 5.5-1, which according the 
Compaq site is the latest to dare go with this older hardware.  It made no 
difference in the outcome.  The de4x5 driver lasts longer than the tulip 
which makes sense because that is what it is (de-500-aa).

On Friday 22 June 2001 04:14 pm, you wrote:
> (I'm at home, and can't netscape the reply :-)
>
> ewa0_mode twist forces the driver to use 10mb half duplex.
> ewa0_mode fast forces the driver to 100mb half duplex
> ewa0_mode fastfd forces 100mb full duplex
> ewa0_mode auto allows the driver to ask the switch what
> they can do together. Hubs are 10 or 100mb half duplex.
> lucent P550 (big $$$) does auto right, and it works.
> lucent P120 usually does auto right.
> most cisco boxes DO NOT do auto right.
> the HP pro-curve switches do auto right.

Ok.  Now I see what you mean.  No, I can't change the router box and my 
building is limited to 10 Mb/s.  We are in the process of upgrading some of 
our site to 100 Mb/s and I will get there eventually.  So it seems like 
ewa0_mode twisted-pair must be correct for my case since de4x5 will not do 
full duplex anyway.   I have better networking at home than this place!

> I agree that there is a problem with the tulip driver
> in tulip mode. the de4x5 driver will not do full duplex,
> but it doesn't lock the machine up doing a re-init
> on "carrier error". The 2.4.0 kernel used the "old" tulip driver
> and worked. 2.4.1 - and later use the new driver, which
> isn't 100%.
> 	I use the de4x5 and ewa0_mode auto. So what if its half
> duplex. On a switch that doesn't really matter much.
>
> With srm 5.8-1 you can drop in an ee-pro-100B (intel) card,
> its called a DE600-aa :-)
> I hate intel cards, they do not autonegotiate at all.
> Usually you have to force them and the switch to use
> the duplex you want.
> If you have access to the switch, try the de4x5 driver,
> and set ewa0_mode to fastfd, set the switch to 100 full,
> no autonegotiate.

This is all good information to know when we do get the 100 Mb/s.
In the mean time, my card is suspect since some errors did show up.  I wonder 
of 2.4.x enhances this and causes the errors to show up prior?

Is there another card you might recommend that would work better and be 
compatible with the 5.5-1 SRM?  Maybe I could give Microway a call and see 
what they have now for the newer kernels.

> 	I dual boot redhat 7.0 and tru64 T5.1A-4 (field test site)
> on one of my XP1000's. I have a up2000 2gb 833mhz dual cpu...
> wow - what a screamer!

ORNL has a Compaq supercomputer, but I think they are running tru64 on it 
also.  I could not see an advantage for a single system running numerically 
intensive code.  If you have Compaq's compilers for Linux, you get the same 
throughput.  Now, if you need to parallel, then the better I/O might make it 
worth it.  Personally, I would like a Microway Beowulf system with lot's of 
processort running Linux for about $50K.

> 	If you suspect the card is bad, try booting tru64
> and watch a "ping -f"... any more than a few lost packets
> says things are BAD.

Don't boot Tru64 anymore (gone!).  I'm running Debian/Linux exclusively now.  
I'll try this under LInux.

> No, you don't need another card, just use 100 half and de4x5.
> post to the developer listed in the linux source tree
> and I'll do the same.

Maybe I can use this to justify the expense of getting to 100 Mb/s for our 
building!

> berkley

-- 
James D. Freels, P.E._i, Ph.D.
Oak Ridge National Laboratory
freelsjd@ornl.gov - work
jdfreels@home.com - home



Reply to: