We haven't had any problems with the 100's: the original eepro (10 Mbps)
seem to be real crap.
The tulip-based cards seem to get better performance when everything
aligns correctly: the eepro100 seems to never perform in a stellar manner,
but it is a lot more reliable and never performance badly either.
Note that our setup has fairly long cables: everything is within
specifications and the cable quality is good, but it may well be that this
is one factor that brings on problems. Maybe the tulip cards work better
on short lines with no line delay, for example - but the fact remains that
the Transmeta network is easily within spec (we're not pushing it, we're
just approaching getting close to the max length) and the tulip cards had
problems while the eepro100 do not.
I'm sure there are other good cards: I _thought_ the tulip ones were
supposed to be among the best. So I'm not saying that the eepro100 is the
best card out there, but I'd personally trust it more than others if only
because I've seen a lot of them work under conditions that broke other
cards..
> We have quite a few of them here, and not once with the PCI NIC's have
> I heard of anyone here having them going at full blast in full-duplex
> mode. The only case where I have heard of them going full blast was a
> test using a the eepro on the motherboard of one of the new p2 LX
> motherboards.
I agree that we haven't gotten top performance with the eepro100, but with
the tulips we had packets just getting lost etc. Losing a few packets due
to the collision detection being flaky means that even if you get perfect
performance for everything else, the actual bandwidth you get at an
application layer is only a small fraction of what it should be.
In short, I'll take a steady 5MB/s every day over a flaky 10MBps that
occasionally drops to 100kBps (yes, it really dropped to just kilobytes
per second when it was doing badly with the tulip thing.. )
Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu