Gregg C Levine <[EMAIL PROTECTED]> writes:
> Can any of you good people answer a question? How is Linux doing his/its
> networking? Is it exactly the way the I386 distribution designers
> suggested it would, on that platform?
Yes and no. Yes, Linux networking on S/390 is the same as
on Intel x86 platforms. All the tools work the same (ping,
ifconfig, traceroute, etc.), routing works the same, etc.
But no, it varies from the i386 distributions in one point:
autoconfiguration. My recent i386 experience is from Red Hat,
but I believe the paradigm is basically the same everywhere.
During installation, the installer probes the system to
determine what network card you have and how to configure
it, and it just works (DON'T ask me about fun with my
ancient NE-200-ish Ethernet card, though!).
For S/390, the choices are fewer - 5 instead of over 100.
They require just as much configuration (about that Ethernet
card ... no, never mind), but it doesn't happen automatically.
You've got a few decisions to make, but they're pretty easy.
> I have been seeing more then the
> usual amount of posts regarding the subject of how to configure a
> specific adapter for the guest, and it is starting to confuse me.
I agree, it's an interesting issue. Every platform seems to have its
Big Problems, and one of the biggest for Linux/390 seems to be
networking (another is ISV support, but even Oracle appears to have
seen the light).
A year or so ago, the networking questions generally took two forms:
1) "Which network interface should I want?" - IUCV, CTC,
or LCS or OSA. If you weren't lucky enough to have a
spare OSA port, you had an actual decision to make,
because Linux under VM can use either IUCV or CTCs just
as easily. Consensus wound up being "use IUCV if you
can find a stable version of the device driver, use
CTC otherwise".
2) "Why can't the rest of the world reach my Linux/390
under VM?" - the answer was almost always a problem with
routing (a technical problem) and getting your network
admin to give you what you need (a communications problem).
The other common answer was both - proxy ARP.
These days all the questions appear to be about problems using the
OSA card and it's associated binary-only device driver. There have
been a variety of incompatibilities, sometimes involving kernel levels
("Does anyone have a copy of qeth.o for 2.4.35-ac32-pre42-greased-turkey?")
or microcode updates. The routing issue seems to have died down,
for which I think we can thank both the mailing-list membership
and the authors of the various Linux/390 redbooks.
I think the kernel-incompatibility problem could be made to die
down, but only IBM can do what's necessary. I think they understand
that - the 21 January 2002 experimental patches to the 2.4.7 kernel
include a version of the "no jiffies" patch that doesn't interfere
with the binary-only drivers as long as you don't enable the
CONFIG_NO_HZ_TIMER option. That's a big step forward. If you want to
run with the option enabled, you need different drivers, but having read
the patch I'm not convinced it could be done any other way except by
shipping some or all of the driver source code.
Ross Patterson
Computer Associates
