[Server-devel] eth0 statii IP

[John Watlington]

On Jul 3, 2008, at 2:41 PM, Tim Moody wrote:

>
> ----- Original Message -----
> From: "Martin Langhoff" <martin.langhoff at gmail.com>
> To: "Tim Moody" <timmoody at sympatico.ca>
> Cc: <server-devel at lists.laptop.org>
> Sent: Thursday, July 03, 2008 12:57 PM
> Subject: Re: [Server-devel] eth0 statii IP

The script which performs network configuration (olpc-scripts/network- 
config) is separate
from the script which performs network configuration at first boot  
(init.d/olpc-network-config)
to allow manual operation and simplify debugging.  It is not called  
by any other script.

>> On Thu, Jul 3, 2008 at 12:09 PM, Tim Moody <timmoody at sympatico.ca>  
>> wrote:
>>> /fsroot.olpc/etc/sysconfig/olpc-scripts/network_config closes the  
>>> loop by
>>> creating /etc/sysconfig/olpc_net_config, but ONLY if it  
>>> succeeds.  It
>>> will
>>
>> Correct. If you change your network hw on the machine, you should run
>> the network_config script with a parameter of "1", or remove the
>> olpc_net_config flag file and reboot.
>>
>>> fail if there are more than two wired ethernet adapters (or  
>>> none); I have
>>> three so it always fails.
>>
>> Ouch. What is the role of the 3rd adapter? Is it valid / useful for
>> our core scenarios?
>
> My machine doubles as a honey wall (2 adapters bridged and 1 to  
> monitor).  I
> don't know if there is a legitimate need for 3 wired adapters in an XS
> scenario, but I have seen a couple of others on this list with the  
> same
> problem (possibly test machines wanting to simulate the mesh over  
> wired

Why would you want to simulate the mesh (outside of doing mesh  
routing alg.
research) ?   Please don't confuse mesh networking with the ability to
collaborate, which works with any network connection between laptops
(WiFi, mesh, wired).  We currently don't recommend that any school  
(except
for ones with less than fifteen students) use mesh networking.   We  
suggest
traditional 802.11g APs, connected to a server over wired ethernet be  
used instead.

> because we can't get an antenna).

Why can't you get an antenna ?   You can request one through our  
developer
program.   Please let me know if you are having problems obtaining  
antennas
through that program --- there have been problems in the past but we  
think
they are fixed!

> Either way, the script could probably accomodate more
> wired adapters.  It already knows what's there.

Sure.  When that script was written, it wasn't clear how long it would
last (I was guessing a month, but it has been almost a year).  If it  
isn't
going to be replaced soon, adding support for three or four adapters
would be easy.

But all adapters beyond the first should be routed subnets, as we are
trying to prevent multicast and broadcast traffic from propagating.
You won't get your honeywall configuration...

> In general a little more flexibility for people who are testing  
> would be
> nice including, while I'm at it, the ability to partition the  
> machine during
> the install.

Disks are cheap.   Why wouldn't we want testers all using the same
partition layout ?   The current partition layout can be critized and
changed, but preserving sanity is easier with a common partitioning
scheme across the XS servers.

BTW, that is a one line change in the kickstart file --- you can  
trivially
build your own school server image that lets you manually set  
partitions...

We need another name for "mega-servers", which provide school server
services, but have a different network, service/server mapping, and  
storage
scheme (partitioning or separate disk servers) than the XS.  Suggestions
accepted

Cheers,
wad