[Planetlab-devel] supporting multiple IP address per node -> "ip_addresses" sliver attribute

[Thierry Parmentelat]

hi marc

On Oct 2, 2008, at 11:08 PM, Marc E. Fiuczynski wrote:

> Hello Faiyaz, Thierry, Reid, and Daniel, (and everyone else who  
> cares):
>
> Daniel added support to give a sliver its own ip address.  This is  
> done by adding a sliver attribute called "ip_addresses" whose value  
> is the string version of the specific IP address.
any comment on why this is called ip_addresses and not ip_address ? is  
it intended to be able to specify several entries here ?
>
>
> NM appears to handle this on the fly, but it requires that the node  
> itself has an interface alias for the specific ip address in  
> question.  What's currently required to do that is the following:
>
> a) create a NodeNetworkSettingType called "alias" AND create a  
> sliver attribute called "ip_addresses"
>    - [This should be done as part of the basic MyPLC initialization]
there's already an 'alias' predefined in db-config; not that the  
description would be crystal clear though....
ip_addresses is still missing
>
> b) for each IP address alias assigned to a node create a new  
> NodeNetwork
>  - the "method" field of this NodeNetwork should be static,
>     which in turns requires the full complement of information
>     (gateway, dns entries, etc.) are required
as much as possible I would advocate that we try and get that by  
ourselves;
otherwise we're putting a huge burden on the operations team;
the ip address should be enough in most cases, and we could deduce the  
rest, that's very likely going to be similar to the what the primary  
interface is using, no ?
>
>
> c) for each of these NodeNetwork entries serving as an alias, do
>  - AddNodeNetworkSetting(NODENETWORKID,"alias","0")
I think the values need to be incremented, so that we use eth0:0 ..  
eth0:<n> for the pseudo-interfaces that are based on the same physical  
one. Daniel ?
>
> d) the GUI needs to be amended to include an alias "checkbox" when  
> creating a NodeNetwork.  If checked, the GUI would make the call  
> outlined in step (c) above.  Of course, it would be good for the GUI  
> to also show that a particular NodeNetwork is considered to serve as  
> an alias.
I don't know what the planetlab GUI does here; the one we have allows  
to manage interface (let's use the new names :) settings; it's basic  
but does the job, in a flat and generic manner (you can set settings  
individually and that's it). Of course this can be greatly improved  
once the feature is getting used.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Picture 3.png
Type: image/png
Size: 79014 bytes
Desc: not available
Url : http://lists.planet-lab.org/pipermail/devel/attachments/20081003/2a185e7c/Picture3-0001.png
-------------- next part --------------



>
>
> If you guys agree this is about right, then I'll create the  
> appropriate tickets for this.
>
> Thierry, please comment on how this fits in with your new {i,n}link  
> work?

ilinks allow to specify arbitrary relations between interfaces; (i  
haven't written the nlink part yet, but the intention is to link nodes  
of course); my initial goal was for modelling dummynets, [[our first  
implementation was relying on specific objects and one of the  
objectives of this 5.0 api was to allow to rewrite the dummynet thing  
based on generic objects instead]]
I had not thought about this use case for ilinks, but it could be used  
to model e.g. the fact that a fake interface is piggybacking a  
physical one (how is this done today if we have several physical  
interfaces ?), which in turn could help autoconfiguration. Or any  
other thing in fact.

>
>
> Marc
>
>
>
> _______________________________________________
> Devel mailing list
> Devel at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/devel