Re: Possible a better order to switch connections
- From: Paul Ionescu <i_p_a_u_l yahoo com>
- To: networkmanager-list gnome org
- Subject: Re: Possible a better order to switch connections
- Date: Mon, 03 Oct 2005 01:17:30 +0300
On Sun, 02 Oct 2005 11:53:21 -0700, Karl Hegbloom wrote:
> On Sun, 2005-10-02 at 15:43 +0300, Paul Ionescu wrote:
>> IMHO, a better approach is to let the old connection running, configure
>> the new connection, switch to the new connection, (optionally notify all
>> apps that there was a net change), and deactivate the old connection.
>>
>> This way, there is no net downtime, and I don't think there is a major
>> impact in NM.
>
> The "downtime" is brief, and the new interface cannot have the same IP as
> the old one if you don't bring the old one down first.
The "downtime" is sometime not so brief, and why not avoid the downtime if
possible ?
Besides, one can have the same IP on multiple interfaces if needed.
You can test it.
> Open TCP "stream"
> network connections will stop working when you bring down the interface
> they are routed over, since their IP goes away. For "connectionless" UDP,
> that's probably not a problem, unless the application layer keeps track of
> your IP address and something happens when that changes.
>
> Question:
>
> If, for instance, the wired connection is brought down, and then the wifi
> is brought up right away and given the same IP that the wired interface
> just had, will open TCP connections continue to function? That is to say,
> will the Linux networking code move the connection to the new interface?
>
> What if there was a configuration option (perhaps a default) to have it
> move the old IP over to the new interface, as a secondary IP, provided the
> network and mask are the same? Given that, what if there was a way to
> flag it to be automatically deleted once all active connections on it are
> gone? Should that be in-kernel, or something NetworkManager does?
>
> It seems to me that this is probably possible and a use-case that may well
> have been considered during the design and implementation of the Linux
> networking stack. Am I correct?
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]