Scaling the Number of Network Interfaces on Linux

David, Roopa,

to follow your talk at nedev11, I already proposed some times ago a patch to
remove sysctl, which was rejected. You can see the thread here:
http://thread.gmane.org/gmane.linux.network/285840

Instead of removing completly the sysctl entries, another idea could be to
manage a group of interfaces which will share the same subtree.


Regards,
Nicolas

On 2/10/16 12:15 PM, Nicolas Dichtel wrote:

to follow your talk at nedev11, I already proposed some times ago a
patch to
remove sysctl, which was rejected. You can see the thread here:
http://thread.gmane.org/gmane.linux.network/285840

Thanks for the reference. I'll take a look.

Instead of removing completly the sysctl entries, another idea could be to
manage a group of interfaces which will share the same subtree.

This come out from a side conversation as well -- for example to have 
interfaces enslaved to a bridge or bond share the same devconf.

This is certainly possible to do and I can give it a look. The key point 
is that we clearly need a means to lighten the overhead of a network 
interface.

David

Le 10/02/2016 12:59, David Ahern a écrit :

On 2/10/16 12:15 PM, Nicolas Dichtel wrote:

[snip]

quoted

Instead of removing completly the sysctl entries, another idea could be to
manage a group of interfaces which will share the same subtree.

This come out from a side conversation as well -- for example to have interfaces
enslaved to a bridge or bond share the same devconf.

It could also be interesting to have another way to group interfaces. For
example, group all ppp interfaces.

This is certainly possible to do and I can give it a look. The key point is that
we clearly need a means to lighten the overhead of a network interface.

Yes, I agree.