Francois Desloges wrote:
> On Mon, 15 Jan 2001, Andi Kleen wrote:
> > On Mon, Jan 15, 2001 at 08:30:12PM +0100, A.N.Kuznetsov wrote:
> > > What's about VLANs, they can be handled as separate virtual devices
> > > provided you have _couple_ of them. It the number is higher, they must
> > > be clustered as single nbma interface via framing (i.e neighbour) level
> > > or via tags in routing tables. The same thing is with MPLs.
> > > That's why I strongly dislike the idea to create zillions of net_devices
> > > and consider that approach to VLANs as stupid one. And this is reason
> > > why hashing device list (being great in principle) is not considered
> > > to be some really required feature.
> > Is there any evidence that people really want to use hundreds of VLANs on a
> > single box in practice?
> We plan to use hundreds if not thousands, on our 2,5 Tbps, linux
> controlled, MAN switch-router.
> We may especially use VIDs as Client (or servce) identifier, for billing
> purposes, once they have been aggegated on a single links past the really
> (typically only L2 switched) hop. Such machines will be used in Metropoles to
> connect a _lot_ of people.
I think my VLAN patch will scale to that without any problem, though
I have not run serious traffic over that many interfaces. I do think
you'll want the device-hashing patch for that application, even if
most users of VLAN will not need it.
The one linear lookup currently in the VLAN code (that I know of), is
based on the number of physical ethernet interfaces that are running
ethernet. That could be fixed with some smarter code on my part,
but since the number has been low, sofar, I have not optimized it yet.
> And I know that there is other hardware startup that plan to use Linux
> on their high speed platform as well (we're not alone :-)
Makes you wonder what a dual CPU Ghz machine with a bunch of Gb NICs in
it could do :)
Ben Greear (greearb@xxxxxxxxxxxxxxx) http://www.candelatech.com
Author of ScryMUD: scry.wanfear.com 4444 (Released under GPL)