> -----Original Message-----
> From: Igor Chubin [mailto:igor@xxxxxxx]
> Sent: 27 March 2007 13:31
> To: Petersson, Mats
> Cc: Igor Chubin; xen-users@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-users] Maximum number of interfaces in DomU
>
> On Di, Mär 27, 2007 at 02:24:20 +0200, Petersson, Mats wrote:
>
> ...
> > > Does anybody know something about
> > > the restriction of 3 network interfaces
> > > in domU that was discussed here:
> > >
> > > http://lists.xensource.com/archives/html/xen-users/2006-05/msg
> > > 00024.html
> > > http://lists.xensource.com/archives/html/xen-users/2006-08/msg
> > > 00968.html
> > >
> > >
> > > Please tell, is the problem solved?
> >
> > No, and there is no plan to "solve" this "problem".
>
>
> Quotes around the words solve and problem
> mean that this is not a problem at all?
Well, that obviously depends on how you look at it. It is of course considered
a problem by some people. But it's not considered a problem by the Xen
developers community, I think, because no one has so far come up with a good
argument why you actually need a GUEST to have more than three virtual
interfaces. It's not limiting the bandwidth (which three physical connectors
could perhaps do), because the speed of each interface is limited only by CPU
and memory access speed in the guest and Dom0. It is of course limiting the
number of different sources that guest can get data from with only the
interface as a the distinguishing factor.
Can you explain what the setup is where you need this, and why you need more
than three virtual devices? It may help forward the argument for a fix, but
there may also be others who have the solved the same sort of problem within
the community here, without using more than three interfaces.
[Just to note: The reason it's limited to 3 interfaces is simply that's how
many fit into one 4KB memory section, which makes it easy to manage in the
code. Adding more interfaces would be possible, but it would complicate matters
by some amount, as the interface between Dom0 and DomU is based on physical
memory pages, and sending a list of those will make not only break backwards
compatibility, but also complicate the common case of using (up to) three
interfaces].
Finally, I'm indeed a Xen developer, but I've got very little to do with
virtual network interfaces, so I may not have grasped the whole reasoning why
this limit is and if/how difficult it is to change, nor any other complications
or other issues involved. I also have not been involved in the decision to "not
solve this problem".
But I believe that if this was actually a REAL PROBLEM to a lot of people, then
it would have been solved a long time ago. It may be a case of "yes, it's
possible to solve, but not enough people are asking for it", or it may simply
be that it's "not a real problem".
--
Mats
>
> --
> WBR, i.m.chubin
>
>
>
>
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home