Why is STP turned off?

Why is STP turned off?

[John Haxby]

I've been having trouble with bridges and I noticed that STP is 
explicitly turned off.  This in spite of the brctl man page:

       brctl stp <bridge> <state> controls this bridge  instance’s  
participa-
       tion in the spanning tree protocol. If <state> is "on" or "yes" 
the STP
       will be turned on, otherwise it will be turned off.  When  
turned  off,
       the  bridge  will not send or receive BPDUs, and will thus not 
partici-
       pate in the spanning tree protocol.  If  your  bridge  isn’t  
the  only
       bridge  on the LAN, or if there are loops in the LAN’s topology, 
DO NOT
       turn this option off. If you turn this option off, please know 
what you
       are doing.

So, I guess someone knew what they were doing

Why did you turn STP off?

jch

Re: Why is STP turned off?

[Keir Fraser]

On 30/4/08 14:50, "John Haxby" <john.haxby@oracle.com> wrote:

> So, I guess someone knew what they were doing
> 
> Why did you turn STP off?

All bridge interfaces but the external interface are guest vif's which are
typically not hiding bridges. This simple topology does not require STP.

 -- Keir

RE: Why is STP turned off?

[Caitlin Bestler]

> -----Original Message-----
> From: xen-devel-bounces@lists.xensource.com [mailto:xen-devel-
> bounces@lists.xensource.com] On Behalf Of Keir Fraser
> Sent: Thursday, May 01, 2008 8:33 AM
> To: John Haxby; xen-devel
> Subject: Re: [Xen-devel] Why is STP turned off?
> 
> On 30/4/08 14:50, "John Haxby" <john.haxby@oracle.com> wrote:
> 
> > So, I guess someone knew what they were doing
> >
> > Why did you turn STP off?
> 
> All bridge interfaces but the external interface are guest vif's which
> are
> typically not hiding bridges. This simple topology does not require
> STP.
> 
>  -- Keir

The guest vifs are indeed very unlikely to be acting as bridges.
And any switch that only has a single uplink and N internal links
(none of which lead to a Bridge) can indeed decide not be an 802.1
Bridge and therefore not run spanning tree.

But if Xen is not running spanning tree and one of the Guest VIFs
*does* run spanning tree the results can be quite messy. An explicit
warning on this might make sense.

Re: Why is STP turned off?

[Keir Fraser]

On 1/5/08 21:29, "Caitlin Bestler" <Caitlin.Bestler@neterion.com> wrote:

>> All bridge interfaces but the external interface are guest vif's which
>> are
>> typically not hiding bridges. This simple topology does not require
>> STP.
>> 
>>  -- Keir
> 
> The guest vifs are indeed very unlikely to be acting as bridges.
> And any switch that only has a single uplink and N internal links
> (none of which lead to a Bridge) can indeed decide not be an 802.1
> Bridge and therefore not run spanning tree.
> 
> But if Xen is not running spanning tree and one of the Guest VIFs
> *does* run spanning tree the results can be quite messy. An explicit
> warning on this might make sense.

Actually I can't remember why we originally turned off STP. It may have been
because it took longer for the bridge to 'settle' when new vifs came online.
On the other hand I may simply have turned off STP along with other
parameters (hello/learning latencies) as part of a blanket effort to make
the bridge dumb but efficient. If others have had good experiences with STP
enabled we could consider re-enabling it in the default bridge
configuration.

 -- Keir

Re: Why is STP turned off?

[Daniel P. Berrange]

On Thu, May 01, 2008 at 10:05:16PM +0100, Keir Fraser wrote:
> On 1/5/08 21:29, "Caitlin Bestler" <Caitlin.Bestler@neterion.com> wrote:
> 
> >> All bridge interfaces but the external interface are guest vif's which
> >> are
> >> typically not hiding bridges. This simple topology does not require
> >> STP.
> >> 
> >>  -- Keir
> > 
> > The guest vifs are indeed very unlikely to be acting as bridges.
> > And any switch that only has a single uplink and N internal links
> > (none of which lead to a Bridge) can indeed decide not be an 802.1
> > Bridge and therefore not run spanning tree.
> > 
> > But if Xen is not running spanning tree and one of the Guest VIFs
> > *does* run spanning tree the results can be quite messy. An explicit
> > warning on this might make sense.
> 
> Actually I can't remember why we originally turned off STP. It may have been
> because it took longer for the bridge to 'settle' when new vifs came online.
> On the other hand I may simply have turned off STP along with other
> parameters (hello/learning latencies) as part of a blanket effort to make
> the bridge dumb but efficient. If others have had good experiences with STP
> enabled we could consider re-enabling it in the default bridge
> configuration.

STP does cause problems with DHCP - particularly during installation we
have found problems with STP taking along time to settle causing the
DHCP requests to time out. So I'd recommend keeping it turned off by 
default.

Dan.
-- 
|: Red Hat, Engineering, Boston   -o-   http://people.redhat.com/berrange/ :|
|: http://libvirt.org  -o-  http://virt-manager.org  -o-  http://ovirt.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: GnuPG: 7D3B9505  -o-  F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|