[ovs-discuss] Strange behavior with paths longer than 5 hops.

Vjekoslav Brajkovic balkan at cs.washington.edu
Wed Jul 18 03:09:19 UTC 2012


I can't reduce the number of hops. We're doing large scale emulations using
topologies that are synthesised from the Rocketfuel dataset. Therefore, the
design can't be more efficient.

The example above was used only to explain the conditions under which this
problem occurs.

All broadcast traffic that we send is forwarded along the spanning tree, so
I don't see any reason why the machine would crash.

On Tue, Jul 17, 2012 at 7:54 PM, Jesse Gross <jesse at nicira.com> wrote:

> If you remove it at some point the machine will crash.  I think your
> time is better spent reducing the number of hops that you have.  I
> don't know what your goal is but your current design is extremely
> inefficient.
>
> On Tue, Jul 17, 2012 at 7:41 PM, Vjekoslav Brajkovic
> <balkan at cs.washington.edu> wrote:
> > Would you happen to know what needs to be patched in order to
> > modify/remove this behavior?
> >
> > On Tue, Jul 17, 2012 at 7:38 PM, Jesse Gross <jesse at nicira.com> wrote:
> >> On Tue, Jul 17, 2012 at 7:32 PM, Vjekoslav Brajkovic
> >> <balkan at cs.washington.edu> wrote:
> >>> Hi,
> >>>
> >>> I ran into this problem and I think it might be a bug. The problem is
> >>> most likely somewhere on the datapath, as it appears to be happening
> >>> with and without a controller.  Here are the steps to reproduce it.
> >>>
> >>> (container 0) <--> sw0 <--> sw1 <--> sw2 <--> sw3 <--> sw4 <--> sw5
> >>> <--> sw6 <--> sw7 <--> (container 1)
> >>>
> >>> It appears to be happening with paths that are above 5 hops. Both with
> >>> OVS 1.4.2 and 1.6.1.
> >>
> >> It's not a bug, it's intentional behavior.  In order to avoid blowing
> >> out the kernel stack in the event of a loop or excessive number of
> >> hops we limit the number of times a packet can traverse the datapath
> >> to 5.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20120717/ae3afcac/attachment.html>


More information about the discuss mailing list