cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Hugo Trippaers <>
Subject Re: [MERGE] network-guru-orchestration into master
Date Thu, 31 Oct 2013 09:44:08 GMT

On 31 okt. 2013, at 04:34, Marcus Sorensen <> wrote:

> Here's a scenario... Lets say we currently have a zone deployed with Vlan
> isolation. There's nothing stopping us from running a vxlan isolation on
> the same hosts, even on the same physical interface. This is actually done
> in devcloud-kvm where public and mgmt traffic are selected by traffic label
> (which are tagged vlans), and guest traffic is handled by vxlan.
> If one wanted to deprecate Vlan in favor of vxlan without a big zone
> shufflr it would be simple if isolation method were a property of network
> offering. I imagine other isolation offerings could also be deployed in
> tandem, even if it required multiple physical interfaces in the host.

Currently the system we have in place is indeed capable of running different isolation types
on a host, it uses the physical network concept in cloudstack to determine which isolation
type to use. This is mainly because the physical network also defines which tag or network
interface label to use on the hypervisors.

I understand the case that it might be possible to combine different isolation types on physical
networks, like the case you describe with vlan and vxlan. However in other cases it should
not happen, like when nicira nvp is used which takes exclusive control of a traffic label
or bridge on the hypervisor. As far as i understood the contrail implementation it actually
takes control of the entire networking of a hypervisor, so its exclusive with any other isolation

Within the database and (most of) the code it is already possible to set multiple isolation
types on a physical network. That gives the admin the ability to link a certain physical interface/traffic
label or bridge to a set of isolation types. For example cloudbr0 with NONE, VLAN and VXLAN
and cloudbr1 with STT. The offering system can be used to indicate which isolation provider
to use for which particular virtual network.



> On Wed, Oct 30, 2013 at 11:36 AM, Pedro Roque Marques
> <> wrote:
>> Why ? That seems to me the most obvious way to do it.
>> There are different networking solutions: e.g. VLANs and overlays such as
> OpenContrail that assume an L3 switching topology.  For a deployment one
> would tend to choose a solution associated with the physical network.
> Sure that would be simple.  The problem is that Zone also implies
> other things.  Secondary/Primary storage for example.  Zone puts other
> limitations on things.  So by tying networking to a zone you can get
> into the situation where somebody has a CloudStack installation, and
> they want to add OpenContrail.  If we tie it to the zone, that may
> mean that in order for them to use OpenContrail they may need to get
> another NFS secondary storage, primary storage, compute nodes, etc and
> then copy all the templates from one zone to another, etc, etc.
> So currently you can't have Basic and Advanced networking in the same
> zone.  I think it should be possible.  Imagine I had a smaller
> dev/test cloud.  Today if I want basic and advance networking (because
> those are two distinict work loads), I have to have servers dedicated
> to each type.  So 10 for basic, 10 or advanced.  Now the networking
> under the hood for basic and advance can mix quite easily but just
> because we decided to implement it as it is today, I'm suddenly forced
> to managed two pools of physical resources, when it would have been
> must easier and more cost effective to have one pool.
> I think this is general thing that should change with CloudStack over
> time.  Zone and Pod should not be so closely tied to network concepts.
> CloudStack should be capable of being looser for people who want more
> hybrid work loads.  We too easily say that a service provider is just
> going to choose one or the other so why support mixed configurations.
> But there are more people that could be using CloudStack but are
> turned off by the strict models and concepts that CloudStack enforces.
> I have to say I was one of those people.  CloudStack forced me to
> manage my infrastructure in a certain way that I didn't particularly
> care for.
> Darren

View raw message