Hi Rafael,

I think you’re overcomplicating things a bit. As Simon said – all switches used 
in the same zone need to be configured to trunk the same VLAN range. If your 
zone uses VLANs 100-200 then all switches serving the PODs and clusters in that 
zone need to be able to trunk VLANs 100-200, such that you maintain traffic 
between VMs in the same isolated network. When UserA spins up a network and a 
VM, he e.g. gets allocated VLAN 101 – and it should not matter where in the 
zone his VMs end up – i.e. they are spread across PODs and clusters – the VM 
should always be able to connect to VLAN 101.

With regards to the nuts and bolts of this – yes you are working at Layer 2 on 
the switches, but the mechanism used to transfer traffic between them has 
nothing to do with CloudStack.

Regards,
Dag Sonstebo
Cloud Architect
ShapeBlue

On 28/02/2017, 18:58, "Rafael Weingärtner" <rafaelweingart...@gmail.com> wrote:

    Ok, let’s see if I understood what we are doing.
    
    Within a switch domain, we send packets using the mac address of network
    interface cards. The switch has a table matching mac addresses to ports
    they are being used in; therefore, it can efficiently execute the
    forwarding. When you say that we expect broadcast domain across a zone, you
    mean that switches of this zone are exchanging information regarding the
    mac address of the network interfaces they "know". Is that it?
    
    If there is a packet (Frame) that is addressed to a mac address to a
    network interface that is not within a domain of a switch, this switch may
    forward this packet to another switch that “knows” the interface that has
    the destination mac address. is that what you guys mean by "expecting
    broadcast domain across a zone"?
    
    On Tue, Feb 28, 2017 at 1:45 PM, Simon Weller <swel...@ena.com> wrote:
    
    > Rafael,
    >
    >
    > ACS expects the same layer 2 broadcast domain across all pods in a zone.
    > So your switches have to trunk all vlans. This can get really nasty for
    > larger typologies (due to vlan limits and spanning tree), hence why
    > technologies like vxlan, stt and gre were invented.
    >
    >
    > - Si
    >
    > ________________________________
    > From: Rafael Weingärtner <rafaelweingart...@gmail.com>
    > Sent: Tuesday, February 28, 2017 12:41 PM
    > To: users@cloudstack.apache.org
    > Subject: Re: CloudStack Advanced networking doubt
    >
    > Great, thanks for the explanation Dag.
    > So, for two VMs that are on the same virtual network, but different PODs
    > (consequently in a different layer -2 domain switch) how does ACS handle 
in
    > this situation?
    >
    >
    > On Tue, Feb 28, 2017 at 1:36 PM, Dag Sonstebo <dag.sonst...@shapeblue.com>
    > wrote:
    >
    > > Hi Rafael,
    > >
    > > in the confines of that zone yes. All switches serving one zone need to
    > > trunk the same VLANs, no matter how you configure your PODs or clusters.
    > >
    > > Regards,
    > > Dag Sonstebo
    > > Cloud Architect
    > > ShapeBlue
    > >
    > > On 28/02/2017, 18:31, "Rafael Weingärtner" <rafaelweingart...@gmail.com>
    > > wrote:
    > >
    > >     You mean, once a user allocates a VLAN’s (let’s say tag 1), in all 
of
    > > the
    > >     switches this VLAN tag is reserved?
    > >
    > >     On Tue, Feb 28, 2017 at 12:48 PM, Dag Sonstebo <
    > > dag.sonst...@shapeblue.com>
    > >     wrote:
    > >
    > >     > Hi Rafael,
    > >     >
    > >     > Keep in mind for an advanced zone the broadcast domain for VLANs 
is
    > > the
    > >     > zone rather than the POD, i.e. VMs in the new POD would use the
    > same
    > > VLANs
    > >     > as the previous VMs in the original POD.
    > >     >
    > >     > Regards,
    > >     > Dag Sonstebo
    > >     > Cloud Architect
    > >     > ShapeBlue
    > >     >
    > >     > On 28/02/2017, 16:16, "Rafael Weingärtner" <
    > > rafaelweingart...@gmail.com>
    > >     > wrote:
    > >     >
    > >     >     Hi folks,
    > >     >     I was checking some information regarding ACS advanced
    > networking
    > >     >     deployment mode, and I ran into this figure [1]. This made me
    > > wonder,
    > >     > what
    > >     >     would happen with the following scenario.
    > >     >
    > >     >     Let`s say I have a similar scenario as the one depicted in
    > > figure [1],
    > >     > a
    > >     >     set of pods with a set of clusters that have a set of hosts.
    > > Each host
    > >     > in a
    > >     >     pod is linked directly using a Layer-2 switch. Let’s assume
    > > there exist
    > >     >     network/aggregation layers that are configured properly and
    > > provide
    > >     > access
    > >     >     to VMs in the cloud using the public IP net. Let’s also assume
    > > that the
    > >     >     public IP net is 1.1.1.0/24; the management and storage
    > > networks are
    > >     > on
    > >     >     isolated networks and are properly set up (Assume also that we
    > > are
    > >     > using
    > >     >     the advanced networking mode).
    > >     >
    > >     >     Now, I create a guest network 2.2.2.0/24. When I deploy a user
    > > VM,
    > >     > ACS will
    > >     >     deploy a VR (let’s call x) with an IP (1.1.1.1) in the public
    > > net, and
    > >     >     other on the guest network (2.2.2.1). Then, this VR(x) will
    > > execute the
    > >     >     firewalling/forwarding for my newly created user VM.
    > >     >
    > >     >     Let’s now imagine that I keep deploying user VMs to a point in
    > > which
    > >     > the
    > >     >     POD gets full. The next VM I create ACS will have to deploy in
    > > other
    > >     > PODs
    > >     >     of the environment. Because this new user VM will be in a
    > > different
    > >     > POD,
    > >     >     the communication with other user VMs is not straightforward
    > > anymore
    > >     > (not a
    > >     >     matter of using the same VLAN and net). What will ACS do to
    > link
    > > users
    > >     > VMs
    > >     >     that are on the same virtual network, but deployed in 
different
    > > PODs?
    > >     >
    > >     >     Will it deploy other VR(y) with an IP (let's say 1.1.1.2) on
    > the
    > > new
    > >     > POD
    > >     >     and create a route between VR(x) and VR(y) using the public
    > > network, so
    > >     >     that the communication for VMs in network 2.2.2.0/24 is
    > > transparent?
    > >     >
    > >     >     http://docs.cloudstack.apache.org/projects/cloudstack-
    > >     > administration/en/4.8/_images/network-setup-zone.png
    > >     >
    > >     >     --
    > >     >     Rafael Weingärtner
    > >     >
    > >     >
    > >     >
    > >     > dag.sonst...@shapeblue.com
    > >     > www.shapeblue.com<http://www.shapeblue.com>
    > ShapeBlue - The CloudStack Company<http://www.shapeblue.com/>
    > www.shapeblue.com
    > Introduction Upgrading CloudStack can sometimes be a little daunting - but
    > as the 5P's proverb goes - Proper Planning Prevents Poor Performance.
    >
    >
    >
    > >     > 53 Chandos Place, Covent Garden, London  WC2N 4HSUK
    > >     > @shapeblue
    > >     >
    > >     >
    > >     >
    > >     >
    > >
    > >
    > >     --
    > >     Rafael Weingärtner
    > >
    > >
    > >
    > > dag.sonst...@shapeblue.com
    > > www.shapeblue.com<http://www.shapeblue.com>
    > ShapeBlue - The CloudStack Company<http://www.shapeblue.com/>
    > www.shapeblue.com
    > Introduction Upgrading CloudStack can sometimes be a little daunting - but
    > as the 5P's proverb goes - Proper Planning Prevents Poor Performance.
    >
    >
    >
    > > 53 Chandos Place, Covent Garden, London  WC2N 4HSUK
    > > @shapeblue
    > >
    > >
    > >
    > >
    >
    >
    > --
    > Rafael Weingärtner
    >
    
    
    
    -- 
    Rafael Weingärtner
    


dag.sonst...@shapeblue.com 
www.shapeblue.com
53 Chandos Place, Covent Garden, London  WC2N 4HSUK
@shapeblue
  
 

Reply via email to