Provider Networking in VMware Cloud Director

This is going to be a bit longer than usual and more of a summary / design option type blog post where I want to discuss provider networking in VMware Cloud Director (VCD). By provider networking I mean the part that must be set up by the service provider and that is then consumed by tenants through their Org VDC networking and Org VDC Edge Gateways.

With the introduction of NSX-T we also need to dive into the differences between NSX-V and NSX-T integration in VCD.

Note: The article is applicable to VMware Cloud Director 10.2 release. Each VCD release is adding new network related functionality.

Provider Virtual Datacenters

Provider Virtual Datacenter (PVDC) is the main object that provides compute, networking and storage resources for tenant Organization Virtual Datacenters (Org VDCs). When a PVDC is created it is backed by vSphere clusters that should be prepared for NSX-V or NSX-T. Also during the PVDC creation the service provider must select which Network Pool is going to be used – VXLAN backed (NSX-V) or Geneve backed (NSX-T). PVDC thus can be backed by either NSX-V or NSX-T, not both at the same time or none at all and the backing cannot be changed after the fact.

Network Pool

Speaking of Network Pools – they are used to create on-demand routed/isolated networks by tenants. The Network Pools are independent from PVDCs, can be shared across multiple PVDCs (of the same backing type). There is an option to automatically create VXLAN network pool with PVDC creation but I would recommend against using that as you lose the ability to manage the transport zone backing the pool on your own. VLAN backed network pool can still be created but can be used only in PVDC backed by NSX-V (same for very legacy port group backed network pool now available only via API). Individual Org VDCs can (optionally) override the Network Pool assigned of its parent PVDC.

External Networks

Deploying virtual machines without the ability to connect to them via network is not that usefull. External networks are VCD objects that allow the Org VDC Edge Gateways connect to and thus reach the outside world – internet, dedicated direct connections or provider’s service area. External network have associated one or more subnets and IP pools that VCD manages and uses them to allocate external IP addresses to connected Org VDC Edge Gateways.

There is a major difference how external networks are created for NSX-V backed PVDCs and for NSX-T ones.

Port Group Backed External Network

As the name suggest these networks are backed by an existing vCenter port group (or multiple port groups) that must be created upfront and is usually backed by VLAN (but could be a VXLAN port group as well). These external networks are (currently) supported only in NSX-V backed PVDCs. Org VDC Edge Gateway connected to this network is represented by NSX-V Edge Service Gateway (ESG) with uplink in this port group. The uplinks have assigned IP address(es) of the allocated external IPs.

Directly connected Org VDC network connected to the external network can also be created (only by the provider) and VMs connected to such network have uplink in the port group.

Tier-0 Router Backed External Network

These networks are backed by an existing NSX-T Tier-0 Gateway or Tier-0 VRF (note that if you import to VCD Tier-0 VRF you can no longer import its parent Tier-0 and vice versa). The Tier-0/VRF must be created upfront by the provider with correct uplinks and routing configuration.

Only Org VDC Edge Gateways from NSX-T backed PVDC can be connected to such external network and they are going to be backed by a Tier-1 Gateway. The Tier-1 – Tier-0/VRF transit network is autoplumbed by NSX-T using 100.64.0.0/16 subnet. The allocated external network IPs are not explicitly assigned to any Tier-1 interface. Instead when a service (NAT, VPN, Load Balancer) on the Org VDC Edge Gateway starts using assigned external address, it will be advertised by the Tier-1 GW to the linked Tier-0 GW.

There are two main design options for the Tier-0/VRF.

The recommended option is to configure BGP on the Tier-0/VRF uplinks with upstream physical routers. The uplinks are just redundant point-to-point transits. IPs assigned from any external network subnet will be automatically advertised (when used) via BGP upstream. When provider runs out of public IPs you just assign additional subnet. This makes this design very flexible, scalable and relatively simple.

Tier-0/VRF with BGP

An alternative is to use design that is similar to the NSX-V port group approach, where Tier-0 uplinks are directly connected to the external subnet port group. This can be useful when transitioning from NSX-V to T where there is a need to retain routability between NSX-V ESGs and NSX-T Tier-1 GWs on the same external network.

The picure below shows that the Tier-0/VRF has uplinks directly connected to the external network and a static route towards the internet. The Tier-0 will proxy ARP requests for external IPs that are allocated and used by connected Tier-1 GWs.

Tier-0 with Proxy ARP

The disadvantage of this option is that you waste public IP addresses for T0 uplink and router interfaces for each subnet you assign.

Note: Proxy ARP is supported only if the Tier-0/VRF is in Active/Standby mode.

Tenant Dedicated External Network

If the tenant requires direct link via MPLS or a similar technology this is accomplished by creating tenant dedicated external network. With NSX-V backed Org VDC this is represented by a dedicated VLAN backed port group, with NSX-T backed Org VDC it would be a dedicated Tier-0/VRF. Both will provide connectivity to the MPLS router. With NSX-V the ESG would run BGP, with NSX-T the BGP would have to be configured on the Tier-0. In VCD the NSX-T backed Org VDC Gateway can be explicitly enabled in the dedicated mode which gives the tenant (and also the provider) the ability to configure Tier-0 BGP.

There are seprate rights for BGP neighbor configuration and route advertisement so the provider can keep BGP neighbor configuration as provider managed setting.

Note that you can connect only one Org VDC Edge GW in the explicit dedicated mode. In case the tenant requires more Org VDC Edge GWs connected to the same (dedicated) Tier-0/VRF the provider will not enable the dedicated mode and instead will manage BGP directly in NSX-T (as a managed service).

Often used use case is when the provider directly connects Org VDC network to such dedicated external network without using Org VDC Edge GW. This is however currently not possible to do in NSX-T backed PVDC. There instead, you will have to import Org VDC network backed by NSX-T logical segment (overlay or VLAN).

Internet with MPLS

The last case I want to describe is when the tenant wants to access both Internet and MPLS via the same Org VDC Edge GW. In NSX-V backed Org VDC this is accomplished by attaching internet and dedicated external network portgroups to the ESG uplinks and leveraging static or dynamic routing there. In an NSX-T backed Org VDC the provider will have to provision Tier-0/VRF that has transit uplink both to MPLS and Internet. External (Internet) subnet will be assigned to this Tier-0/VRF with small IP Pool for IP allocation that should not clash with any other IP Pools.

If the tenant will have route advertisement right assigned then route filter should be set on the Tier-0/VRF uplinks to allow only the correct prefixes to be advertised towards the Internet or MPLS. The route filters can be done either in NSX-T direclty or in VCD (if the Tier-0 is explicitly dedicated).

The diagram below shows example of an Org VDC that has two Org VDC Edge GWs each having access to Internet and MPLS. Org VDC GW 1 is using static route to MPLS VPN B and also has MPLS transit network accessible as imported Org VDC network, while Org VDC GW 2 is using BGP to MPLS VPN A. Connectivity to the internet is provided by another layer of NSX-T Tier-0 GW which allows usage of overlay segmens as VRF uplinks and does not waste physical VLANs.

One comment on usage of NAT in such design. Usually the tenant wants to source NAT only towards the Internet but not to the MPLS. In NSX-V backed Org VDC Edge GW this is easily set on per uplink interface basis. However, that option is not possible on Tier-1 backed Org VDC Edge GW as it has only one transit towards Tier-0/VRF. Instead NO SNAT rule with destination must be used in conjunction with SNAT rule.

An example:

NO SNAT: internal 10.1.1.0/22 destination 10.1.0.0/16
SNAT: internal 10.1.1.0/22 translated 80.80.80.134

The above example will source NAT 10.1.1.0 network only to the internet.

VMware Cloud Director Cells Behind Internet Proxy

VMware Cloud Director cells are usually deployed in the management cluster and their access to Internet might be limited due to security considerations. This can be a problem because certain features do require outgoing access to external (Internet) resources:

  • Catalog subscription: the cell will need access to the published catalog URL
  • Multisite: if you associate multiple Organizations together, some API calls are fan-out by the cell to the respective associated API endpoints, therefore the cell needs to be able to access them (even its own external API endpoint)
  • Cell Appliance VAMI repository for patches or upgrades

The latest VCD release 10.2.1 now does support internet proxy which means there is no need to have full internet access to the management environment.

On the VCD Appliance the proxy can be configured by editing /etc/sysconfig/proxy file:

 

root@vcloud1 [ ~ ]# cat /etc/sysconfig/proxy
# Enable a generation of the proxy settings to the profile.
# This setting allows to turn the proxy on and off while
# preserving the particular proxy setup.
#
PROXY_ENABLED="yes"

# Some programs (e.g. wget) support proxies, if set in
# the environment.
# Example: HTTP_PROXY="http://proxy.provider.de:3128/"
HTTP_PROXY="http://proxy.fojta.com:3128"

# Example: HTTPS_PROXY="https://proxy.provider.de:3128/"
HTTPS_PROXY="http://proxy.fojta.com:3128"

You need to restart vmware-vcd service to apply the configuration.

NSX-T 3.1: Sharing Transport VLAN between Host and Edge Nodes

When NSX-T 3.1 was released a few days ago, the feature that I was most looking for was the ability to share Geneve overlay transport VLAN between ESXi transport nodes and Edge transport nodes.

Before NSX-T 3.1 in a collapsed design where Edge transport nodes were running on ESXi transport nodes (in other words NSX-T Edge VMs were deployed to NSX-T prepared ESXi cluster) you could not share the same transport (TEP) VLAN unless you would dedicate separate physical uplinks for Edge traffic and ESXi underlay host traffic. The reason is that the Geneve encapsulation/decapsulation was happening only on the physical uplink in/egress and that point would be skipped for intra-host datapath between the Edge and host TEP VMkernel port.

This was quite annoying because the two transport VLANs need to route between each other at full jumbo MTU>1600 frame size. So in lab scenarios you had to have additional router taking care of that. And I have seen multiple time issues due to  misconfigured router MTU size.

After upgrading my lab to NSX-T 3.1 I was eager to test it.

Here are the steps I used to migrate to single transport VLAN:

  1. The collapsed Edge Nodes will need to use trunk uplinks created as NSX-T logical segment. My Edge Nodes used regular VDS port group so I renamed the old ones in vCenter and created new trunks in NSX-T Manager.
  2. (Optional) Create new TEP IP Address Pool for the Edges. You can obviously use the ESXi host IP Pool as now they will share the same subnet, or you can use static IP addressing. I opted for new IP Address Pool with the same subnet as my ESXi host TEP IP Address Pool but a different range so I can easily distinguish host and edge TEP IPs.
  3. Create new Edge Uplink Profile VLAN to match the ESXi transport VLAN.
  4. Now for each Edge node repeat this process: edit the node in the Edge Transport Node Overview tab, change its Uplink Profile, IP Pool and uplinks to the created ones in steps #1, #2 and #3. Refresh and observe the Tunnel health.
  5. Clean up now unused Uplink Profile, IP Pool and VDS uplinks.
  6. Deprovision now unused Edge Transport VLAN from physical switches and from the physical router interface.

During the migration I saw one or two pings to drop but that was it. If you see tunnel issues try to put the edge node briefly into NSX Maintenance Mode.

New Networking Features in VMware Cloud Director 10.2

The 10.2 release of VMware Cloud Director from networking perspective was a massive one. NSX-V vs NSX-T gap was closed and in some cases NSX-T backed Org VDCs now provide more networking functionality than the NSX-V backed ones. UI has been redesigned with new dedicated Networking sections however some new features are currently available only in API.
Let me dive straight in so you do not miss any.

NSX-T Advanced Load Balancing (Avi) support

This is a big feature that requires its own blog post. Please read here. In short, NSX-T backed Org VDCs can now consume network load balancer services that are provided by the new NSX-T ALB / Avi.

Distributed Firewall and Data Center Groups

Another big feature combines Cross VDC networking, shared networks and distributed firewall (DFW) functionality. The service provider first must create Compute Provider Scope. This is basically a tag – abstraction of compute fault domains / availability zones and is done either at vCenter Server level or at Provider VDC level.

The same can be done for each NSX-T Manager where you would define Network Provider Scope.

Once that is done, the provider can create Data Center Group(s) for a particular tenant. This is done from the new networking UI in the Tenant portal by selecting one or multiple Org VDCs. The Data Center Group will now become a routing domain with networks spanning all Org VDCs that are part of the group, with a single egress point (Org VDC Gateway) and the distributed firewall.

Routed networks will automatically be added to a Data Center Group if they are connected to the group Org VDC Edge Gateway. Isolated networks must be added explicitly. An Org VDC can be member of multiple Data Center Groups.

If you want the tenant to use DFW, it must be explicitly enabled and the tenant Organization has to have the correct rights. The DFW supports IP Sets and Security Groups containing network objects that apply rules to all connected VMs.

Note that only one Org VDC Edge Gateway can be added to the Data Center Group. This is due to the limitation that NSX-T logical segment can be attached and routed only via single Tier-1 GW. The Tier-1 GW is in active / standby mode and can theoretically span multiple sites, but only single instance is active at a time (no multi-egress).

VRF-Lite Support

VRF-Lite is an object that allows slicing single NSX-T Tier-0 GW into up to 100 independent virtual routing instances. Lite means that while these instances are very similar to the real Tier-0 GW they do support only subset of its features: routing, firewalling and NATing.

In VCD, when tenant requires direct connectivity to on-prem WAN/MPLS with fully routed networks (instead of just NAT-routed ones), in the past the provider had to dedicated a whole external network backed by Tier-0 GW to such tenant. Now the same can be achieved with VRF which greatly enhances scalability of the feature.

There are some limitations:

  • VRF inherits its parent Tier-0 deployment mode (HA A/A vs A/S, Edge Cluster), BGP local ASN and graceful restart setting
  • all VRFs will share its parent uplinks physical bandwidth
  • VRF uplinks and peering with upstream routers must be individually configured by utilizing VLANs from a VLAN trunk or unique Geneve segments (if upstream router is another Tier-0)
  • an an alternative to the previous point EVPN can be used which allows single MP BGP session for all VRFs and upstream routers with data plane VXLAN encapsulation. Upstream routers obviously must support EVPN.
  • the provider can import into VCD as an external network either the parent Tier-0 GW or its child VRFs, but not both (mixed mode)

IPv6

VMware Cloud Director now supports dual stack IPv4/IPv6 (both for NSX-V and NSX-T backed networks). This must be currently enabled via API version 35 either during network creation or via PUT on the OpenAPI network object by specifying:

“enableDualSubnetNetwork”: true

In the same payload you also have to add the 2nd subnet definition.

 

PUT https://{{host}}/cloudapi/1.0.0/orgVdcNetworks/urn:vcloud:network:c02e0c68-104c-424b-ba20-e6e37c6e1f73

...
    "subnets": {
        "values": [
            {
                "gateway": "172.16.100.1",
                "prefixLength": 24,
                "dnsSuffix": "fojta.com",
                "dnsServer1": "10.0.2.210",
                "dnsServer2": "10.0.2.209",
                "ipRanges": {
                    "values": [
                        {
                            "startAddress": "172.16.100.2",
                            "endAddress": "172.16.100.99"
                        }
                    ]
                },
                "enabled": true,
                "totalIpCount": 98,
                "usedIpCount": 1
            },
            {
                "gateway": "fd13:5905:f858:e502::1",
                "prefixLength": 64,
                "dnsSuffix": "",
                "dnsServer1": "",
                "dnsServer2": "",
                "ipRanges": {
                    "values": [
                        {
                            "startAddress": "fd13:5905:f858:e502::2",
                            "endAddress": "fd13:5905:f858:e502::ff"
                        }
                    ]
                },
                "enabled": true,
                "totalIpCount": 255,
                "usedIpCount": 0
            }
        ]
    }
...
    "enableDualSubnetNetwork": true,
    "status": "REALIZED",
...

 

The UI will still show only the primary subnet and IP address. The allocation of the secondary IP to VM must be either done from its guest OS or via automated network assignment (DHCP, DHCPv6 or SLAAC). DHCPv6 and SLAAC is only available for NSX-T backed Org VDC networks but for NSX-V backed networks you could use IPv6 as primary subnet (with IPv6 pool) and IPv4 with DHCP addressing as the secondary.

To enable IPv6 capability in NSX-T the provider must enable it in Global Networking Config.
VCD automatically creates ND (Neighbor Discovery) Profiles in NSX-T for each NSX-T backed Org VDC Edge GW. And via /1.0.0/edgeGateways/{gatewayId}/slaacProfile API the tenant can set the Edge GW profile either to DHCPv6 or SLAAC. For example:
PUT https://{{host}}/cloudapi/1.0.0/edgeGateways/urn:vcloud:gateway:5234d305-72d4-490b-ab53-02f752c8df70/slaacProfile
{
    "enabled": true,
    "mode": "SLAAC",
    "dnsConfig": {
        "domainNames": [],
        "dnsServerIpv6Addresses": [
            "2001:4860:4860::8888",
            "2001:4860:4860::8844"
        ]
    }
}

And here is the corresponding view from NSX-T Manager:

And finally a view on deployed VM’s networking stack:

DHCP

Speaking of DHCP, NSX-T supports two modes. Network mode (where DHCP service is attached directly to a network and needs an IP from that network) and Edge mode where the DHCP service runs on Tier-1 GW loopback address. VCD now supports both modes (via API only). The DHCP Network mode will work for isolated networks and is portable with the network (meaning the network can be attached or disconnected from the Org VDC Edge GW) without DHCP service disruption. However, before you can deploy DHCP service in Network mode you need to specify Services Edge Cluster (for Edge mode that is not needed as the service runs on the Tier-1 Edge GW).  The cluster definition is done via Network Profile at Org VDC level.

In order to use DHCPv6 the network must be configured in Network mode and attached to Org VDC Edge GW with SLAAC profile configured with DHCPv6 mode.

Other Features

  • vSphere Distributed Switch support for NSX-T segments (also known as Converged VDS), although this feature was already available in VCD 10.1.1+
  • NSX-T IPSec VPN support in UI
  • NSX-T L2VPN support, API only
  • port group backed external networks (used for NSX-V backed Org VDCs) can now have multiple port groups from the same vCenter Server instance (useful if you have vDS per cluster for example)
  • /31 external network subnets are supported
  • Org VDC Edge GW object now supports metadata

NSX-V vs NSX-T Feature Parity

Let me conclude with an updated chart showing comparison of NSX-V vs NSX-T features in VMware Cloud Director 10.2. I highlighted new additions in green.

Enable MAC Learning as Default on vSphere Distributed Switch

This short PowerCLI script will change the vSphere Distributed Switch default port group configuration to enable MAC learning policy. This means every port group on such switch inherits this configuration and will have MAC learning enabled unless specifically disabled.

For more information why would you need that read William’s Lam blog.

$vds = get-vdswitch 'DSwitch1'
$spec = New-Object VMware.Vim.VMwareDVSConfigSpec
$spec.DefaultPortConfig = New-Object VMware.Vim.VMwareDVSPortSetting
$spec.DefaultPortConfig.MacManagementPolicy = New-Object VMware.Vim.DVSMacManagementPolicy
$spec.DefaultPortConfig.MacManagementPolicy.MacLearningPolicy = New-Object VMware.Vim.DVSMacLearningPolicy

$spec.DefaultPortConfig.MacManagementPolicy.MacLearningPolicy.Enabled = $True
$spec.DefaultPortConfig.MacManagementPolicy.MacLearningPolicy.AllowUnicastFlooding = $True
$spec.DefaultPortConfig.MacManagementPolicy.MacLearningPolicy.Limit = 4000
$spec.DefaultPortConfig.MacManagementPolicy.MacLearningPolicy.LimitPolicy = "DROP"
$spec.ConfigVersion = $vds.ExtensionData.Config.ConfigVersion
$vds.ExtensionData.ReconfigureDvs_Task($spec)

 

Update 08/07/2020

In case you are using this approach for nested vSphere lab instead of the old promiscuous mode, make sure the vmk0 vmkernel port has a different MAC address than the MAC address of the vmnic of the nested ESXi host. This is because when the vmk0 is migrated to a different ESXi host uplink the vDS will not learn the MAC address on the new switch port as it conflicts with the assigned MAC on the first uplink port (same MAC cannot be learnt on two ports).

The vmkernel port MAC can be easily changed by editing /etc/vmware/esx.conf file.