Zero to SDN in under five minutes

Deploy a cloud application quickly with the new Microsoft SDN stack

You might have seen this blog post recently published on common data center challenges. In that article, Ravi talked about the challenges surrounding deployment, flexibility, resiliency, and security, and how our Software Defined Networking (SDN) helps you solves those challenges.

In this blog post series we will go deeper so you’ll know how you can use Microsoft SDN with Hyper-V to deploy a classic application network topology. Think about how long it takes you to deploy a three-tier web application in your current infrastructure. Ok, do you have a figure for it? How long, and how many other people did you need to contact?

This series focuses on a deployment for a lab or POC environment. If you decide to follow along with your own lab setup you’ll interact with the Microsoft network controller, build an overlay software defined network, define security policy, and work with the Software Load Balancer.

  • In Part 1 you’ll be introduced to the SDN stack and the three-tier workload app
  • In Part 2 you’ll learn about the front end web tier and tenant configuration
  • In Part 3 you’ll get into the application tier and the back end data tier

Here’s what you’ll need in your lab environment:

The first step in deploying the cloud application is to install and configure the servers and infrastructure. You will need to install Windows Server 2016 Technical Preview 4 on a minimum of three physical servers. Use the Planning Software Defined Networking TechNet article for guidance in configuring the underlay networking and host networking. The environment I used while writing this post and deploying the three-tier app has the following configuration:

  • Three physical servers each with Dual 2.26 GHz CPUs, 24 GB of memory, 550 GB of storage, two 10Gb Ethernet cards
  • Each host uses a Hyper-V Virtual Switch in a Switch-Embedded Team configuration
  • All hosts are connected to an Active Directory domain named “SDNCloud”
  • Each server is attached to a Management VLAN, and the default gateway is an SVI on a switch
  • The upstream physical switch is configured with the same VLAN tags as the Hyper-V virtual switch, and uses trunk mode so that management and host network traffic can share the same switch ports

Introduction

Enterprise and hosting providers use their IT tool kits to address similar and reoccurring problems:

  • Deploy new services quickly with enough flexibility to accommodate incremental demand for capacity and performance
  • Maintain availability despite multiple failure modes
  • Ensure security

Windows Server 2016 helps you address these challenges in the application platform itself, and for the networking technology we’ll cover in this blog it’s the same technology that services the 1.5 million+ network requests per second average in Microsoft Azure.

The scenario for the series is for a new product that our fictitious firm “Fabrikam” is launching to meet the demands for convenience and self-service in requesting a new passport, renewing an expired passport, or updating a citizen’s personal information. The application is called “Passport Expeditor” and it removes the need for a citizen to go to the passport agency and execute a paper-based process that uses awkward government-speak.

Passport Expeditor is based on a three-tier architecture, which consists of a front-end web tier to present the interface to the user, the application tier to validate inputs and contains the application logic, and a back-end database tier to store passport information. The software in each tier runs in the in a virtual machine, and is connected to one or more networks with associated security policies.

Figure 1: Passport Expeditor application architecture

External users will access Fabrikam’s Passport Expeditor cloud application through a hostname registered to an IP address that is routable on the public Internet. In order to handle the thousands of requests Fabrikam expects to see at launch, load balancing services are required and will be provided in the network fabric using Microsoft’s in-box Server Load Balancer (SLB). The SLB will distribute incoming TCP connections among the web-tier nodes providing both performance and resiliency. To do this the SLB will monitor the health probes installed on each VM and take any VMs which are “down” out of rotation until they become healthy again. The SLB can also increase the number of  VMs servicing the application during periods of peak load and then scale back down when load decreases.

Core concepts

Before we dive in, let’s spend a moment talking about some core concepts and technologies we will be using:

  • PowerShell scripting: We will use PowerShell scripts to create the network policy and resources and use the HTTP verbs PUT and GET to inform the Network Controller of this policy
  • Network Controller: The Microsoft Network Controller is the “brains” of the SDN Stack. Network policy is defined through a set of resources modeled using JSON objects and given to the Network Controller through a RESTful API. The Network Controller will then send this policy to the SDN Host Agent running on each Hyper-V Host (server)
  • SDN Host Agent: The Network Controller communicates directly with the SDN Host Agent running on each server. The Host Agent then programs this policy in the Hyper-V Virtual switch.
  • Hyper-V Virtual Switch: Microsoft’s vSwitch is responsible for enforcing network policy (such as VXLAN based overlay networks, access control lists, address translation rules, etc) provisioned by the network controller.
  • Software Load Balancer: The SLB consists of a multiplexer which advertises a Virtual IP (VIP) address to external clients (using BGP) and distributes connections across a set of Dynamic IP (DIP) addresses assigned to VMs attached to a network.
  • North/South and East/West traffic: These terms refer to where network traffic originates and is destined. North/South indicates the network traffic is going outside of the virtual network or data center. East/West indicates the network traffic is coming from inside the virtual network or within the data center.


Figure 2: Windows Server 2016 SDN stack

Perform the following steps to configure Windows Server Technical Preview 4 for the scenario:

  1. Install the operating system on the physical server
  2. Enable the Hyper-V Role on each host
  3. Create a Hyper-V Virtual Switch on each host. Be sure to use the same name for each virtual switch on each host and bind it to a network interface
  4. Ensure the virtual switch’s Management virtual network interface (vNIC) is connected to the Management VLAN and has an IP address assigned to it
  5. Verify connectivity via the Management IP address between all servers
  6. Join each host to an active directory domain

The system is now ready to receive the SDN Stack components, software infrastructure, and inform the Network Controller about the fabric resources. If you haven’t already retrieved the scripts from GitHub, download them now. All scripts are available on the Microsoft SDN GitHub repository and can be downloaded as a zip file from the link referenced (for more details on Git, please reference this link).

Fabric resource deployment

The Network Controller must be informed of the environment it is responsible for managing by specifying the set of servers, VLANs, and service credentials. These fabric resources will also be the endpoints on which the controller enforces network policies. The resource hierarchy and dependency graph for these fabric resources is shown in Figure 3.

Figure 3: Network controller northbound API fabric resource hierarchy

The variables in the FabricConfig.psd1 file (see sample here) configuration file must be populated with the correct values to match your environment. Insert the appropriate configuration parameter anywhere you see the mark “<<Replace>>”. You will do this for credentials, VLANs, Border Gateway Protocol Autonomous System Numbers (BGP ASNs) and peers, and locations for SDN service VMs.

When customizing the FabricConfig.psd1 file:

  • Ensure the directory specified by the InstallSrcDir variable is shared with Everyone and has Read/Write access.
  • The value of the NetworkControllerRestName variable must be registered in DNS with the value of the floating IP address of the Network Controller specified by the NetworkControllerRestIP parameter.
  • The value of the vSwitchName variable must be the same for all Hyper-V Virtual Switches in each server.
  • The LogicalNetworksarraycontains the fabric resources which correspond to specific VLANs and IP prefixes in the underlay network. In this post series, we will only be configuring and using:
    • Hyper-V Network Virtualization Provider Address (HNVPA): Used as the underlay network for hosting HNV overlay virtual networks.
    • Management: Used for communication between Network Controller and Hyper-V Hosts (and SDN Host Agent)
    • Virtual IP (VIP): Used as the public (routable) IP prefix through which external users will access the HNV overlay virtual network (e.g. Web-Tier). Routes to the VIPs will be advertised using internal BGP peering between the SLB Multiplexer and BGP Router.
    • The Transit and GREVIP networks are used by the Gateways (not covered in this post series). In the future, the SLB Multiplexer will also connect to the Transit logical network.
  • The Hyper-V host section is an array of NodeNameswhich must correspond to the physical hosts registered in DNS. This section determines where to place the infrastructure VMs (Network Controller, SLB Multiplexer, etc.).
    • The IP Addresses for the Network Controller VMs (e.g. NC-01) must come from the Management logical network’s IP prefix.
    • The IP Addresses for the Software Load Balancer VMs (e.g. MUX-01) must come from the HNVPA logical network’s IP prefix.
  • The Management and HNVPA logical network prefixes must be routable between each other.

Figure 4: Deployment environment

After customizing this file and running the SDNExpress.ps1 script documented in the TechNet article, validate your configuration by testing that the requisite fabric resources, e.g. logical networks, servers, and SLB Multiplexer, are correctly provisioned in the Network Controller by following the steps in the TechNet article. As a first step, you should be able to ping the Network Controller (NetworkControllerRestIP) from any host. You should also verify that you can query resources on the Network Controller using the REST Wrappers Get-NC<ResourceName> (e.g. PS > Get-NCServer) and validate that the output includes provisioningState = succeeded.

Note: The deployment script creates multi-tenant Gateway VMs. These will not be used in this blog series.


Figure 5: Network controller provisioning 

The final check is to ensure that the load balancers are successfully peering with the BGP router (either a VM with Routing and Remote Access Server (RRAS) role installed or Top of Rack (ToR) Switch. Border Gateway Protocol (BGP) is used by SLB to advertise the VIP addresses to external clients and then route the client requests to the correct SLB Multiplexer. In my lab I am using the BGP router in the ToR and the switch validation output is shown below:

Figure 6: Successful BGP peering

Summary and validation

In this blog post, we introduced the Passport Expeditor service which can be installed as a cloud application using the new Microsoft Software Defined Networking (SDN) Stack. We walked through the host and network pre-requisites and deployed the underlying SDN infrastructure, including the Network Controller and SLB. The fabric resources deployed will be used as the basis to instantiate and deploy the tenant resources in part II of this blog series. The Network Controller REST Wrapper scripts can be used to query the fabric resources as shown in the TechNet article here.

In the next blog post: Front-end Web Tier Deployment and Tenant Resources

The SDN fabric is now ready to instantiate and deploy tenant resources. In the next part in this blog series, we will be creating the following tenant resources for the front-end web tier of the three-tier application shown in Figure 1 above:

  1. Access Control Lists
  2. Virtual Subnets
  3. VM Network Interfaces
  4. Public IP

We’d love to hear from you. Please let us know if you have any questions in the comments!