Horizon 8 Architecture
This chapter is one of a series that make up the , a framework that provides guidance on the architecture, design considerations, and deployment of Workspace ONE and Horizon solutions. This chapter provides information about architecting VMware Horizon 8. A companion chapter, , provides information about common configuration and deployment tasks for VMware Horizon.
is a platform for managing and delivering virtualized or hosted desktops and applications to end users. Horizon allows you to create and broker connections to Windows virtual desktops, Linux virtual desktops, Remote Desktop Server (RDS)–hosted applications and desktops, Linux-hosted applications, and Windows physical machines.
This chapter of the covers the architecture and design considerations for Horizon 8 for vSphere. Horizon 8 can be deployed on-premises or on other supported cloud platforms. This chapter covers the foundational and common architectural information for deploying Horizon and is applicable across all supported platforms. Separate chapters give the additional design considerations for Horizon on supported cloud platforms, including , , , , and .
Table 1: Horizon Environment Setup Strategy
A Horizon deployment was designed, deployed, and integrated with the VMware Workspace ONE platform.
The environment was designed to be capable of scaling to 8,000 concurrent connections for users.
This strategy allowed the design, deployment, and integration to be validated and documented.
The core components of Horizon include a VMware Horizon® Client™ authenticating to a Connection Server, which brokers connections to virtual desktops and apps. The Horizon Client then forms a protocol session connection to a Horizon Agent running in a virtual desktop, RDSH server, or physical machine. The protocol session can also be configured to be tunneled via the Connection Server, although this is not generally recommended as it makes the ongoing session dependent on the Connection Server.
Figure 1: Horizon Core Components
External access includes the use of VMware Unified Access Gateway™ to provide secure edge services. The Horizon Client authenticates to a Connection Server through the Unified Access Gateway. The Horizon Client then forms a protocol session connection, through the gateway service on the Unified Access Gateway, to a Horizon Agent running in a virtual desktop or RDSH server. This process is covered in more detail in .
Figure 2: Horizon Core Components for External Access
The following figure shows the high-level logical architecture of the Horizon components, with other Horizon components shown for illustrative purposes.
Figure 3: Horizon Logical Components
The components and features of Horizon are described in the following table.
Table 2: Components of Horizon
From a data center perspective, several components and servers must be deployed to create a functioning Horizon environment to deliver the desired services.
Figure 4: Horizon Logical Architecture
In addition to the core components and features, other products can be used in a Horizon deployment to enhance and optimize the overall solution:
- Workspace ONE Access – Provides enterprise single sign-on (SSO), securing and simplifying access to apps with the included identity provider or by integrating with existing identity providers. It provides application provisioning, a self-service catalog, conditional access controls, and SSO for SaaS, web, cloud, and native mobile applications. See for design and implementation details.
- App Volumes Manager – Orchestrates application delivery by managing assignments of application volumes (packages and writable volumes) to users, groups, and target computers. See for design and implementation details.
- Dynamic Environment Manager – Provides profile management by capturing user settings for the operating system and applications. See for design and implementation details.
- VMware vSAN storage – Delivers high-performance, flash-optimized, hyper-converged storage using server-attached flash devices or hard disks to provide a flash-optimized, highly resilient, shared datastore.
- VMware NSX-T Data Center – Provides network-based services such as security, virtualized networking, routing, and switching in a single platform. With micro-segmentation, you can set application-level security policies based on groupings of individual workloads, and you can isolate each virtual desktop from all other desktops as well as protect the Horizon management servers.
Note: VMware NSX-T Data Center is licensed separately from Horizon.
Pod and Block
One key concept in a Horizon environment design is the use of pods and blocks, which gives us a repeatable and scalable approach.
A pod is made up of a group of interconnected Connection Servers that broker connections to desktops or published applications.
- A pod can broker up to 20,000 sessions, including desktop and RDSH sessions.
- Multiple pods can be interconnected by either using the or by using (CPA).
- A single Cloud Pod Architecture can scale to a maximum of 250,000 sessions. For numbers above that, separate CPAs can be deployed.
The resource capacity for a pod is provided by one or more resource blocks. Each block is made up of one or more resource vSphere clusters, and each block has its own vCenter Server. The number of virtual machines (VMs) a block can typically host depends on the type of Horizon VMs used. See for details. Resource blocks can be either in the same location as the Connection Servers or in a different location, using the deployment model.
Figure 5: Horizon Pod and Block Design
To add more resource capacity, we simply add more resource blocks. We also add more Connection Servers to add the capability for more session connections within the pod.
Depending on the types of VMs (instant clones, full clones, and if using App Volumes), a resource block could host a different number of VMs (see ). Typically, we have multiple resource blocks and up to seven Connection Servers in a pod capable of hosting up to 20,000 sessions. For numbers above that, we deploy additional pods.
As you can see, this approach allows us to design a single block capable of thousands of sessions that can then be repeated to create a pod capable of handling 20,000 sessions. Multiple pods grouped using either the Universal Broker or Cloud Pod Architecture can then be used to scale the environment as large as needed.
Important: All Connection Servers in a single pod must be located within a single data center and cannot span locations.
Options regarding the location of management components, such as Connection Servers, include:
- Co-located on the same vSphere hosts as the desktops and RDSH servers that will serve end-users
- On a separate vSphere cluster
- On separate cloud compute resources, in line with the recommendations of the given cloud platform
In large environments, for scalability and operational efficiency, it is normally best practice to have a separate vSphere cluster to host the management components. This keeps the VMs that run services such as Connection Server, Unified Access Gateway, vCenter Server, and databases separate from the desktop and RDSH server VMs.
Management components can be co-hosted on the same vSphere cluster as the end-user resources if desired. This architecture is more typical in smaller environments or where the use of converged hardware is used and the cost of providing dedicated hosts for management is too high. If you place everything on the same vSphere cluster, you must configure the setup to ensure resource prioritization for the management components. Sizing of resources (for example, virtual desktops) must also take into account the overhead of the management servers. See for more information.
Table 3: Pod and Block Design for this Reference Architecture
A pod was formed in each site.
Each pod contained one or more resource blocks.
Deploying multiple, separate pods in different sites provides site redundancy and allows an equivalent service to be delivered to the user from an alternate location.
This allowed the design and deployment of the block and pod architecture to be validated and documented.
Cloud Pod Architecture
As an alternative to , and when multiple Horizon pods are being deployed, you can use Cloud Pod Architecture (CPA). This allows multiple Horizon pods to be joined together into a federation and then global entitlements (GE) to be assigned that include entitlements from multiple pods. Participating pods can be located in the same physical site or location or in different sites and locations.
This feature allows you to provide users and groups with a global entitlement that can contain desktop pools or RDSH-published applications from multiple different pods that are members of this federation construct.
The following figure shows a logical overview of a basic two-site CPA implementation.
Figure 6: Cloud Pod Architecture
Important: This type of deployment is not a stretched deployment. Each pod is distinct, and all Connection Servers belong to a specific pod and are required to reside in a single location and run on the same broadcast domain from a network perspective.
As well as being able to have desktop pool members or published applications from different pods in a global entitlement, this architecture allows for a property called scope. Scope allows us to define where new sessions should or could be placed and also allows users to connect to existing sessions (that are in a disconnected state) when connecting to any of the pod members in the federation.
CPA can also be used within a site:
- To use global entitlements that span multiple resource blocks and pools
- To federate multiple pods on the same site, when scaling above the capabilities of a single pod
Table 4: Implementation Strategy for Using Cloud Pod Architecture
Cloud Pod Architecture was not used.
Universal Broker was used in preference to Cloud Pod Architecture.
Scalability and Availability
One key design principle is to remove single points of failure in the deployment. The numbers, limits, and recommendations given in this section were correct at the time of writing. For the most current numbers for Horizon 8, see the .
A single Connection Server supports a maximum of 4,000 sessions. This is reduced to 2,000 sessions when tunneling connections through the Connection Server by using the Blast Secure Gateway, PCoIP Secure Gateway, or the HTTP(s) Secure Tunnel on the Connection Server.
Up to seven Connection Servers are supported per pod, with up to 20,000 active sessions in total per pod.
To satisfy the requirements that the proposed solution be robust and able to handle failure, deploy one more server than is required for the number of connections (n+1).
Important: The Connection Servers in a single pod cannot span physical locations. They must be located within a single data center and run on a well-connected LAN segment.
Table 5: Strategy for Deploying Connection Servers
Three Horizon Connection Servers were deployed per pod.
These ran on dedicated Windows Server 2022 VMs located in the internal network.
One Connection Server supports up to 4,000 connections.
Sessions are not tunneled through the Connection Servers and the secure gateways and secure tunnel are disabled on the Connection Servers.
Two Connection Servers are required to handle the load of the target 8,000 users for each pod.
A third server provides redundancy and availability (n+1).
Load Balancing Connection Servers
For high availability and scalability, VMware recommends that multiple Connection Servers be deployed in a load-balanced replication cluster. This ensures that user load is evenly distributed across all available servers and a single namespace can be used by end users. Using a load balancer also facilitates greater flexibility by enabling IT administrators to perform maintenance, upgrades, and configuration changes while minimizing impact to users.
Connection Servers broker client connections, authenticate users, and direct incoming requests to the correct agent resource. Although the Connection Server helps form the connection for authentication, it typically does not act as part of the data path after a protocol session has been established.
The load balancer serves as a central aggregation point for traffic flow between clients and Connection Servers, sending clients to the best-performing and most available Connection Server instance. Using a load balancer with multiple Connection Servers also facilitates greater flexibility by enabling IT administrators to perform maintenance, upgrades, and changes in the configuration without impacting users. To ensure that the load balancer itself does not become a point of failure, most load balancers allow for the setup of multiple nodes in an HA or active/passive configuration.
Figure 7: Connection Server Load Balancing
Connection Servers require the load balancer to have a session persistence setting. This is sometimes referred to as persistent connections or sticky connections, and ensures data stays directed to the relevant Connection Server. For more information, see the VMware Knowledge Base article .
Table 6: Strategy for Using Load Balancers with Connection Servers
An NSX Advanced load balancer was used in front of the Connection Servers for internal connections.
Source IP was configured for the persistence or affinity type.
This provides an internal common namespace for the Connection Servers, which allows for ease of use, scale, and redundancy.
vCenter Server is the delimiter of a resource block.
The recommended number of VMs that a vCenter Server can typically host depends on the type of Horizon VMs used. The following limits have been tested.
- 20,000 instant-clone VMs
- 4,000 full-clone VMs
Just because VMware publishes these configuration maximums does not mean you should necessarily design to that limit. Using a single vCenter Server does introduce a single point of failure that could affect too large a percentage of the VMs in your environment. Therefore, carefully consider the size of the failure domain and the impact should a vCenter Server become unavailable.
A single vCenter Server might be capable of supporting your whole environment, but to reduce risk and minimize the impact of an outage, you will probably want to include more than one vCenter Server in your design. You can increase the availability of vCenter Server by using VMware vSphere® High Availability (HA), which restarts the vCenter Server VM in the case of a vSphere host outage. can also be used to provide an active-passive deployment of vCenter Server appliances, although caution should be used to weigh the benefits against the added complexity of management.
Sizing can also have performance implications because a single vCenter Server could become a bottleneck if too many provisioning tasks run at the same time. Do not just size for normal operations but also understand the impact of provisioning tasks and their frequency.
For example, consider a use case with non-persistent, parent-based, instant-clone desktops, which are deleted after a user logs off and are provisioned when replacements are required. Although a non-persistent, floating desktop pool can be pre-populated with spare desktops, it is important to understand how often replacement VMs would need to be generated and when that happens. Are user logoffs and the demand for new desktops spread throughout the day? Or are desktop deletion and replacement operations clustered at certain times of day? If these events are clustered, can the number of spare desktops satisfy the demand, or do replacements need to be provisioned? How long does provisioning desktops take, and is there a potential delay for users?
Understanding provisioning tasks, like this, helps understand the demand placed on the vCenter Server and whether it is better to scale out rather than scale up.
Table 7: Implementation Strategy for vCenter Server
Two resource blocks were deployed per site, each with their own vCenter Server virtual appliance, located in the internal network.
A single resource block and a single vCenter Server are supported for the intended target of 8,000 instant-clone VMs; however, having a single vCenter Server for the entire user environment presents too large a failure domain.
Splitting the environment across two resource blocks, and therefore over two vCenter Servers reduces the impact of any potential outage.
This approach also allows each resource block to scale to a higher number of VMs and allow for growth, up to the pod recommendation, without requiring us to rearchitect the resource blocks.
Secure, external access for users accessing resources is provided through the integration of Unified Access Gateway (UAG) appliances. We also use load balancers to provide scalability and allow for redundancy. A Unified Access Gateway appliance can be used in front of Connection Servers to provide access to on-premises Horizon desktops and published applications.
External Access Architecture
When using Unified Access Gateway to provide external access to Horizon, the same Connection Servers can be used for both external and internal connections.
With the preferred architecture for traffic flow and load balancing of Unified Access Gateways and Connection Servers, a load balancer is not placed in-line between the Unified Access Gateways and the Connection Servers. The architecture simplifies the design and makes it easier to troubleshoot.
Figure 8: Unified Access Gateway and Connection Server Architecture
While the previous diagram shows a 1-to-1 mapping of Unified Access Gateway to the Connection Server, it is also possible to have a N-to-1 mapping, where more than one Unified Access Gateway maps to the same Connection Server.
High availability is maintained by having at least two Unified Access Gateways and ensuring that they specify different Connection Servers as targets for the Connection Server URL. A Unified Access Gateway can detect if its target Connection Server is not responding and will notify the external load balancer that it can no longer handle any new connections.
Note: It is still a valid architecture and supported to have a load balancer inline between the Unified Access Gateways and the Connection Servers. However, when a load balancer is placed between the two, the Unified Access Gateway cannot detect if an individual Connection Server is down.
When required for internally routed connections, a load balancer for the Connection Servers can be either:
- Located so that only internal users use it as an FQDN.
- Placed in between the Unified Access Gateways and the Connection Server and used as an FQDN target for both internal users and the Unified Access Gateways.
Figure 9: Options for Load Balancing Connection Servers for Internal Connections
For on-premises deployment of Horizon within a data center of an organization, it is common to install Unified Access Gateway appliances in a single DMZ, which provides a network isolation layer between the Internet and the customer data center.
Figure 10: Single DMZ Deployment showing Blast Extreme ports
Unified Access Gateway has built-in security mechanisms for all the Horizon protocols to ensure that the only network traffic entering the data center is traffic on behalf of an authenticated user. Any unauthenticated traffic is discarded in the DMZ.
Some organizations have two DMZs (often called a double DMZ or a double-hop DMZ) that are sometimes used to provide an extra layer of security protection between the Internet and the internal network. In a double DMZ, traffic has to be passed through a specific reverse proxy in each DMZ layer. Traffic cannot simply bypass a DMZ layer.
Note that in a Horizon deployment, a double DMZ is not required, but for environments where a double DMZ is mandated, an extra Unified Access Gateway appliance acting as a Web Reverse Proxy can be deployed in the outer DMZ.
Figure 11: Double DMZ Deployment showing Blast Extreme ports
Unified Access Gateway Scaling
When deployed for Horizon Edge services a standard size Unified Access Gateway appliance is sized for up to 2,000 sessions. When designing Unified Access Gateway appliance numbers, a balance with the number of Connection Servers should be considered to ensure that overall availability is maintained in the event of either server component.
Figure 12: Scaling Out Unified Access Gateways
Table 8: Implementation Strategy for External Access
Six standard-size Unified Access Gateway appliances were deployed as part of the Horizon solution.
These were located in the DMZ network.
UAG provides secure external access to internally hosted Horizon desktops and applications.
One standard-size UAG appliance is recommended for up to 2,000 concurrent Horizon connections.
Four UAG appliances are required to handle the load of the target 8,000 users.
We elected not to place a load balancer in-line between the UAGs and the Connection Server, so each UAG targets a particular Connection Server. To match up with the requirement for three Connection Servers, we should look to have a 2-to-1 ratio of UAG to Connection Server. This leads to an overall count of six UAGs, to ensure redundancy and availability.
Load Balancing Unified Access Gateway
It is strongly recommended that end users connect to Unified Access Gateway using a load-balanced virtual IP (VIP). This ensures that user load is evenly distributed across all available Unified Access Gateway appliances and facilitates greater flexibility by enabling IT administrators to perform maintenance, upgrades, and configuration changes while minimizing impact to users.
To implement them in a highly available configuration, provide a virtual IP address (VIP), and present users with a single namespace, the following options can be used.
- A third-party load balancer, such as the (Avi) can be used.
- Unified Access Gateway (HA) can be used.
When load balancing Horizon traffic to multiple Unified Access Gateway appliances, the initial XML-API connection (authentication, authorization, and session management) needs to be load balanced. The secondary Horizon protocols must be routed to the same Unified Access Gateway appliance to which the primary Horizon XML-API protocol was routed. This allows the Unified Access Gateway to authorize the secondary protocols based on the authenticated user session.
If the secondary protocol session is misrouted to a different Unified Access Gateway appliance from the primary protocol one, the session will not be authorized. The connection would therefore be dropped in the DMZ, and the protocol connection would fail. Misrouting secondary protocol sessions is a common problem if the load balancer is not configured correctly.
The load balancer affinity must ensure that XML-API connections made for the whole duration of a session (with a default maximum of 10 hours) continue to be routed to the same Unified Access Gateway appliance.
With the use case of providing secure, external access to resources, there is no need to provide a single namespace to the Horizon Connection Servers because only external users will be connecting. This means that there is no need to provide a load balancer VIP in front of the Connection Servers.
Although the secondary protocol session must be routed to the same Unified Access Gateway appliance as was used for the primary XML-API connection, there is a choice of whether the secondary protocol session is routed through the load balancer or not. This normally depends on the capabilities of the load balancer.
For more detail on load balancing of Unified Access Gateway appliances, see:
Table 9: Strategy for Using Load Balancers with Connection Servers
For the on-premises deployments of Horizon, an NSX Advanced load balancer was used in front of the Unified Access Gateways.
Source IP was configured for the persistence or affinity type.
This provides a common namespace for the Unified Access Gateways, which allows for ease of use, scale, and redundancy.
Unified Access Gateway High Availability
As an alternative to using a third-party load balancer, Unified Access Gateway provides, out-of-the-box, a high-availability solution for the Unified Access Gateway edge services. The solution supports up to 10,000 concurrent connections in a high-availability (HA) cluster and simplifies HA deployment and configuration of the services.
For more information on the Unified Access Gateway High Availability component and configuration of edge services in HA, see the following resources:
To ensure correct communication between the components, it is important to understand the network port requirements for connectivity in a Horizon deployment. The has more detail and includes diagrams illustrating the traffic. It has specific sections and diagrams on internal, external, and tunneled connections.
- The network ports shown are destination ports.
- The arrows indicate the direction of traffic initiation (source to destination).
- Horizon UDP protocols are bidirectional, so stateful firewalls should be configured to accept UDP reply datagrams.
Horizon is a multi-protocol solution. Three remoting protocols are available when creating desktop pools or RDSH-published applications: Blast Extreme, PCoIP, and RDP.
Table 10: Display Protocol for Virtual Desktops and RDSH-Published Apps
For this design, we leveraged Blast Extreme.
This display protocol supports multiple codecs, both TCP and UDP from a transport protocol perspective, and the ability to do hardware encoding with NVIDIA GRID vGPU.
Blast Extreme is configured through Horizon when creating a pool. The display protocol can also be selected directly on the Horizon Client side when a user selects a desktop pool.
See the following guides for more information, including optimization tips:
With internal connections, the Horizon Client authenticates to a Connection Server, which brokers connections to virtual desktops and apps. The Horizon Client then forms a protocol session connection to a Horizon Agent running in a virtual desktop, RDSH server, or physical machine.
The following diagram shows the ports required to allow an internal Blast Extreme connection.
Figure 13: Internal Connection with Blast Extreme Network Ports
The following diagram shows the ports required to allow an internal PCoIP connection.
Figure 14: Internal Connection with PCoIP Network Ports
The following diagram shows the ports required to allow an internal RDP connection.
Figure 15: Internal Connection with RDP Network Ports
External connections include the use of VMware Unified Access Gateway to provide secure edge services. The Horizon Client authenticates to a Connection Server through the Unified Access Gateway. The Horizon Client then forms a protocol session connection, through the secure gateway service on the Unified Access Gateway, to a Horizon Agent running in a virtual desktop or RDSH server.
The following diagram shows the ports required to allow an external Blast Extreme connection.
Figure 16: External Connection with Blast Extreme Network Ports
The following diagram shows the ports required to allow an external PCoIP connection.
Figure 17: External Connection with PCoIP Network Ports
The following diagram shows the ports required to allow an external RDP connection.
Figure 18: External Connection with RDP Network Ports
Horizon Cloud Service
There are currently two versions of Horizon Cloud Service. To enable the use of subscription licensing, and other Horizon Cloud Services, requires the Horizon pod to be connected to the appropriate Horizon Cloud Service. If you are planning to deploy or connect Horizon 8 for the first time, Horizon Cloud Service – next-gen is recommended. However, confirm that the services and features you want to use are currently supported to work with Horizon 8.
Horizon Cloud Service – next-gen
- Can be used to enable subscription-based licensing for Horizon 8 environments.
- Provides monitoring of Horizon 8 environments. It is planned to offer other services in the future.
- A is used to connect a Horizon 8 pod to Horizon Cloud Service – next-gen.
- For more information on Horizon Cloud Service – next-gen and Next-gen Horizon Control Plane Services, see .
Horizon Cloud Service – first-gen
- Can be used to enable subscription-based licensing for Horizon 8 environments.
- Provides other first-gen Horizon Cloud Services for Horizon 8 environments. For more information on the services available from Horizon Cloud Service – first-gen, see .
- A appliance is used to connect a Horizon 8 pod to Horizon Cloud Service – first-gen.
Depending on which version of the Horizon Cloud Service you intend to connect a Horizon pod to, you will need to deploy the corresponding connector appliance. A separate connector appliance is required for each individual Horizon pod. A Horizon pod should only be connected to either next-gen or first-gen Horizon Cloud Service, and not both at the same time.
Horizon Edge Gateway Appliance
The Horizon Edge Gateway is a virtual appliance that is used to connect a Horizon 8 pod with Horizon Cloud Service – next-gen services and features. A separate Horizon Edge Gateway is required for each Horizon pod that you connect to Horizon Cloud Service - next-gen.
Figure 19: Horizon 8 Pods Connected to Horizon Cloud Service – next-gen with Horizon Edge Gateway Appliances
Providing availability for Horizon Edge Gateway appliances.
- vSphere High Availability (HA) can be used to restart Horizon Edge Gateway appliances in the event of a vSphere host failure. For more information, see .
- To provide continuous availability, vSphere Fault Tolerance can be used with Horizon Edge Gateway appliances. For more information, see .
The documentation also has several resources to assist in the Horizon Edge Gateway appliance deployment:
Horizon Cloud Connector
The Horizon Cloud Connector is a virtual appliance that is used to connect a Horizon 8 pod with Horizon Cloud Service – first-gen services and features. The Horizon Cloud Connector is deployed from the VMware vSphere® Web Client and paired to one of the Connection Servers in the pod. This enables the use of control plane services from Horizon Cloud Service – first-gen, including licensing, Universal Broker, Image Management, Help Desk, and Cloud Monitoring.
As part of the pairing process, the Horizon Cloud Connector virtual appliance connects the Connection Servers to the Horizon Cloud Service – first-gen to manage the subscription license. With a subscription license for Horizon, you do not need to retrieve or manually enter a license key for Horizon product activation. However, license keys are still required for supporting the components, which include vSphere, vSAN, and vCenter Server. These keys are emailed to the VMware Customer Connect contact.
You must have an active VMware Customer Connect account to purchase a Horizon license from . You then receive a subscription email with the link to download the Horizon Cloud Connector as an OVA (Open Virtual Appliance) file.
One active, or primary, Horizon Cloud Connector VM is supported per pod.
To support service-level fault tolerance you can create a two-node Horizon Cloud Connector cluster by adding a worker node to the cluster containing the primary node. The worker node contains a replica of the Horizon Cloud Connector application services. For more information on Horizon Cloud Connector Clusters and to understand which services can currently be protected, see .
High availability of each Cloud Connector node is also provided by vSphere HA, which restarts the Cloud Connector VM in the case of a vSphere host outage.
Horizon Universal Broker
The Horizon Universal Broker is a feature of Horizon Cloud Service – first-gen, which allows the brokering of desktops and applications to end users across all cloud-connected Horizon pods. For more information, see in the First-Gen Horizon Control Plane Architecture chapter.
Note: At the time of writing, the brokering service from Horizon Cloud Service – next-gen is not supported with Horizon 8 pods and resources.
- A Horizon Cloud Connector is required for each Horizon 8 pod that will use the Horizon Cloud Service -first-gen and its features, which include the Horizon Universal Broker.
- The Universal Broker plug-In should be installed on every Connection Server in each participating pod, as described in .
Figure 20: Horizon Universal Broker Architecture and Components
See the following documents for the latest on limitations when using the Universal Broker:
Instant Clone Smart Provisioning
An automated instant clone pool or farm is created from a golden image VM using the vSphere instant clone API. Instant clone technology replaces View Composer linked clone as the process for creating automated farms in Horizon.
Horizon creates several types of internal VMs (Internal Template, Replica VM, and Parent VM) to manage these clones in a more scalable way. Instant clones share the virtual disk of the replica VM and therefore consume less storage than full VMs. In addition, instant clones share the memory of the parent VM when they are first created, which contributes to fast provisioning. After the instant clone VM is provisioned and the machine starts to be used, additional memory is utilized. After it is provisioned, the clone is no longer linked to the parent VM.
Figure 21: Instant Clones with Parent VMs
Although helpful in speeding up the provisioning speed, the use of a parent VM does increase the memory requirement across the cluster. When there is a low density of VM clones per host, either with small desktop pools or with RDS Host farms, the benefit of having more memory outweighs the increase in provisioning speed. In this case, Horizon can automatically choose to provision instant clones directly from the replica VM, without creating any parent VM.
- No running parent VMs are created on the vSphere hosts.
- Each clone has a vSphere snapshot taken, after cloning from the replica, which it is reverted to at logoff.
Figure 22: Instant Clones without Parent VMs
A single instant clone pool or farm can have both instant clones that are created with parent VMs (mode A) or without parent VMs (mode B). The default behavior can be overridden at a pool or farm level. See .
There are a variety of methods for authenticating users in Horizon to control access to desktops and published applications.
Workspace ONE Access Authentication
One of the methods of accessing Horizon desktops and applications is through Workspace ONE Access. This requires integration between Connection Servers and Workspace ONE Access using the SAML 2.0 standard to establish mutual trust, which is essential for single sign-on (SSO) functionality.
When SSO is enabled, users who log in to Workspace ONE Access with Active Directory credentials can launch remote desktops and applications without having to go through a second login procedure. If you set up the True SSO feature, users can log in using authentication mechanisms other than AD credentials. See and see .
When defining the SAML authenticator on the Connection Servers, you can choose to set the delegation of authentication to allowed or required. Allowed makes this optional, whereas required enforces the use of the SAML authentication source. When configure as required you can also enable Workspace ONE which redirects any direct authentication attempts into Workspace ONE Access. See .
Table 11: Strategy for Authenticating Users Through Workspace ONE Access
A SAML authenticator for Workspace ONE was configured to be required on the Connection Servers.
Workspace ONE mode was enabled on the Connection Servers.
With this configuration, Connection Servers allow Workspace ONE Access to be a dynamic SAML authenticator. This strategy facilitates the launch of Horizon resources only from Workspace ONE Access and redirects any attempts to authenticate directly to Horizon back to Workspace ONE Access.
Unified Access Gateway Authentication
Unified Access Gateway supports multiple authentication options; for example, pass-through, RSA SecurID, RADIUS, SAML, and certificates, including smart cards. Pass-through authentication forwards the request to the internal server or resource. Other authentication types enable authentication at the Unified Access Gateway, before passing authenticated traffic through to the internal resource.
You can also use SAML to authenticate Horizon users against a third-party identity provider (IdP), leveraging Unified Access Gateway as the service provider (SP).
Table 12: Strategy for Authenticating Users Through Unified Access Gateway
Unified Access Gateway was left with the default pass-through authentication and no additional authentication methods were implemented on Unified Access Gateway.
This configuration facilitates the launch of Horizon resources from Workspace ONE Access and will use that as the user authentication point.
Implementing further authentication on the Unified Access Gateway would force users to have to authenticate twice.
Many user authentication options are available for users logging into Horizon including using Workspace ONE Access or Unified Access Gateway. Active Directory credentials are only one of these many authentication options. Ordinarily, using anything other than AD credentials would prevent a user from being able to single-sign-on to a Horizon virtual desktop or published application. After selecting the desktop or published application from the catalog, the user would be prompted to authenticate again, this time with AD credentials.
True SSO provides users with SSO to Horizon desktops and applications regardless of the authentication mechanism used. If a user authenticates by using Active Directory credentials, the True SSO feature is not necessary, but you can configure True SSO to be used even in this case so that the AD credentials that the user provides are ignored and True SSO is used.
True SSO uses SAML, where Workspace ONE is the Identity Provider (IdP) and the Horizon Connection Server is the Service Provider (SP). True SSO generates unique, short-lived certificates to manage the login process.
Figure 23: True SSO Logical Architecture
Table 13: Implementation Strategy for SSO
True SSO was configured and enabled.
This feature allows SSO to Horizon resources when launched from Workspace ONE Access, even when the user does not authenticate with Active Directory credentials.
True SSO requires the Enrollment Server service to be installed using the Horizon installation media.
True SSO Components
For True SSO to function, several components must be installed and configured within the environment. This section discusses the design options and details the design decisions that satisfy the requirements.
The Enrollment Server is responsible for receiving certificate signing requests (CSRs) from the Connection Server. The enrolment server then passes the CSRs to the Microsoft Certificate Authority to sign using the relevant certificate template. The Enrollment Server is a lightweight service that can be installed on a dedicated Windows Server instance, or it can co-exist with the MS Certificate Authority service. It cannot be co-located on a Connection Server.
The components of Horizon True SSO are described in the following table.
Table 14: Components of Horizon True SSO
A server that delivers True SSO functionality by ensuring a user can single-sign-on to a Horizon resource when launched from Workspace ONE Access, regardless of the authentication method.
The Enrollment Server is responsible for receiving certificate signing requests from the Connection Server and then passing them to the Certificate Authority to sign.
True SSO requires Microsoft Certificate Authority services, which it uses to generate unique, short-lived certificates to manage the login process.
Active Directory Certificate Services (AD CS) role running on a Windows server.
Used for issuing short-lived certificates that are used as part of the SSO process.
Load Balancing of Enrollment Servers
Two Enrollment Servers were deployed in the environment, and the Connection Servers were configured to communicate with both deployed Enrollment Servers. The Enrollment Servers can be configured to communicate with two Certificate Authorities.
By default, the Enrollment Servers use an Active/Failover method of load balancing. It is recommended to change this to round-robin when configuring two Enrollment Servers per pod to achieve high availability.
Table 15: Strategy for Load Balancing Between the Enrollment Servers
The Connection Server was configured to load balance requests using round robin between the two Enrollment Servers.
With two Enrollment Servers per pod, this is the recommendation when designing for availability.
vSphere HA and VMware vSphere® Storage DRS™ can be used to ensure the maximum availability of the Enrollment Servers. DRS rules are configured to ensure that the devices do not reside on the same vSphere host.
True SSO Scalability
A single Enrollment Server can handle all the requests from a single Horizon pod. The constraining factor is usually the Certificate Authority (CA). A single CA can generate approximately 35 certificates per second.
To ensure availability, a second Enrollment Server should be deployed per pod (n+1). Additionally, ensure that the certificate authority service is deployed in a highly available manner, to ensure complete solution redundancy.
Figure 24: True SSO High Availability
With two Enrollment Servers, and to achieve high availability, it is recommended to:
- Co-host the Enrollment Server service with a Certificate Authority service on the same machine.
- Configure the Enrollment Server to prefer to use the local Certificate Authority service.
- Configure the Connection Servers to load-balance requests between the two Enrollment Servers.
Table 16: Implementation Strategy for Enrollment Servers
Two Enrollment Servers were deployed per Pod.
These ran on dedicated Windows Server 2022 VMs located in the internal network.
These servers also had the Microsoft Certificate Authority service installed.
One Enrollment Server is capable of supporting a pod of 20,000 sessions.
A second server provides availability (n+1).
Figure 25: True SSO High Availability Co-located
Active Directory Domains
True SSO is supported in both single-domain and in multi-domain environments. With multiple domains, two-way trusts should be in place between the domains.
Looking at an example with two active directory domain trees (A & X) that are in the same active directory forest. Each of the domain trees has transitive trusts between all domains in that tree. Additionally, domain A tree and domain X tree have a two-way, transitive trust relationship between each other.
True SSO is supported in this scenario, and the Enrollment Servers can be placed in any domain.
Figure 26: Two Domain Trees in the Same Forest
Looking at another example with two active directory forests each containing its own active directory domain trees. In each of the forests, each of the domain trees has transitive trusts between all domains in the tree. Additionally, the two forests have a two-way, forest-level trust in place.
The Enrollment Servers can be placed within any domain of any forest.
Figure 27: Two Domain Trees in Separate Forests
Scaled Single-Site Architecture
The following sample diagram shows the server components and the logical architecture for a single-site deployment of Horizon. For clarity, the focus of this diagram is to illustrate the core Horizon server components, so it does not include additional and optional components such as App Volumes, Dynamic Environment Manager, and Workspace ONE Access.
Figure 28: Single-Site Scaled Horizon Pod
Horizon now supports the deployment of Horizon Agents in a remote location from the Horizon Connection Servers. This is useful when you need to consume capacity in another data center or cloud platform without having to stand up a full Horizon pod in that location.
The recommended scale for this feature is up to 1,000 VMs in the remote location. With larger numbers, although this feature would still work, the scale becomes large enough to warrant a separate Horizon pod deployed in the cloud platform.
Networking and routing are required between the existing Horizon environment and the location where capacity is going to be consumed and the remote agents installed. One technical constraint to be aware of is that the Horizon Agents (running in the virtual desktops and RDSH hosts) must be within 120 milliseconds of latency of the Horizon Connection Servers.
The two common use cases for this feature are:
- Centralize Horizon Pods for Private Datacenters – Minimize the number of Horizon pods across multiple locations and private data centers. This is useful when virtual desktops are required to be physically in multiple locations, such as branch offices. Normally this would require separate pods per location but with the remote agent architecture, the management infrastructure can be centralized reducing the number of pods that needs to be deployed.
- Extend an existing Horizon pod to consume capacity from a cloud platform – Existing Horizon pods can be extended to manage and consume capacity deployed in the cloud.
If Horizon desktops and RDS hosts are deployed across multiple geographic locations, instead of having to deploy and manage a separate Horizon pod next to the virtual desktop / RDSH capacity in each geographic location, this feature allows consolidation down to fewer Horizon pods that are centrally located.
For this use case, this feature applies to any currently supported version of Horizon 8.x.
Figure 29: Centralized Horizon Pod Consuming Capacity in Other Private Datacenters
In this scenario, existing Horizon pods, typically deployed in private data centers, can burst out to the public cloud without having to create and manage a new Horizon pod in that cloud platform. This allows the extension of existing Horizon pods to manage and use cloud capacity on one of the supported cloud platforms such as , , , , or .
Figure 30: Extending a Horizon Pod to Manage Cloud Capacity
To add cloud capacity to an existing Horizon pod:
- Use your cloud subscription and create a new software-defined data center (SDDC) on the desired cloud platform with the number of desired hosts.
- Configure networking between the data center that hosts the existing Horizon pod and the desired cloud data center and ensure that latency is 120 milliseconds or less.
- Add the vCenter of the newly created SDDC to the existing Horizon pod.
- The Horizon pod can start managing the newly added cloud capacity.
For the purpose of disaster recovery, we recommend at least 2 Horizon Pods located in two separate locations. If consolidating multiple locations to use a centralized Horizon Pod using this feature, consider the impact on disaster recovery and ensure that at least two locations have independent Horizon pods with their own sets of Connection Servers.
If the goal of going to the cloud is to provide a disaster recovery solution for an existing Horizon pod, it should be recognized that remote agents alone will not provide site redundancy. For disaster recovery, we recommend that you set up a separate Horizon pod in the desired cloud platform.
Using remote capacity (other data centers or cloud capacity) to host virtual desktops or RDSH servers, and remotely deploying Horizon agents does not change the size limitations or recommendations for a single Horizon pod. See .
Horizon Connection Servers within a single pod must be deployed in a single location and cannot be spread across private data centers and cloud data centers. See for more detail on why this is not supported.
Unified Access Gateways
Unified Access Gateways, if in use, should be deployed in the same location as the Horizon Connection Servers.
The location of the Unified Access Gateways is important as all external connections go through them and can impact latency. It is not supported to have the Unified Access Gateways for one pod spread across more than one location.
If you choose to locate the Unified Access Gateways in the same location as the remote agents, you should investigate deploying a standard Horizon pod in that location with all the management infrastructure, including the Connection Servers.
Latency between the data center where the existing Horizon pod with the Connection Servers are deployed and every single remote location where the Horizon Agents are deployed must be 120 milliseconds or less.
Networking connections between private data centers and cloud data centers vary depending on the cloud provider, we recommend that you work with your VMware technical team as well as your cloud provider for optimal configuration.
As with any Horizon deployment, correct networking and routing must be in place to ensure the components communicate properly.
Figure 31: Remote Agents Networking Considerations
With a remote agent deployment, attention should be paid to ensure that the components that are deployed remotely from each other can communicate properly. The diagram above highlights the key items with the numbers in circles referenced in the required traffic flows, below.
- Connection Servers to vCenter Servers.
- Horizon Agents in virtual desktops or RDS Hosts to the Connection Servers.
- For internal user sessions, the Internal Horizon Clients to Horizon Agents.
- For external user sessions, the Unified Access Gateways to Horizon Agents.
- Horizon Agents to Active Directory Domain Controllers/file servers.
The ability to use remote agents in a private data center applies to any currently supported version of Horizon 8.x.
When extending a Horizon pod to manage cloud capacity, the existing Horizon pod must be running Horizon 8 2006 or later.
- If Horizon 2106 and later is in use, when you add the cloud vCenter, you must select the correct deployment type. If this is selected incorrectly, instant clones may not provision properly. For more details, see .
- If Horizon 2103 or earlier is in use, you can set the deployment type of the cloud vCenter in the ADAM DB.
Horizon Control Plane services are not yet able to differentiate that capacity is deployed across multiple sites. As a result, Image Management Service functionality will not work when used on a vCenter Server that is remote to the Connection Servers.
This reference architecture documents and validates the deployment of all features of Horizon across two data centers.
The architecture has the following primary tenets:
- Site redundancy – Eliminate any single point of failure that can cause an outage in the service.
- Data replication – Ensure that every layer of the stack is configured with built-in redundancy or high availability so that the failure of one component does not affect the overall availability of the desktop service.
To achieve site redundancy,
- Services built using Horizon are available in two data centers that are capable of operating independently.
- Users are entitled to equivalent resources from both the primary and the secondary data centers.
- Some services are available from both data centers (active/active).
- Some services require failover steps to make the secondary data center the live service (active/passive).
To achieve data replication,
- Any component, application, or data required to deliver the service in the second data center is replicated to a secondary site.
- The service can be reconstructed using the replicated components.
- The type of replication depends on the type of components and data, and the service being delivered.
- The mode of the secondary copy (active or passive) depends on the data replication and service type.
Active-passive architecture uses two or more pods of Connection Servers, with at least one pod located in each data center. Pods are joined together using Cloud Pod Architecture configured with global entitlements.
Active-passive service consumption should be viewed from the perspective of the user. A user is assigned to a given data center with global entitlements, and user home sites are configured. The user actively consumes Horizon resources from that pod and site and will only consume from the other site if their primary site becomes unavailable.
Figure 32: Active-Passive Architecture
Active-active architecture also uses two or more pods of Connection Servers, with at least one pod located in each data center. The pods are joined using Cloud Pod Architecture, which is configured with global entitlements.
As with an active-passive architecture, active-active service consumption should also be viewed from the perspective of the user. A user is assigned global entitlements that allow the user to consume Horizon resources from either pod and site. No preference is given to which pod or site they consume from. The challenges with this approach are usually related to replication of user data between sites.
Figure 33: Active-Active Architecture
Stretched Active-Active - Unsupported
This architecture is unsupported and is only shown here to stress why it is not supported. Connection Servers within a given site must always run on a well-connected LAN segment and therefore cannot be running actively in multiple geographical locations at the same time.
Figure 34: Unsupported Stretched Pod Architecture
Multi-site Global Server Load Balancing
A common approach is to provide a single namespace for users to access Horizon pods deployed in separate locations. A Global Server Load Balancer (GSLB) or DNS load balancer solution can provide this functionality and can use placement logic to direct traffic to the local load balancer in an individual site. Some GSLBs can use information such as the user’s location to determine connection placement.
The use of a single namespace makes access simpler for users and allows for administrative changes or implementation of disaster recovery and failover without requiring users to change the way they access the environment.
Note the following features of a GSLB:
- GSLB is similar to a Domain Name System (DNS) service in that it resolves a name to an IP address and directs traffic.
- Compared to a DNS service, GSLB can usually apply additional criteria when resolving a name query.
- Traffic does not flow through the GSLB to the end server.
- Similar to a DNS server, the GLSB does not provide any port information in its resolution.
- GSLB should be deployed in multiple nodes in an HA or active/passive configuration to ensure that the GSLB itself does not become a point of failure.
Table 17: Strategy for Global Load Balancing
A global server load balancer was deployed.
This provides a common namespace so that users can access both sites.
Multi-site Architecture Diagram
The following diagram shows the server components and the logical architecture for a multi-site deployment of Horizon. For clarity, the focus of this diagram is to illustrate the core Horizon server components, so it does not include additional and optional components such as App Volumes, Dynamic Environment Manager, and Workspace ONE Access.
Figure 35: Multi-site Horizon Architecture
The following diagram shows the server components and the logical architecture for multiple pod deployment of Horizon, including additional components. This can be applied to pods deployed in either the same or different locations.
Figure 36: Multiple Horizon Pods with Additional Services
Universal Broker is used to provide a single FQDN (Fully Qualified Domain Name) for users to connect to and then access assignments in any Horizon pod. For more information, see Horizon Universal Broker in .
Each site has a set of file shares for Dynamic Environment Manager. The IT Config share can be replicated and made available in both sites as users only require read access. The Profile Archive shares are active to users in one site only, although they can be replicated to an alternative location, to be used in an outage event. For more information, see Multi-site Design in .
The Composer server is only required when using linked clones and is considered the legacy method for creating clones. It was deprecated in Horizon 8 2006, and from Horizon 8 2012 onwards is no longer supported. Instant clones do not require a Composer server. The recommendation is to use instant clones in preference to linked clones.
The Composer service works with the Connection Servers and a vCenter Server. Each Composer server is paired with a vCenter Server in a one-to-one relationship. For example, in a block architecture where we have one vCenter Server per 4,000 linked-clone VMs, we would also have one Composer server.
High availability is provided by vSphere HA, which restarts the Composer VM in the case of a vSphere host outage. VM monitoring with vSphere HA can also attempt to restart the VM in the case of an operating system crash.
If the VMware View Composer service becomes unavailable, all existing desktops can continue to work just fine. While vSphere HA is restarting the Composer VM, the only impact is on any provisioning tasks within that block, such as image refreshes or recomposes, or creating new linked-clone pools.
Summary and Additional Resources
Now that you have come to the end of this design chapter on Horizon 8, you can return to the and use the tabs, search, or scroll to select your next chapter in one of the following sections:
- Overview chapters provide understanding of business drivers, use cases, and service definitions.
- Architecture chapters give design guidance on the products you are interested in including in your platform, including Workspace ONE UEM, Workspace ONE Access, Workspace ONE Assist, Workspace ONE Intelligence, Horizon Cloud Service, Horizon, App Volumes, Dynamic Environment Manager, and Unified Access Gateway.
- Integration chapters cover the integration of products, components, and services you need to create the platform capable of delivering the services that you want to deliver to your users.
- Configuration chapters provide reference for specific tasks as you build your platform, such as installation, deployment, and configuration processes for Workspace ONE, Horizon Cloud Service, Horizon, App Volumes, Dynamic Environment Management, and more.
The following updates were made to this guide:
Author and Contributors
This chapter was written by:
- , Senior Staff End-User-Computing (EUC) Architect in End-User-Computing Technical Marketing, VMware.
Your feedback is valuable.