Cloud Connector sizing and scale considerations for HDX proxy

Proper sizing and configuration of Citrix Cloud Connectors™ are crucial for high availability, stability, and efficient performance in HDX™ proxy deployments that do not use the Rendezvous v1 or v2 protocol (non-Rendezvous mode). Cloud Connectors serve as a vital proxy between the Citrix Gateway service and VDAs, making their optimization essential for seamless operations.

Tested maximum capacities and recommended machine configurations given in this article provide guidelines to ensure optimal performance and stability in HDX proxy deployments.

Note:

The recommendations apply to Cloud Connector deployments functioning as proxies for HDX proxy deployments in non-Rendezvous mode.

Improperly sized or misconfigured Cloud Connectors can result in performance bottlenecks, service disruptions, and instability. These challenges become more significant when handling large volumes of HDX proxy traffic between the Citrix Gateway service and VDAs. Proper sizing is essential to avoid these challenges and maintain a reliable user experience.

Recommendations

The following are the recommendations to determine the connector sizing:

• Minimum Cloud Connectors: 3 Cloud Connectors to ensure high availability.

• Processors: Intel Ice Lake, Cascade Lake, and newer processors.

• Network: Azure Accelerated Networking or equivalent network settings in other cloud environments or hypervisors.

Ensure you test in your specific environment to confirm the appropriate connector sizing.

Test conditions

• CPU and memory requirements are for the base Operating System (OS) and Citrix services only. Third-party applications and services might require more resources.

• Sessions are launched at a sustained rate.

• Tests are performed on a single Windows connector. A minimum of three connectors is recommended to ensure high availability.

• Tests are performed with the Cloud Connector configured with Intel Ice Lake or Cascade Lake processors.

• The client, Citrix Gateway service PoP, Cloud Connector, and VDAs are hosted on Azure in the same region. They establish end-to-end HDX sessions in non-Rendezvous mode. The sessions run through the test Cloud Connector while continuously simulating data packets.

• Tests are performed with the D2ds_v5, D4ds_v5, D8s_v5, and D16s_v4 virtual machines.

Test methodology

Tests are performed to apply load and evaluate the performance of the connector component. During testing, performance data such as CPU usage on the Cloud Connector and end-to-end latency are collected. Citrix proprietary simulation tools are used to simulate VDAs and sessions. These tools replicate the behavior of traditional VDAs and sessions while requiring fewer resources than hosting actual sessions and VDAs.

Recommendations for Cloud Connector sizing in this article are based on data gathered from the following tests:

• Tests performed to determine the number of connections and throughput supported without exceeding the following predefined conditions:

  • CPU usage of the Citrix Gateway service connector process < 70%

  • End-to-end latency < 50 ms

• Tests performed with minimum, average, and maximum packet sizes, using average values for results.

Summary of test findings

The following table lists the minimum recommended Cloud Connector CPU and other important configurations:

Note:

The preceding concurrent HDX sessions and throughputs are recommendations and not hard limits. We recommend you to test your own HDX session limits in your specific environment.