Dell Reference Architecture for Microsoft Lync on Citrix XenDesktop 5.6

Transcription

1 Dell Reference Architecture for Microsoft Lync on Citrix XenDesktop 5.6

2 Table of Contents 1 Introduction Purpose of this document Scope Solution Architecture Overview Introduction Architectural Overview Overview of Solution Components Design Challenges Hardware Components Dell PowerEdge R720 Servers Local Storage Dell Wyse Endpoints Dell Wyse R Class Cloud Clients Dell Wyse Z Class Clients Dell Wyse D Class Thin Clients Force10 S55 Network Switch Network Topology Software Components Overview Microsoft Windows Server 2008 R2 SP1 and 2012 with Hyper V and Microsoft SCVMM Microsoft Lync Server Citrix XenDesktop Citrix ICA Protocol and HDX Citrix HDX RealTime Optimization Pack 1.2 for Microsoft Lync How the Optimization Pack Works Data Flow with the Optimization Pack Without the Optimization Pack Test Description, Methodology, and Results Overview Test Configuration Configuration Details Test procedures Test Results and Data Analysis Data Analysis: Audio Conferencing Workload Data Analysis: Video Conferencing Workload Conclusion References About the Authors... 28

3 1 Introduction 1.1 Purpose of this document 1.2 Scope This document describes a reference architecture and performance testing for Microsoft Lync 2010 client applications running within a Citrix XenDesktop 5.6 virtual desktop environment using Dell servers and thin clients. Dell PowerEdge servers hosted the Citrix XenDesktop virtual desktops, as well as the application and infrastructure services. Dell Wyse thin client devices were the user endpoint devices. The purpose of the testing was two fold: (1) to assess Microsoft Lync 2010 client application performance using Citrix virtual desktops on the thin clients, and (2) to examine the impact of the Citrix HDX RealTime Optimization Pack for Microsoft Lync, Version 1.2. This paper highlights performance metrics and user densities measured with and without the Citrix HDX RealTime Optimization Pack. Relative to delivering Microsoft Lync 2010 applications within a Citrix XenDesktop environment, the objectives of this document are to: Define the technical architecture for the solution. Define the hardware requirements that supported the design. Define any design constraints. Discuss relevant risks, issues, assumptions and concessions. Provide a breakdown of key design elements such that the reader receives an incremental or modular explanation of the design. Provide solution sizing, scaling, and component selection guidelines.

4 2 Solution Architecture Overview 2.1 Introduction Collaboration is increasingly important to today s workforce and operational practices. Organizations find that improved communication and strong collaboration among employees, business partners, and customers improves operational efficiency, increases productivity, and reduces costs, helping to contribute to business success. To facilitate effective collaboration, many companies are implementing Microsoft Lync to provide Unified Communications services: instant messaging (IM), application sharing, audio/video/web conferencing, and VoIP telephony. Microsoft Lync can help companies reduce travel and mobile phone expenses, lowering operational budgets while increasing communication and creativity among employees and key stakeholders. Dell and Citrix have developed a Unified Communications (UC) architecture to help organizations achieve cost savings and efficiencies when enabling collaborative services. This virtualized UC architecture supports Microsoft Lync 2010 client applications that execute within a Citrix XenDesktop VDI environment. Citrix XenDesktop provides on demand enterprise delivery of Microsoft Windows applications, including Microsoft Lync, allowing desktops to be virtualized, centralized, and managed within the corporate datacenter for better security, administrative simplicity, and reduced costs. This paper describes a virtualized UC architecture for deploying Microsoft Lync 2010 client applications within Citrix XenDesktop. It presents performance and scalability tests jointly conducted by Citrix and Dell engineers, and analyzes the results of that testing to help customers confidently size virtualized deployments for Microsoft Lync client delivery. 2.2 Architectural Overview In testing of this architecture, three Dell PowerEdge server platforms hosted infrastructure services, Microsoft Lync services, and Citrix XenDesktop services (Figure 1). A variety of Dell Wyse thin client devices acted as user endpoints for the provided application services. VMs Hosted DC1 DDC1 SQL1 VMM1 Dell Poweredge 720 Infrastructure Hyper V server 1 10 x Win7 VMs without Optimization pack Wyse Thin Clients. ICA / HDX. Dell Poweredge 720 Compute Hyper V server for VDI VM s VMs Hosted 10 x Win7 VMs with Optimization pack 1.2 LYNCFE LYNCBE LYNCFS LYNCMA VMs Hosted Dell Poweredge 720 Infrastructure Hyper V server 2

5 Figure 1. Dell/Citrix architecture for Microsoft Lync. The architecture was tested to demonstrate the efficient delivery of Microsoft Lync client applications, and to determine the impact of the add on Citrix HDX RealTime Optimization Pack. Test results presented in this paper can help to guide customers in sizing actual deployments Overview of Solution Components The key components in the reference architecture include: Dell PowerEdge R720 Servers. These dual socket platforms run the Intel Xeon E family of processors, with up to 24 DIMMs for a maximum of 768GB RAM, and support for up to SAS disks, providing uncompromising performance and scalability in a compact 2U form factor. The Intel Xeon E processor family features 32 nanometer process technology with up to 8 cores per processor, delivering fast processing for compute intensive multimedia applications. Dell Wyse Thin and Cloud Clients. Dell Wyse thin client devices are well suited to virtual desktop environments because they bring the benefits of simplified security and centralized management. A variety of thin and cloud client devices are available, including those that support high definition multimedia graphics, voice, and video for collaborative services. Citrix XenDesktop. Citrix XenDesktop is a desktop virtualization solution that transforms Windows desktops and applications into an on demand VDI service. With XenDesktop, you can securely deliver Windows, web, and Software as a Service (SaaS) applications, or full virtual desktops to PCs, Macs, tablets, smartphones, laptops, and thin clients all with a high definition user experience. In this architecture, Citrix XenDesktop provided pooled desktops for a variety of Dell Wyse thin client devices. Citrix HDX RealTime Optimization Pack for Microsoft Lync, Version 1.2. This package supports clear, crisp high definition video and audio conferencing in conjunction with Microsoft Lync. Testing of this architecture compared the performance of Microsoft Lync workloads with and without the Optimization Pack to understand impact on user densities. Microsoft Windows Server 2008 R2 Service Pack 1 and Microsoft Windows Server 2012 Enterprise Editions with Hyper V and SCVMM. Virtual machines (VMs) were hosted on Microsoft Windows Server 2008 R2 Service Pack 1 and Microsoft Windows Server 2012 with Hyper V. Microsoft System Center 2012 Virtual Machine Manager (SCVMM) was used to manage the environment. Microsoft Lync Server 2010 Enterprise Edition. Microsoft Lync supports instant messaging, shared applications, audio and video conferencing, and interactive voice. To take advantage of these collaborative capabilities, users access a Lync client that connects to a frontend Lync server to establish communications with other users Design Challenges Microsoft Lync imposes a computationally intensive workload on servers, especially because of its requirement to drive real time audio and video output across low bandwidth networks. To meet requirements for voice and video conferencing, the VDI architecture must be capable of fast compute processing, low latencies, high throughput, and fast video rendering. To understand the capabilities of the design and the user densities that it can support, testing focused on the highly demanding workloads of audio and video conferencing. Since instant messaging (IM) imposes a nominal impact on performance and user densities, it was not part of the tested workload Citrix ICA Protocol and Citrix HDX RealTime Optimization Pack 1.2 To deliver high quality multimedia capabilities for collaborative tools like Microsoft Lync, Citrix XenDesktop uses an efficient Citrix developed core protocol, ICA (Independent Computing Architecture), which includes HDX technology to create a rich end user experience. Citrix XenDesktop relies on the ICA protocol and HDX to pass data between XenDesktop servers and client endpoint devices. The protocol natively applies compression and optimizations to address many challenges associated with network latencies, taking advantage of endpoint functionality and rendering locally or

6 remotely depending on endpoint capabilities. Installing the Citrix HDX RealTime Optimization Pack for Microsoft Lync, Version 1.2 adds further optimizations that take advantage of the local capabilities of Dell Wyse thin clients, accelerating performance of collaborative voice/video services Dell Infrastructure Technologies for VDI and Unified Communications While the Citrix XenDesktop optimizations are important, the scaling of Microsoft Lync clients also depends on powerful servers and capabilities that are inherent in the user endpoint devices. In this architecture, Dell PowerEdge servers and Dell Wyse thin clients provide the fast CPU, I/O, and rendering speeds to support efficient VDI and Unified Communications services. Implementing a successful virtualized UC initiative requires a comprehensive and proven architecture. Dell products and infrastructure services are designed to help deploy a cost effective, reliable, and responsive solution while: Lowering complexity and getting the VDI environment deployed quickly, with less risk. Centralizing management and efficiently distributing virtual and collaborative service workloads across the datacenter. Improving application continuity and meeting service level agreements (SLAs) Powered by Intel Xeon processors, the latest generation of high speed Dell PowerEdge R720 servers maximizes performance, providing fast processing speeds, large memory capacities, and flexible I/O. Dell Force10 high performance Ethernet switch/routers provide industry leading density and resiliency to simplify network delivery at low cost. All of these industry standard technologies come together in compact building blocks that help to simplify desktop virtualization and unified communications service deployments.

7 3 Hardware Components 3.1 Dell PowerEdge R720 Servers For purposes of architecture testing, the design incorporates best in class Dell Generation 12 server platforms Dell PowerEdge R720 servers. These dual socket platforms run the fastest Intel Xeon E family of processors, host up to 768GB RAM, and support up to SAS disks or SAS disks, providing uncompromising performance and scalability in a compact 2U form factor. In this test configuration, there were three Dell PowerEdge R720 servers used: one to host virtual machines (VMs) for the infrastructure, one to host VMs for Microsoft Lync server components, and one to host virtual desktops. All three servers had similar configurations as outlined below. Configuration of Dell PowerEdge R720 Servers 2 x Intel Xeon E Processor (2.5 GHz, 6 cores/12 threads per CPU) 64GB Memory (4 x 16GB 1600Mhz) Microsoft Windows Server 2008R2 Enterprise with Hyper V 6 x 146GB SAS 6Gbps 15k Disks PERC H710 Mini Integrated 1GB RAID Controller Broadcom Gb QP NDC (LAN) Broadcom Gb DP NIC (LAN) idrac7 Enterprise w/ vflash, 8GB SD 2 x 750W PSUs For more information on the Dell PowerEdge R720 servers (and other server offerings from Dell), please visit: LINK Local Storage The servers relied on local storage to store operating system images and virtual machines for the infrastructure servers, Microsoft Lync servers, and the Citrix XenDesktop virtual desktops. On each server, the configuration of local storage was as follows: Two 146GB SAS 6Gbps 15k disks were configured as RAID 1 volumes to host Windows Server 2008 with Hyper V.

8 The remaining four 146GB SAS 6Gbps 15k disks were configured as RAID 10 volumes to store the VMs. 3.2 Dell Wyse Endpoints Citrix XenDesktop can deliver virtual desktops to a variety of endpoint device types. Dell Wyse offers a wide selection of secure, reliable, and cost effective thin and cloud clients that integrate easily into any virtualized infrastructure while meeting budget and performance requirements. The test environment used several different models of Dell Wyse cloud client devices, specifically Wyse R class (Wyse R10L and R00LX), Wyse Z class (Wyse Z50D and Z90D7), and the Wyse D class (Wyse D90D7) clients. For more information on all Dell Wyse client devices, please visit: LINK Dell Wyse R Class Cloud Clients Wyse R class cloud clients contain powerful processors, dual monitor support, fast graphics, and a multiple USB ports for peripheral support. To support the demands of multimedia applications, these clients incorporate the advanced ATI 690E chipset from AMD. They typically use between 12 and 15 watts of power, compared to a PC that typically uses 70 to 150 watts. These clients use the Dell Wyse ThinOS, a thin client operating system. For more information on Dell Wyse R Class client devices, please visit: LINK Dell Wyse Z Class Clients The Wyse Z class clients have an efficient processor and a silent, fanless design that conserves power usage and emissions. AMD G Series Accelerated Processing Units create an ideal VDI platform for Unified Communications services. These client devices deliver excellent performance for HD multimedia graphics, voice, and video applications, and support hardware accelerated DirectX 11 graphics with OpenGL 4.0 and OpenCL support. For more information on Dell Wyse Z Class client devices, please visit: LINK. The Wyse Z90D7 is a high performance and Windows Embedded Standard 7 thin client for virtual desktop environments. Featuring a dual core AMD processor, its revolutionary design eliminates performance constraints, allowing it to achieve incredible speed and power for demanding multimedia applications and HD video. The Wyse Z50D is designed for power users, and is for demanding multimedia applications. It combines an embedded Wyse enhanced SUSE Linux Enterprise with a dual core AMD 1.6 GHz processor and a revolutionary graphics engine to accelerate 2D/3D graphics and HD video.

9 3.2.3 Dell Wyse D Class Thin Clients Running Windows Embedded Standard 2009, the Dell Wyse D90D7 client is a high performance thin client for virtual desktop environments. Featuring a unified engine that eliminates performance constraints, this client achieves outstanding speed and power for demanding VDI and embedded Windows applications, rich graphics, and HD video. Driving the high speed and performance is a powerful energy saving AMD G Series dual core 1.4GHz processor, creating a solid platform to support a range of applications. For more information on Dell Wyse D Class client devices, please visit: LINK. 3.3 Force10 S55 Network Switch In this architecture, all servers share a single Force10 S55 switch for network connections. As user density increases, adding another Force10 S55 switch is recommended to add redundancy. Model Features Options Uses Force10 S55 44 x BaseT (10/100/10 00) + 4 x SFP Redundant PSUs 4 x 1Gb SFP ports the support copper or fiber 12Gb or 24Gb stacking (up to 8 switches) 2 x modular slots for 10Gb uplinks or stacking modules Top of Rack (ToR) switch for LAN Public and Management Networks The diagram below shows features on the front and back of the Force 10 S55 switch. An optimal design uses the 10Gb uplink modules, creating 10Gb uplinks to a core or distribution switch. If 10Gb to a core or distribution switch is unavailable, the front 4 x 1Gb SFP ports can be used. The front 4 SFP ports can support copper cabling and can be upgraded to optical fiber to support longer distances. For more information on the Dell Force10 S55 switch, please visit: LINK

10 3.4 Network Topology As shown below, the network topology defines a management network as well as the network that supports VDI traffic between the servers and the thin clients.

11 4 Software Components 4.1 Overview The software technologies in this architecture deliver a rich multimedia experience for collaborative applications in a VDI environment. The software components include: Microsoft Windows Server 2008 R2 SP1 with Hyper V role Microsoft Windows Server 2012 with Hyper V role Microsoft System Center 2012 Virtual Machine Manager (SCVMM) Microsoft Lync Server 2010 Citrix XenDesktop 5.6 FP1 (feature pack) Citrix HDX RealTime Optimization Pack 1.2 for Microsoft Lync This section introduces the general functionality of each component and discusses configuration settings within Citrix XenDesktop relative to the reference architecture for Microsoft Lync. 4.2 Microsoft Windows Server 2008 R2 SP1 and 2012 with Hyper V and Microsoft SCVMM On two physical servers, Microsoft Windows Server 2012 Datacenter Edition was configured with Hyper V to host VMs for infrastructure and Microsoft Lync software services. All virtual servers on these physical machines were installed with Microsoft Windows Server 2008 R2 SP1. The third physical server was installed with Microsoft Windows 2008 R2 SP1 to host virtual desktops. Citrix XenDesktop can be hosted bare metal or on a choice of hypervisors (Citrix XenServer, VMware ESXi, or Microsoft Hyper V ). For this architecture, Citrix XenDesktop was deployed on Microsoft Server 2008 R2 SP1 with Hyper V. Best practices were applied by leveraging the Hyper V Best Practice Analyzer as noted in MS KB Citrix XenDesktop was configured to host Microsoft Windows 7 SP1 for all virtual desktops. Microsoft System Center 2012 Virtual Machine Manager (SCVMM) was used to manage the environment. 4.3 Microsoft Lync Server 2010 Microsoft Lync Server 2010 gives users a single interface for a variety of communications tools: voice, IM, and audio, video, and Web conferencing. It tracks presence information, such as user pictures, skills, and locations, giving users the context they need for communications. The same presence and contact information can be used across both Microsoft Lync 2010 and Microsoft Office applications. Users can collaborate more effectively using desktop and application sharing, Microsoft PowerPoint uploads, and rich whiteboarding. Microsoft offers several licensing options to scale as required for each site s Unified Communications needs. A Microsoft Lync Server 2010 license is required for each operating system environment running a server instance and there are two edition types: Microsoft Lync Server 2010 Standard Edition: Standard Edition supports IM, presence, conferencing, and an option for voice. It requires a single system to host the server components and database for storing user and conferencing information. Microsoft Lync Server 2010 Enterprise Edition: Enterprise Edition supports more sophisticated conferencing capabilities, and enables separation of server functionality and data storage to achieve higher densities and load balancing for better availability. Enterprise Edition was used in the testing of this reference architecture.

12 A Client Access License (CAL) is also required for each user or device accessing the Microsoft Lync Server. (For more information on licensing Microsoft Lync 2010, see licensing.aspx.) Microsoft Lync requires the following logical servers: A domain controller, installed with Active Directory Domain Services (AD DS), Active Directory Certificate Services (AD CS), and DNS Server roles. A Microsoft Lync Front End Server installed with Microsoft Lync Server 2010 Enterprise Edition. Front end services include Session Initiation Protocol (SIP) Registrar, SIP proxy, conferencing and other services such as A/V conferencing, Web conferencing, instant messaging, application sharing, response group, bandwidth policy, call park, conferencing announcement, and audio test. A Microsoft SQL Server Back End Server, installed with Microsoft SQL Server 2008 SP1. The Back End Server provides database services for the Front End pool. The Microsoft Lync FileShare Server, for storing files for the Microsoft Lync Server. A Monitoring/Archiving Server, installed with Microsoft Lync Server 2010 Enterprise Edition and Microsoft SQL Server 2008 SP1. Figure 2 shows virtual servers for Microsoft Lync services (as well as SCVMM, SQL, and other infrastructure services) hosted on two physical machines. Addresses are also shown for one of the two networks.

13 Figure 2. Two physical machines host Microsoft Lync and other infrastructure and management services. 4.4 Citrix XenDesktop Citrix XenDesktop is a desktop virtualization solution that delivers individual applications or complete hosted desktops to users across the entire enterprise. It combines the benefits of centralized management and security with a personalized user experience. Key features include: Personal vdisk. Citrix XenDesktop 5.6 gives IT the ability to deliver personal VDI desktops through innovative Personal vdisk technology that enhances personalization and reduces storage costs. This feature centrally stores a single copy of Microsoft Windows and combines it with a personal vdisk for each user s apps, data, and settings, simplifying enterprise wide deployment of virtual desktops. High definition user experience (HDX) technology. XenDesktop 5.6 contains significant enhancements to HDX technology, delivering virtual desktops over WANs to mobile workers and

14 branch office employees up to three times faster than previous releases. Key enhancements include faster printing and scanning, faster app launch, and flexible Quality of Service (QoS) controls to optimize the user experience. XenDesktop 5.6 also features significant multimedia, voice and video enhancements, including new Flash redirection technology that enhances video and audio performance over WANs. Broad device support. Using Citrix XenDesktop 5.6, customers can deliver self service applications and desktops to more than one billion devices, including PCs, Macs, tablets, smartphones, and thin clients and all major device operating platforms, including Apple ios, Google Android, and Google ChromeOS. XenDesktop 5.6 enables a rich, native experience on each device, including support for gestures and multi touch features, customizing the experience based on the type of device and leveraging device features to optimize performance. Support for Microsoft RemoteFX. XenDesktop with HDX and RemoteFX (a key capability of Microsoft Hyper V ) delivers a great user experience for rich content, regardless of whether execution occurs on the server or client device. A pooled VDI desktop model was used in this architecture to deploy desktop images to the Dell Wyse thin clients. Pooled VDI desktops use a single OS image to create multiple thinly provisioned or streamed desktops. In this test scenario, pooled desktops were thinly provisioned from a single OS image and streaming was not used. Using hypervisor APIs, Machine Creation Services (MCS) in XenDesktop were used to deliver desktop images and to create, start, stop, and delete virtual machines. Citrix XenDesktop 5.6 Feature Pack 1 supports the optional use of Personal vdisks. A pooled VDI desktop can use a Personal vdisk to maintain application, profile and data differences that are not part of the base image. This creates a persistent personal desktop, reducing the need for independent and dedicated desktops in the enterprise. For the purpose of testing this Microsoft Lync architecture, Personal vdisks were not configured. Supporting Remote PC FlexCast delivery, Citrix XenDesktop 5.6 Feature Pack 1 includes enhancements to Citrix Receiver and HDX technologies to take advantage of local endpoint rendering where possible. To support the Citrix HDX RealTime Optimization Pack 1.2, Citrix XenDesktop 5.6 Feature Pack 1 is required. 4.5 Citrix ICA Protocol and HDX For less intensive application workloads, the native Citrix ICA and HDX optimizations make desktop virtualization scalable and practical, even over low bandwidth and high latency WAN connections. HDX technology includes these features that help to optimize performance and provide a rich highdefinition user experience: MultiStream ICA Protocol splits virtual desktop traffic into 5 streams real time, interactive, background, bulk and RTP Voice to enable network administrators to prioritize traffic by type and maintain QoS. RTP Voice delivers optimal audio performance, even over high latency networks. Adaptive compression and de duplication algorithms optimize for network efficiency. Seamless isochronous plug and play provides support for webcams and USB audio devices. Client side webcam video compression, where possible, reduces bandwidth requirements. Voice over IP SDK supports leading telepresence applications. The optimization offloads voice traffic from virtual desktops and processes the voice stream locally using advanced voice routing. Peer to peer connections enable enterprise scale video conferencing. For more info on HDX, see defexperience.html

15 4.6 Citrix HDX RealTime Optimization Pack 1.2 for Microsoft Lync Citrix HDX RealTime Optimization Pack 1.2 supports features that enable a highly scalable solution for delivering real time audio video conferencing and USB or VoIP enterprise telephony using Microsoft Lync in Citrix XenDesktop, XenApp, and VDI in a Box environments. Users can participate in audiovideo or audio only calls to and from other HDX RealTime users and other standards based video desktop and conference room systems. The installation files for the Optimization Pack are available for download via the XenDesktop 5.6 Feature Pack 1, which is available from The Optimization Pack contains both client and server components: The client component, called Citrix HDX RealTime Media Engine, is integrated with the Citrix Receiver on the endpoint device and performs all signalling and media processing directly on the user device itself, offloading the server for maximum scalability, minimizing network bandwidth consumption, and ensuring optimal audio video quality. The server side (i.e., virtual desktop) component, Citrix HDX RealTime Connector, is a connector to the Microsoft Lync client that drives the RealTime Media Engine on the endpoint. The Connector runs in the virtual server environment alongside Microsoft Lync client and communicates signalling information over a Citrix ICA virtual channel to the RealTime Media Engine running on the user device How the Optimization Pack Works To understand how the architecture works when the Optimization Pack is used, picture a typical softphone architecture, as illustrated below. At the top, a UI layer allows users to make and answer calls. There is some business logic in the middle, and a media engine at the bottom that handles the audio video encoding/decoding and Session Initiation Protocol (SIP) signalling. User Interface Business Logic Media Engine By pulling the media engine out from this architecture and moving it over to the user device, the workload of media processing also moves to the user device. In such an optimized architecture, the communications between the media engine and the layer of software above it, which would normally happen on the same machine, now take place over a virtual channel. This inter process communication, however, consists only of orchestration commands (e.g., Make a call ). The actual media traffic flows directly from the media engine on the user device to the other party on the call (or to a conferencing bridge in the case of a multi party call). As a result, there is no media processing happening on the server, only on the user device. Server Side (Virtual Desktop) User Interface User Device Media Engine Virtual Business Logic When the Citrix HDX RealTime Optimization Pack 1.2 is configured, it changes where processing occurs and how data is routed between solution components. The next sections illustrate how the

16 Optimization Pack reroutes media data from user device to user device, moving processing of media data off of the server that hosts the virtual desktops Data Flow with the Optimization Pack Provided as part of the Optimization Pack, the RealTime Media Engine is installed on each thin client. When a user opens the Microsoft Lync client in a virtual desktop, the RealTime Media Engine on the user device is initialized and registers with the Microsoft Lync server. Session Initiation Protocol (SIP) is used to place and initialize the call session, and only signalling information is sent over the ICA protocol to the XenDesktop server (Steps 1 4 in Figure 3). After the call session is established, all media traffic flows directly peer to peer (from one user endpoint directly to the other, as shown in Step 5 in Figure 3). In the case of a multi party call, media traffic is routed directly to the Microsoft Lync Audio Video Conferencing Server. Figure 3. Data Flow with Citrix HDX RealTime Optimization Pack 1.2. The RealTime Media Engine running on the endpoint device transmits compressed video directly from the endpoint and avoids sending uncompressed video over the network. This reduces network load as well as decreases processing on the virtual desktop server. In this way the Optimization Pack makes the solution work well even in WAN environments and allows LAN based deployments to scale. In addition, the RealTime Media Engine performs decompression on the endpoint and not in the Citrix server, which helps to significantly increase VDI scalability. Of course, for optimal performance, the thin clients should feature codecs that accelerate compression and decompression operations on the endpoints. Citrix has validated the Optimization Pack for Microsoft Lync on a number of Dell Wyse thin client devices. To see the list of validated devices, see realtime optimization pack 12/hdx realtimeoptimization pack 12 system requirements.html. As Figure 3 illustrates, the media traffic does not go through the Citrix XenDesktop server at all. The RealTime Media Engine routes audio/video directly between clients over UDP and bypasses TCP

17 based VDI protocols entirely, thereby allowing voice and video traffic to flow with minimal latency and no delay spikes Without the Optimization Pack Figure 4 illustrates the media data flow when the Optimization Pack is not used. When the user opens Microsoft Lync in the virtual desktop, the user registers with the Microsoft Lync server. Similar to when the Optimization Pack is used, Session Initiation Protocol (SIP) is used to place and initialize the call session. Signalling information is again sent over the ICA protocol to the virtual desktops on the Citrix XenDesktop server. Once the call is established, however, media traffic continues to flow through the virtual desktop, with every inbound and outbound packet passing through the VDI Hyper V host. Figure 4. Data Flow with no Optimization Pack. Because all audio/video traffic is flowing over ICA (from endpoint1 to Virtual Desktop1 to Virtual Desktop2 to endpoint2, as shown in Step 4 in Figure 4), additional latency is introduced. In addition, all compression and decompression operations occur on the virtual desktops and not in the endpoints, resulting in a significant CPU impact on the VDI Hyper V host. The collected test metrics in the next section demonstrate the performance tradeoffs that occur without the use of the Optimization Pack.

18 5 Test Description, Methodology, and Results 5.1 Overview To simulate the desktop experience of a VDI environment with a Microsoft Lync 2010 workload, Citrix and Dell engineers configured a small test environment. The goal was to assess VDI user experience and analyze performance of Microsoft Lync audio and video conferencing applications running within virtual desktops with and without the Optimization Pack. Two test scenarios were created: Test case #1 simulated shared audio, establishing three separate audio conferences Test case #2 simulated shared video, supporting two separate video conferences With each test scenario, engineers conducted multiple test runs: first using Citrix XenDesktop without the Optimization Pack, and then again using Citrix XenDesktop with the Optimization Pack. The intent was to understand the impact of the Optimization Pack on VDI user densities with Microsoft Lync 2010 workloads. 5.2 Test Configuration As shown back in Figure 1 (page 5), three Dell PowerEdge R720 servers were configured with the following virtual machines on Microsoft Server 2008 with Hyper V: Citrix XenDesktop VDI Server: 10 x Windows 7 VMs (no Optimization Pack) 10 x Windows 7 VMs with Optimization Pack 1.2 Infrastructure Server: Domain Controller (DC1) XenDesktop Controller (DDC1) System Center Virtual Machine Manager (VMM1) SQL Database used for SCVMM and XenDesktop (SQL1) Microsoft Lync Server: Microsoft Lync Front End Server (LYNCFE) Microsoft Lync Back End Server (LYNCBE) Microsoft Lync File Server (LYNCFS) Microsoft Lync Monitoring and Archiving Server (LYNCMA) Configuration Details Engineers performed the following procedures to configure the Dell PowerEdge R720 servers and create the test environment: Installing the Microsoft Windows Server OS (Microsoft Windows Server 2008 R2 SP1 was installed on the compute host while Microsoft Windows Server 2012 was installed on the two infrastructure servers) Configuring the Hyper V hypervisor on the three compute and infrastructure servers Creating the required infrastructure VMs Setting up the infrastructure Domain Controller Setting up the infrastructure Microsoft SQL servers Setting up the Microsoft SCVMM 2012 server Setting up the Citrix XenDesktop 5.6 FP1 server Setting up the Microsoft Lync infrastructure

19 Creating the target VDI VMs with Microsoft Windows 7 x86 installed on them Setting up the Citrix HDX RealTime Optimization Pack for Microsoft Lync on both the server and on the Dell Wyse thin client devices Citrix has previously validated the Optimization Pack for Microsoft Lync on a number of thin client devices. To see the list, visit realtime optimizationpack 12/hdx realtime optimization pack 12 system requirements.html Infrastructure and Desktop VM Configuration The table below summarizes configuration parameters for the various infrastructure Microsoft Lync VMs. VM Name Processor RAM DC1 2 vcpus 2 GB VMM1 4 vcpus 4 GB DDC1 4 vcpus 4 GB SQL1 4 vcpus 8 GB LyncFE 4 vcpus 4 GB LyncMA 4 vcpus 4 GB LyncFS 4 vcpus 4 GB LyncBE 4 vcpus 8 GB The table below summarizes configuration parameters used to configure target virtual desktops for the XenDesktop pools. One pool contains 10 VMs that are installed with the Optimization Pack; the other pool contains 10 VMs without the Optimization Pack. Configuration Parameter CPU resource Memory Write Cache Desktops Configured XenDesktop Setting 1 vcpu Dynamic Memory: 1024MB 2048MB 2GB (Fixed) 10 with Optimization Pack; 10 without Optimization Pack 5.3 Test procedures The testing was largely a manual process. For each test run of either a voice or video conferencing workload, test engineers performed this sequence of steps: 1) Logging into each thin client and connected to its desktop VM. 2) Starting a script to collect performance metrics (using perfmon) at the hypervisor level. (Due to time limitations, data was not collected at the thin client endpoints.) 3) Manually initiating the Microsoft Lync voice or video connection. 4) Simulating the conferencing workload. (An MP3 player or tablet played an audio or video stream to simulate a voice or video source.) 5) Stopping the Lync session and stopping the performance monitoring. 6) Logging off the desktop VM. 7) Analyzing the collected data and graphing the performance results.

20 5.4 Test Results and Data Analysis The tests compared performance of voice and video conferencing workloads on thin clients running virtual desktops. Performance data was collected for the desktops configured with and without the Citrix HDX RealTime Optimization Pack for Microsoft Lync. Results and analysis are presented in the following pages Data Analysis: Audio Conferencing Workload The test simulated three audio conferences occurring concurrently between three pairs of users, each using one of the six virtual desktops running on a thin client device. Network bandwidth, logical processor utilization, memory utilization, and disk I/O datapoints were collected for each of the virtual desktops as well as the Hyper V host Audio Without Optimization Pack The following graphs show network bandwidth, logical processor utilization, and memory utilization recorded under an audio conferencing workload using six virtual desktops without the Citrix HDX RealTime Optimization Pack for Microsoft Lync. Disk I/O metrics are not included because disk I/O was not impacted. In the data below, the audio conference sessions started at minute 3:00 and stopped at minute 8:00. Please note that in the test scenario, a total of eight virtual desktops were in a running state on the host although only six participated in audio conferencing. Network Bandwidth The graph below shows the cumulative network bandwidth utilization for the six virtual desktops measured at the Hyper V host network adapter. The six VMs consumed approximately 150,000 bytes, using an average total network bandwidth of approximately 1171 kbps (195 kbps per VM).

21 Logical Processor Utilization The graph below shows the aggregate logical processor utilization consumed by the six virtual desktops. It shows the %Guest Runtime on the Hyper V server. The %Guest Runtime is the average percentage of time the guest code is running across all logical processors. Prior to the start of the audio conferences, utilization was approximately 1% and increased to 8% at peak, indicating that the six virtual desktops consumed approximately 7% (or about 1.17% per VM) of the available logical processing power during the audio calls. Thus, a single server can support a theoretical maximum of approximately 84 VMs running voice calls. Memory Utilization Dynamic memory was configured for the Hyper V VMs. Each VM started at 1 GB of RAM and dynamically grew to a maximum of 2 GB. The graph below shows that memory utilization did not impact the performance results. With eight virtual desktops running, memory consumption of the desktops consisted of approximately 8GB, resulting in a total memory consumption of about 10 GB including the parent partition. Total Memory (GBytes) Memory Utilization Audio With Optimization Pack The following graphs show network bandwidth, logical processor utilization, and memory utilization recorded under an audio conferencing workload using six virtual desktops with the Citrix HDX RealTime Optimization Pack for Microsoft Lync. Metrics for disk I/O are not included because disk I/O was not impacted. In the data below, the audio conferencing sessions started at minute 5:00 and stopped at minute 10:00. Please note that in this scenario, a total of 16 virtual desktops were in a running state on the host (twice the number as in the previous scenario). Network Bandwidth The graph below shows the cumulative network bandwidth utilization for the six virtual desktops measured at the Hyper V host network adapter. In the graph below, there are a few network spikes due to changes on the screen requiring redraws. The overall network utilization is close to zero.

22 Logical Processor Utilization The graph below shows the aggregate logical processor utilization consumed by the six virtual desktops. It shows the %Guest Runtime on the Hyper V server. The %Guest Runtime is the average percentage of time the guest code is running across all logical processors. Prior to the start of the audio conferences, utilization was about 2%. This is twice that of the previous scenario (which saw about 1% utilization) because there were double the number of machines running. With the Optimization Pack, utilization during the conferences fluctuated but the average remained around 2 3%. This is largely below the 8% utilization level that was consistently observed in the previous test run without the Optimization Pack.

23 Memory Utilization The graph below shows that memory utilization did not impact the performance results. The 16 running virtual desktops consumed about 16 GB, resulting in a total memory consumption of about 20 GB including the parent partition. Total Memory (GBytes) Memory Utilization Data Analysis: Video Conferencing Workload The test simulated two video conferences occurring concurrently between two pairs of users, each using one of four virtual desktops running on a thin client device. Data was collected for each of the virtual desktops at the Hyper V level. All of the virtual desktop sessions were connected at a resolution of 1680x Video Without Optimization Pack The following graphs show network bandwidth, logical processor utilization, and memory utilization recorded under the video conferencing workload using four virtual desktops without the Citrix HDX RealTime Optimization Pack for Microsoft Lync. In the data below, the video conferencing sessions started at minute 7:00 and stopped at minute 12:00. Please note that in this scenario a total of eight virtual desktops were in a running state on the host. Network Bandwidth The graph below shows the cumulative network bandwidth utilization for the four virtual desktops measured at the Hyper V host network adapter. The 4 VMs consumed approximately 550,000 bytes, using an average total network bandwidth of about 4300 kbps (1074 kbps per VM).

24 Logical Processor Utilization The graph below shows the aggregate logical processor utilization for the four virtual desktops. It shows overall processor utilization of the Hyper V server of 12% (3% utilization per VM). Thus, a single server can support a theoretical maximum of approximately 33 VMs running video conferences. Memory Utilization The graph below shows that memory utilization did not impact the performance results. The eight running virtual desktops consumed approximately 8 GB, resulting in a total memory consumption of about GB including the parent partition. Total Memory (GBytes) Memory Utilization Video With Optimization Pack The following graphs show network bandwidth, logical processor utilization, and memory utilization recorded under the video conferencing workload using four virtual desktops with the Citrix HDX RealTime Optimization Pack for Microsoft Lync. In the data below, the video sessions started at minute 5:45 and stopped at minute 11:00. Please note that in this scenario, a total of 16 Virtual desktops were in a running state on the host (twice the number as in the previous scenario). Network Bandwidth The graph below shows the cumulative network bandwidth utilization for the four virtual desktops measured at the Hyper V host network adapter. In the graph below we can see few network spikes due to changes on the screen requiring redraws. The overall network utilization is close to zero.

25 Logical Processor Utilization The graph below shows the aggregate logical processor utilization for the four virtual desktops. It shows the overall processor utilization of the Hyper V server. Memory Utilization The graph below shows that memory utilization did not impact the performance results. The 16 running virtual desktops consumed approximately 16 GB, resulting in a total memory consumption of about 20 GB including the parent partition. Total Memory (GBytes) Memory Utilization

26 6 Conclusion As shown in the collected performance metrics, using the Citrix HDX RealTime Optimization Pack for Microsoft Lync decreased the processing load on the virtual desktops and the amount of network bandwidth consumed. The Optimization Pack allows the solution to take advantage of the capabilities of the Dell Wyse thin clients and scale collaborative services more effectively.

27 References For more information, see the following resources: Dell PowerEdge servers, r720/pd Dell Wyse thin clients, xenith class/pd Citrix XenDesktop, Citrix HDX RealTime Optimization Pack for Microsoft Lync, realtime optimization packwrapper.htm Microsoft Lync, Documentation: Wyse Enhanced Microsoft Windows Embedded Standard 7 WFR2 Administrators Guide A Sizing Study of Microsoft Lync Server 2010 and its Back End SQL Database on Dell PowerEdge Servers

28 About the Authors Loay Shbeilat is a Solutions Architect for Worldwide Alliances with Citrix. Loay has over 13 years of experience and expertise on the broader Microsoft and Citrix solutions software stack, as well as in enterprise virtualization, storage, networking, and enterprise data center design. Senthil Baladhandayutham is the Solutions Development Manager in the Desktop Virtualization Solutions Group at Dell, managing the development and delivery of enterprise class desktop virtualization solutions based on Dell data center components and core virtualization platforms. Gus Chavira is a Principle Solution Architect in the Desktop Virtualization Solutions Group at Dell in charge of building Reference Architecture. Solution direction, testing, planning are all responsibilities of the S.A. In addition to this responsibility, Gus has over 20 years working in Virtualization and related fields. Neetu Arora is a Lead Solution Engineer in the Desktop Virtualization Solutions Group at Dell building, testing, validating, and optimizing enterprise VDI stacks. In addition to this responsibility, Neetu has 10 years of experience working in Virtualization and related fields.

29 THIS WHITE PAPER IS FOR INFORMATIONAL PURPOSES ONLY, AND MAY CONTAIN TYPOGRAPHICAL ERRORS AND TECHNICAL INACCURACIES. THE CONTENT IS PROVIDED AS IS, WITHOUT EXPRESS OR IMPLIED WARRANTIES OF ANY KIND Dell Inc. All rights reserved. Reproduction of this material in any manner whatsoever without the express written permission of Dell Inc. is strictly forbidden. Dell, the Dell logo, Wyse, PowerEdge, Force10, and EqualLogic are trademarks of Dell Inc. Citrix, FlexCast, XenDesktop, XenApp, and Citrix Receiver are trademarks of Citrix Systems, Inc. and/or one or more of its subsidiaries, and may be registered in the United States Patent and Trademark Office and in other countries. Microsoft, Windows, Windows Server 2012, and Hyper V are either trademarks or registered trademarks of Microsoft Corporation in the United States and/or other countries. Intel is a registered trademark of Intel Corporation in the U.S and other countries. Other trademarks and trade names may be used in this document to refer to either the entities claiming the marks or names of their products.