Optimizing Desktop Virtualization Solutions wit te Cisco UCS Accelerator Solution Brief February 2013 Higligts Delivers linear virtual desktop storage scalability wit consistent, predictable performance Reduces te need for expensive networked storage Simplifies solution manageability wit fewer arcitectural components required Improves TCO wit reduced costs for storage, servers, power, cooling, rack space, and operations Dramatically improves application response time Reduces te marginal cost of virtual desktops A cost-effective, ig-performance, scalable solution for non-persistent virtual desktops deployed on te Cisco Unified Computing System For virtual desktop infrastructure (VDI) applications, offloading a portion of storage IO to te Cisco UCS Accelerator in te local to server can result in a lower per-desktop cost for storage, lower overall costs, significantly improve predictable performance for scaling VDI at lower infrastructure cost at significantly lower and power and center footprints compared to traditional sared-storage approaces. Te Cisco UCS B200 M3 Blade Server wit Cisco UCS Accelerator delivers tese important aspects along industry leading management capabilities for organizations seeking to implement VDI. Callenge: Desktop Virtualization and te I/O Bottleneck Desktop and application virtualization are increasingly popular ways for enterprises to reduce capital and operating expenses, improve efficiency, increase control, and expand connectivity. Wit virtual desktops, users can access teir desktop images for laptops, tin-clients, smartpones, or oter devices from a osted, centralized infrastructure in a data center. Wile many enterprises are increasingly turning to desktop virtualization to provide increased business agility, risk mitigation, and support for bring-your-own-device (BYOD) initiatives, many organizations quickly realize tat te newfound success of virtual desktop infrastructure (VDI) as demonstrated in a small pilot or proof-ofconcept (POC) does not readily extend to a full-scale production environment. Wit undreds or tousands of users logging onto teir virtual desktops at te same time at te start of te work day, boot storms tat drain network performance can become a regular occurrence. Te back-end data center infrastructure becomes deluged wit storage I/O traffic as desktop operating systems initialize and users log in. An activity tat was previously confined to a user s laptop or desktop (and te embedded ard disk serving te desktop operating image) is now stretced across te network and into te back-end storage infrastructure, wic may not ave been adequately sized to meet te storage I/O needs of te end-user population. 1 2013 Cisco and/or its affiliates. All rigts reserved.
Te I/O Balancing Act For example, for a typical VDI environment wit 150 virtual desktops booted witin a four-minute window, a peak aggregate demand from all desktops of 39,000 read I/O operations per second (IOPS) and 2250 write IOPS or approximately 260 read IOPS and 15 write IOPS per virtual desktop were observed. Compare tese numbers wit a steady-state environment, were IOPS ave settled down after te boot and login storm. Te same study saw approximately 11 read IOPS and 7 write IOPS per virtual desktop. Tese numbers present an extremely wide variance across wic a VDI arcitect must build a solution for consistent performance and optimized cost. Many organizations struggle wit te task of predicting ow to appropriately size teir storage I/O performance needs. It s difficult to balance peak loads seen during boot and login storms, antivirus scans, and normal steady-state operations tat consume muc fewer IOPS. Oversizing for peak demand results in a uge capital expenditure (CapEx) burden and wasted infrastructure resources. Undersizing results in dissatisfied end users wo compare te lackluster performance against te more responsive desktops tey used before. Tis situation is likely to result in te early demise of any desktop virtualization initiative or to at least stall its progress in becoming a full-fledged production-ready solution. A related issue for traditional sared-storage approaces is lumpiness of scale. Put anoter way, traditional approaces for VDI storage typically incur a non-linear cost increase as implementers expand teir user populations, going from teir first 100 users to 1000 users and beyond. Typically, saredstorage array controllers can support a presized number of IOPS. Wit te growt in te number of virtual desktops, one of two approaces is followed. Eiter te number of controllers is increased or te scale of te existing controller is increased. Wit eiter approac, significant additional capital expenditure is required, increasing te marginal cost of adding virtual desktops for te next tier of users. An additional concern for VDI implementers is related to management complexity. Wit traditional sared- storage arrays, manageability of te SAN along wit te various virtual and pysical control points may overburden desktop virtualization deployments. Eac virtual desktop can ave four or more logical storage elements (including te image, write cace, profiles, and user data). Mapping and placing tese appropriately on te storage array logical unit numbers (LUNs) can create a storage manageability callenge, especially as tese elements grow along wit an increasing number of desktop users. Te Cisco Unified Computing System: Optimized for Desktop Virtualization IT organizations seeking to implement desktop virtualization are increasingly turning to te Cisco Unified Computing System (Cisco UCS) as teir compute platform of coice. Te industry s first unified data center platform, Cisco UCS delivers a converged, programmable infrastructure tat simplifies and speeds enterprise-class application and service deployment in bare metal, virtualized, and cloud-computing environments. Unified, model-based management, end-to-end provisioning, and migration support come togeter in tis next-generation data center platform to accelerate and simplify application deployment wit greater reliability and security. Cisco UCS provides tese important features: Integration of Cisco servers and network and I/O resources into one system Improvement of enterprise application availability and performance Scalability of service delivery to increase business agility Streamlining of data center resources to reduce total cost of ownersip Radical reduction in te number of devices requiring setup, management, power, cooling, and cabling Cisco UCS B-Series Blade Server: Te VDI Workorse Cisco UCS is canging te economics and performance curve of server-osted client computing, delivering a robust, igperformance computing fabric on wic desktop virtualization can be deployed. Te latest generation of Intel Xeon processors in Cisco UCS blade and rack-mount servers is used to ost user desktops. Te enterprise-class Cisco UCS B200 M3 Blade Server furter extends te capabilities of te Cisco UCS portfolio in a alf-blade form factor. It arnesses te power of te Intel Xeon E5-2600 processor product family, offering up to 768 GB of RAM, up to two SSD/HDD drives, and up to 80 Gb/s Eternet server I/O connectivity to deliver exceptional levels of performance, memory expandability, and I/O trougput for nearly all applications. 2 2013 Cisco and/or its affiliates. All rigts reserved.
Te Cisco UCS Accelerator Solution: Taking te Guesswork out of Sizing I/O for VDI Te Cisco UCS B200 M3 Blade Server offers new innovations wit greater performance and value for VDI workloads based on a large, on-server cace storage footprint (referred to as Tier-0 storage). Instead of relying on back-end sared storage to ost te golden master image and associated clone images for users, te same image can now be stored locally on Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapter installed on te B200 M3 Blade Server. A central copy of te golden master image can still be kept on back end-sared storage, along wit user sares, profiles, and persona. Te UCS Accelerator solution is an ideal server cacing solution to deliver uncompromised I/O to support a guaranteed number of users at not only lower cost but also wit predictable performance compared wit a SAN-based infrastructure (Figure 1). Now VDI administrators can linearly scale te number of supported users and deliver consistent performance for undreds to tens of tousands of users in a pooled, nonpersistent VDI implementation. Because te on-server cace storage approac involves a directly mapped flas storage cace tat supports muc iger IOPS tan conventional disk-based storage, VDI environments can now boot seamlessly witout bottlenecks and in a fraction of te time and cost of networked storage solutions. Implementing VDI on te Cisco UCS B200 M3 Blade Server wit te Cisco UCS Accelerator consists of te following steps: Direct te ypervisor to store virtual desktop images on te Cisco UCS B200 M3 Blade Server s on-board server cace provided by te on-board Fusion-io flas module Store user data and profile information on back-end networked storage (typically a NAS or SAN environment). Note: Tis can include specific persona management solutions tat can be layered on top of non-persistent, pooled desktop implementations Wit VMware vspere 5.1 (wit vmotion wit a unified arcitecture), live migration of virtual desktops, including memory and storage, is supported witout te requirement of sared storage. Tis makes it possible for te vmotion caracteristics of virtual desktops in te Cisco UCS Accelerator solution to be similar to tat of sared storage. For ig availability, in te unlikely event tat a server fails, users simply reconnect and te VDI broker reinstantiates teir desktop on anoter server. Rebooting takes a fraction of te time consumed by traditional networked storage. Figure 1. Cisco UCS Accelerator Solution for VDI Cisco UCS Accelerator: Fusion-io iodrive2 Module Networked Golden Master Image User Sares Profiles Cisco UCS B-Series wit Server Cacing Solution Golden Master Image (local copy) Clones VDI Connection Brokers Clients 3 2013 Cisco and/or its affiliates. All rigts reserved.
VMware View VDI Solution wit Cisco UCS Accelerator Used wit te popular VMware View VDI solution, te Cisco UCS Accelerator on te Cisco UCS B200 M3 Blade Server osts VMware linked clones for faster performance, lower cost, and easier scaling (Figure 2). Citrix VDI Solution wit Cisco UCS Accelerator For tose organizations implementing a Citrix VDI solution, te Cisco UCS Accelerator solution can provide significant benefits. In a Citrix XenDestkop solution, te Cisco UCS Accelerator will ost te write cace (and optionally te vdisk) for optimal performance and scale (Figure 3). Figure 2. Cisco UCS Accelerator Solution for VDI used wit VMware View VMware B200 M3 Blade Server VDI Clones x 140+/ Blade View Manager Replica (WRITE/READ I/O) Virtual Disk Total 6 GB (Per Clone) Linked Clone Windows Swap File also Placed Here Volumes (Mounted on Flas LUNs assigned to Hypervisors) Figure 3. Cisco UCS Accelerator Solution for VDI Used wit Citrix VDI Solution UCS Server 128GB RAM, 2 CPU OS caces vdisk file data at te block level, 64-Bit OS provides better cacing performance, typically up to 1000 streams per server. CITRIX Provisioning Server Farm B200 M3 Blade Server VDI Clones x 140+/Blade (Streamed from PVS vdisk) XenDesktop 5.6 Brokers vdisk1 vdisk1 vdisk1 vdisks Syncronized Manually on Eac PVS Server (vdisk Standard Mode) (WRITE/READ I/O) PVS 6.1 Farm (Provides Resilience and Load Balancing for vdisk Stream) Set to Cace on Device Disk Volumes (Mounted on Flas LUNs assigned to Hypervisors) Virtual Disk Total 4 GB (Per Clone) PVS Write-Cace Windows Swap File also Placed Here Automatically by PVS Target Device Agent Write-Cace Deleted upon VM Sutdown 4 2013 Cisco and/or its affiliates. All rigts reserved.
Benefits of Employing VDI Solutions Based on Cisco UCS Accelerator Lower Cost of troug Lower Cost of IOPS: Te Cisco UCS Accelerator-based VDI solution delivers a dramatically lower-cost storage solution, wit predictable, consistent storage IOPS performance, as compared to sared enterprise storage solutions. As VDI deployments scale to larger user populations, te cost of storage scales more affordably wit a server-based cace solution suc as te Cisco UCS Accelerator. For example, in a deployment of 2000 virtual desktops, comparing SAN storage wit te server-based Cisco UCS Accelerator, storage costs were reduced by 50 percent (Figure 4). Figure 4. Cisco UCS Accelerator Solution for VDI: Cost Estimates for 2000 Virtual Desktops $350,000 $300,000 $250,000 $200,000 $150,000 $100,000 $50,000 1/2 te cost Figure 5. Performance of Traditional SAN-based Solution for VDI Virtual Desktops Server Platform + OS/Hypervisor Network Write latency= ~600 s- 15 ms Read latency= ~600 s- 10 ms Controller Media Te Cisco UCS Accelerator solution provides te disaggregation of te storage environment by removing te need to traverse te network, storage controller, and media, and secondly by containing te I/O requests for eac server witin te server. Performance of a VDI solution using te Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapter results in reduced latency because reads and writes occur on te Adapter (Figure 6). Figure 6. Cisco UCS Accelerator Solution Arcitecture for VDI 0 SAN wit Enterprise * Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapters** Virtual Desktops Server Platform + OS/Hypervisor Media * Includes industry standard enterprise storage to support up to 2000 desktops wit one year support ($286,000), four ports of Cisco Nexus 5000 Series, and four storage port licenses for Cisco UCS Fabric Interconnect, all at Average Selling Price prices. ** Includes Cisco UCS Accelerator at ASP *** Doesn t include support and maintenance costs Network Controller Write latency= ~30 s- 2 ms Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapters can provide up to 500,000 writes (512 KB) or approximately 10 times more write IOPS and 100,000 reads (512 KB) or approximately tree times more read IOPS tan a typical sared storage solution (sized for 2000 desktops). Improved Performance Wit SAN-based storage solutions for VDI, reads and writes are serviced by te storage array, requiring communications across te server, storage network, storage controller, and storage media. Tere is latency associated wit eac of tese elements (Figure 5). Additionally, for IOPS-intensive workloads, backlogs occur at te access fabric, storage controller, and storage media layers due to te serialization of IOPS, furter slowing te VDI application. Read latency= ~70 s- 20 ms Te Cisco UCS Accelerator arcitecture delivers in 10 times more IOPS (as compared to 20 solid state drives beind a RAID controller) and consistent performance of random reads/writes. Traditional queue depts and oters forms of I/O serialization are avoided. Te Cisco UCS Accelerator solution for VDI complements SAN investments and provides investment protection, consistent performance, and a ig degree of scalability because te IOPS workload is distributed across multiple servers as VDI solutions scale. 5 2013 Cisco and/or its affiliates. All rigts reserved.
Server-based Manageability: In traditional application arcitectures, tere are multiple management tools for network, storage, and compute environments. Wit Cisco UCS, Cisco UCS Manager provides unified, embedded management of all software and ardware components of te Cisco UCS across multiple blade cassis, rack servers, LAN connectivity, SAN networking, systems identity management, monitoring and troublesooting. Wit te Cisco UCS Accelerator in te server, te UCS Management capabilities are extended to cover Tier 0 storage as well. Tis dramatically simplifies storage management for VDI deployments. Figure 7 sows tat te same Cisco UCS environment for a single cassis and a single blade can scale to 20 cassis and 160 blades troug dynamic provisioning. And now, troug te addition of te Cisco UCS Accelerator, te Cisco UCS Manager manages te Tier 0 storage on te server in addition to all te many oter components it manages. Figure 7. Cisco UCS Accelerator Solution Arcitecture for VDI Mgmt LAN SAN A SAN B Controller Citrix Provisioning Server Citrix XenApp Farm Application Tier Servers Citrix XenDesktop Farm Citrix 3D PRO wit NVDIA GRID Mgmt LAN SAN A SAN B One Logical Cassis to Manage* LAN Connectivity SAN Networking Blade Cassis Server Blades Rack Servers + Tier 0 On Server Server Identity Management Monitoring, Troublesooting *160 servers managed in a single domain as a logical cassis Controller Citrix Provisioning Server Citrix XenApp Farm Application Tier Servers Citrix XenDesktop Farm Citrix 3D PRO wit NVDIA GRID 6 2013 Cisco and/or its affiliates. All rigts reserved.
Predictable Performance as te VDI Solution Scales By disaggregating te monolitic storage and placing te storage close to te virtual desktop workload in smaller blocks, te scalability of te VDI solution dramatically improves. Wit te smaller block of storage on te server (wic can support up to ten times more IOPS requirements) satisfying te desktop workload requirement, tis VDI block approac scales linearly witout external dependencies. As a VDI solution scales from undreds to tousands of virtual desktops, eac server wit a Cisco UCS Accelerator becomes a building block, enabling more cost-effective scalability wit predictable performance. Lower Power Consumption and Floor Space A VDI solution using te Cisco UCS Blade Server wit Cisco UCS Accelerator uses dramatically less power (about 16 times less power per virtual desktop) and as a muc smaller data center footprint tan traditional SANbased storage solutions (Figure 8). Summary As organizations seek to embrace te expected benefits of VDI, it is critical for tem to understand te importance of correctly sizing storage IOPS. Oversizing leads to wasted CapEx and underutilized infrastructure. Undersizing IOPS capacity leads to a disappointing end-user experience. Figure 8. Cisco UCS Accelerator Solution for VDI Compared to SAN-based : Power Consumption Industry-Standard Sared * Power Consumption on per Virtual Desktop 2.4 Te Cisco UCS Accelerator Solution for VDI eliminates te guesswork and balancing act associated wit correctly sizing a VDI implementation by delivering an onserver SSD-based storage footprint tat provides expansive local storage IOPS capacity. Te result is linearly scalable and predictable performance at a fraction of te cost of igperformance networked storage. VDI arcitects can now successfully eliminate storage I/O storms, maximize performance, and optimize teir capital and operating expenses, resulting in a quicker pat to VDI return on investment. For More Information Cisco Desktop Virtualization ttp://www.cisco.com/go/vdi Cisco Unified Computing System Programmable Infrastructure ttp://www.cisco.com/en/us/netsol/ ns1166/index.tml Unified Model-based Management ttp://www.cisco.com/en/us/netsol/ ns1169/index.tml Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapters** 0.15 0.00 0.50 1.00 1.50 2.00 2.50 3.00 Cisco Servers for te Cisco Unified Computing System ttp://www.cisco.com/en/us/prod/ ps10265/cisco_servers_for_ucs.tml Watts Per Virtual Desktop * Typical 15,000 RPM SAS drive on industry-standard enterprise storage consumes 8W, assuming 20 percent additional overeads for controller, selves, and management nodes. ** Te Cisco UCS 785GB MLC Fusion-io iodrive 2 Adapters typically consumes 20W and on average supports 130 virtual desktops on te Cisco UCS B200 M3 Blade Server. Americas Headquarters Cisco Systems, Inc. San Jose, CA Cisco Systems (USA) Pte. Ltd. Singapore Europe Headquarters Cisco Systems International BV Amsterdam, Te Neterlands. URL: www.cisco.com/go/trademarks. Tird party trademarks mentioned are te property of teir respective owners. Te use of te word partner does not imply a partnersip relationsip between Cisco and any oter company. (1110R) C22-726648-00 2/13