Rapidly deploy VMware View : Cisco HyperFlex HX220c M4 Node for Virtual Desktop Infrastructure with VMware Horizon View

Posted on Posted in Alıntı Dökümanlar

Executive Summary

A picture of a man working on his laptop in a library

The lack of a simple and efficient IT platform designed for easy consumption by application workloads is adding to the cost of doing business. Cisco HyperFlex Systems addresses this need while also including the desirable features of flexibility and consistent management.

Cutting the deployment time of the required compute, storage and network infrastructure for desktop virtualization to two hours or less is an order of magnitude faster than typical two-week deployment times for some converged systems.

When using VMware Horizon as the broker for virtual desktop infrastructure (VDI) for pooled, non-persistent Microsoft Windows virtual desktops and new with Horizon 7, for hosted Apps and desktops via integration with Microsoft RDS (Remote Desktop services) on Windows Server; allows the near immediate use of new capacity as it is brought online and provides seamless desktop image maintenance.

VMware Horizon pooled floating assigned linked-clone or instant-clone desktop performance on Cisco HyperFlex provides industry leading baseline and sustained Login VSI response times, 2 times faster than the leading hyperconverged competitor.[1] End users can expect one second or less response times with the cluster at full 1000 user load.

We have updated the original release of this paper to include performance data for recently released Intel Broadwell processors and faster memory and for Microsoft Windows 10 and Office 2016 workload.

It is clear that Cisco HyperFlex delivers industry leading end-user experience for VMware Horizon virtual desktop users on the latest desktop operating system and office suite on the customer’s provisioning platform of choice.

Reference Architecture Overview

This section defines current IT Infrastructure challenges and presents a high-level view of the solution and its benefits.

Document Purpose

This document describes a virtualized solution platform with eight Cisco HyperFlex HX220c M4 Nodes for up to 1000 VMware Horizon View linked-clone desktops (VDI). The document provides design and sizing guidelines to help you build solution platforms similar to the one tested. This document is not intended to be a comprehensive guide to every aspect of this solution.

Solution Purpose

Cisco HyperFlex Systems are built on the Cisco Unified Computing System™ (Cisco UCS®) platform. They offer faster deployment and greater flexibility and efficiency at a competitive price while lowering risk for the customer. Proven components from Cisco are integrated to form a software-defined storage (SDS) platform. This approach eliminates or reduces the need for planning and configuration decisions, while allowing customization to meet customer workload needs. The platform and management model adopted is an extension of the established Cisco UCS data center strategy, with familiar components managed in a consistent manner through a policy-based framework using Cisco UCS Manager.

Business Challenges

An efficient IT Infrastructure is integral to the initial success and continued competitiveness of most businesses. IT efficiency can be expressed in capital and operating costs to the business. Two major components of operating costs for all businesses are human resources and optimal utilization of purchased IT resources.

The underlying issues that contribute to these operating costs are as follows:

  • Complexity: Complex systems take longer to deploy and require a greater number of highly skilled technical staff members. The multitude of technologies and tools required to keep the infrastructure running and the nonstandard methods introduced by this approach have a direct effect on failure rates, contributing to even more costs to the business.
  • Stranded capacity: Even with virtualization, IT resource consumption is not optimal. The business requirements and computing and storage needs of workloads change over time, potentially resulting in unused computing or storage resources in the enterprise. One way to prevent this underutilization of resources is to introduce flexibility into the architecture so that you can expand computing and storage resources independently.

 

Efforts to reduce management complexity through consolidation of native element managers on pre-integrated and converged IT infrastructure have resulted in only limited improvements. These factors and the short depreciation cycles of capitalized IT resources point to the need for simpler and more precisely controlled components to achieve the necessary utilization levels.

The Solution

The Cisco HyperFlex solution focuses on simplicity of deployment and operation. It delivers a hyperconverged platform that has the advantage of allowing you to start small and grow in small increments without the need for expensive storage devices connected to computing resources by either SAN or network-attached storage (NAS) methods. A basic cluster requires three hyperconverged nodes managed by Cisco UCS Manager. Beyond this, a Cisco HyperFlex cluster can increase computing and storage resources for flexible scaling according to workload needs. Flexibility is introduced by creating a cluster with a mix of Cisco UCS B200 M4 Blade Servers as computing-only nodes connected to a set of Cisco HyperFlex HX240c M4 Nodes operating as hyperconverged nodes. In this scenario, the hyperconverged node provides storage for the Cisco UCS B200 M4 computing-only nodes. This feature allows either storage or computing capacity to be added independently to achieve optimal levels of cluster resources.

The Cisco HyperFlex solution also delivers storage efficiency features such as thin provisioning, data deduplication, and compression for greater capacity and performance improvements. Additional operational efficiency is facilitated through features such as cloning and snapshots.

This solution uses Cisco HyperFlex HX220c M4 Nodes, Cisco UCS fabric interconnects, Cisco UCS Manager, Cisco Nexus® 9372 platform switches, Cisco HyperFlex HX Data Platform (SDS) software, and VMware Horizon 6.2 and the ESXi 6.0 Update 1 (U1) hypervisor. The HX220c M4 Nodes provide computing, cache, and storage resources and are centrally managed by Cisco UCS Manager. The HX Data Platform software serves computing and networking resources and a shared pool of storage resources from separate Cisco HyperFlex HX-Series nodes for consumption by a mix of workloads. This SDS platform is managed by a plug-in in the VMware vCenter WebClient.

 

 

Solution Benefits

This solution provides the following benefits to customers:

  • Simplicity: The solution is designed to be deployed and managed easily and quickly through familiar tools and methods. No separate management console is required for the Cisco HyperFlex solution.
  • Centralized hardware management: The cluster hardware is managed in a consistent manner by service profiles in Cisco UCS Manager. Cisco UCS Manager also provides single-point console and firmware management capabilities. Cisco HyperFlex HX Data Platform (SDS) clusters are managed through a plug-in to VMware vCenter.
  • High availability: Component redundancy is built in to most levels at the node. Cluster-level tolerance to node, network, and fabric interconnect failures is implemented as well.
  • Efficiency: Complementing the other management efficiencies are features such as thin provisioning, data deduplication, compression, cloning, and snapshots to address concerns related to overprovisioning of storage.
  • Flexibility: “pay as you grow”: Customers can purchase the exact amount of computing and storage they need and expand one node at a time up to the supported cluster node limit.

 

Customers who have already invested in Cisco® products and technologies have the opportunity to mitigate their risk further by deploying familiar and tested Cisco UCS technology.

Main Components

A Cisco HyperFlex cluster can consist of three to eight nodes that are similarly configured, except in a mixed configuration that includes blades and hyperconverged nodes. The best practice is to create a highly available cluster with N+1 resiliency, in which the cluster can sustain all virtual machines with one node in maintenance mode or in a failed state. This solution requires a minimum of four converged nodes per cluster. Converged nodes have processing, cache, and capacity layers together in a unit such as a Cisco HyperFlex HX220c M4 Node.

Each node has a 120-GB solid-state disk (SSD) drive, used for Cisco HyperFlex HX Data Platform housekeeping and logs, and a larger 480-GB high-endurance SSD drive, used for write logs and for caching read and write data. Storage capacity is provided by a set of six 1.2-terabyte (TB) 10,000-rpm 12-Gbps SAS hard disk drives (HDDs). Cisco HyperFlex HX-Series nodes are managed by Cisco UCS Manager hosted in a pair of fault-tolerant, low-latency Cisco UCS fabric interconnects.

The network layer can be any pair of switches with 10-Gbps bandwidth. In this case, a pair of Cisco Nexus 9372 platform switches in standalone mode is used for connection to the existing network. The hypervisor used is VMware ESXi 6.0 U1.

 

 

Table 1 lists the main components used in a Cisco HyperFlex cluster and the versions tested.

Table 1.       Main Components of Cisco HyperFlex Cluster

Layer Device Image
Computing Cisco HyperFlex HX220c M4 Node
Cisco UCS 6248UP 48-Port Fabric Interconnect Release 2.2(6f)
Network Cisco Nexus 9372 platform switch Release 6.1(2)I3(1)
Storage SSD: 1 x 120 GB and 1 x 480 GB SSD drives
HDD: 6 x 1.2-TB HDDs
Software VMware vSphere ESXi Release 6.0 U1
VMware vCenter Release 6.0 U1
Cisco HyperFlex HX Data Platform Release 1.7.1

 

The description of the solution architecture provides more details about how these components are integrated to form the Cisco HyperFlex HX Data Platform.

Cisco HyperFlex HX-Series Hardware

The Cisco HyperFlex HX220c M4 Node is an essential component in the design of Cisco HyperFlex software-defined infrastructure. It combines the Cisco UCS C220 M4 Rack Server with the SDS functions of the Cisco HyperFlex HX Data Platform. Cisco HyperFlex HX-Series nodes are then integrated into a converged fabric by connecting to a pair of Cisco UCS 6248UP fabric interconnects. Cisco UCS Manager, hosted in the fabric interconnects, is used to manage all hardware in the cluster from a single console.

Important innovations in the platform include a standards-based unified network fabric, Cisco virtualized interface card (VIC) support, and Cisco Extended Memory technology. The system uses a single-connect architecture, which eliminates the need for multiple Ethernet connections from each node in the cluster. Thus, cluster expansion is simplified, resulting in quicker deployment and fewer errors.

Cisco Nexus 9000 Series Switches

The solution requires a redundant set of low-latency, 10-Gbps switches for connection to shared services such as Microsoft Active Directory, Domain Name System (DNS), Network Time Protocol (NTP), Dynamic Host Configuration Protocol (DHCP), and VMware vCenter. The switches used for this purpose are a pair of Cisco Nexus 9372PX Switches operating in standalone mode.

The Cisco Nexus 9372PX and 9372PX-E Switches provide a line-rate Layer 2 and Layer 3 feature set in a compact form factor. Each switch offers a flexible switching platform for both three-tier and spine-and-leaf architectures as a leaf node. With the option to operate in either NX-OS mode (for Cisco NX-OS Software) or ACI mode (for Cisco Application Centric Infrastructure [Cisco ACI™]), these switches can be deployed in small business, enterprise, and service provider environments.

The Cisco Nexus 9372PX and 9372PX-E Switches have forty-eight 1- and 10-Gbps Enhanced Small Form Pluggable (SFP+) ports and six Quad SFP+ (QSFP+) uplink ports. All the ports are line rate, delivering 1.44 terabits per second (Tbps) of throughput in a 1-rack-unit (1RU) form factor.

To provide investment protection, a Cisco 40-Gbps bidirectional transceiver allows reuse of an existing 10 Gigabit Ethernet multimode cable plant for 40 Gigabit Ethernet. The solution also supports 1- and 10-Gbps access connectivity for data centers that are migrating access switching infrastructure to faster speeds.

Storage

Physical storage in Cisco HyperFlex Systems is provided by individual hyperconverged nodes in the cluster. A converged node provides computing and memory resources, an SSD-based cache layer for staging read and write operations, and a capacity layer that includes varying numbers of spinning media (HDDs) for persistent storage.

Cisco HyperFlex software consolidates isolated pockets of storage from individual converged nodes into a log-structured file system called the Cisco HyperFlex HX Data Platform. The log-structured file system assembles blocks to be written to the cache until a configurable write log is full or until workload conditions dictate that it be destaged to a spinning disk. When existing data is (logically) overwritten, the log-structured approach appends a new block and updates the metadata. When the cache is destaged, the write operation consists of a single disk seek operation with a large amount of data written. This approach improves performance significantly compared to the traditional read-modify-write model, which is characterized by numerous seek operations and small amounts of data written at a time.

Data blocks written to disk are compressed into objects and sequentially laid out in fixed-size segments. The objects are distributed across all nodes in the cluster to make uniform use of storage capacity. By using a sequential layout, the platform helps increase flash-memory endurance and makes the best use of the read and write performance characteristics of HDDs, which are well suited for sequential I/O. The platform includes enterprise features such as thin provisioning, space-efficient clones, and snapshots for data protection. Inline deduplication and compression are turned on by default and contribute to significant increases in resource utilization.

Because read, modify, and write operations are not used, compression and snapshot operations have little or no performance impact. The file system has been designed from the foundation to provide very efficient mechanisms for deduplication, compression, and snapshotting.

VMware vSphere

VMware vSphere provides a common virtualization layer (the hypervisor) for a computer’s physical resources: the VMware ESX host. The hypervisor allows provisioning of precisely controlled and fully functional virtual machines with the required CPU, memory, disk, and network connectivity characteristics. Virtual machines can run the operating system and application workload of choice in an isolated manner for increased utilization of the underlying hardware.

The high-availability features of VMware vSphere 6.0 that are relevant to this solution include the following:

  • VMware vMotion: Provides live migration of virtual machines within a virtual infrastructure cluster, with no virtual machine downtime or service disruption
  • VMware Storage vMotion: Provides live migration of virtual machine disk (vmdk) files between data stores whether within or across storage arrays, with no virtual machine downtime or service disruption
  • VMware vSphere High Availability: Detects and provides rapid recovery for a failed virtual machine in a cluster
  • VMware Distributed Resource Scheduler (DRS): Provides load balancing of computing capacity in a cluster
  • VMware Storage Distributed Resource Scheduler (SDRS): Provides policy-based storage management, such as load balancing across multiple data stores based on space use and I/O latency

 

VMware vCenter Server provides a scalable and extensible platform that forms the foundation for virtualization management for the vSphere cluster. vCenter manages all vSphere hosts and their virtual machines.

Cisco HyperFlex Data Platform

In a Cisco HyperFlex System, the data platform requires a minimum of three Cisco HyperFlex HX-Series converged nodes for the default three-way mirroring of data. To create a highly available cluster with N+1 resiliency, the solution considers a minimum of four hyperconverged nodes per cluster. Each node includes a Cisco HyperFlex HX Data Platform controller that implements the distributed file system using internal flash-based SSD drives and high-capacity HDDs to store data. The controllers communicate with each other over 10 Gigabit Ethernet to present a single pool of storage that spans the nodes in the cluster. Individual nodes access data through a data layer using file, block, object, or API plug-ins. As nodes are added, the cluster scales to deliver computing, storage capacity, and I/O performance.

In the VMware vSphere environment, the controller occupies a virtual machine with a dedicated number of processor cores and memory, allowing it to deliver consistent performance and not affect the performance of the other virtual machines in the cluster. The controller can access all storage resources without hypervisor intervention through the VMware VMDirectPath feature. It uses the node’s memory and SSD drives as part of a distributed caching layer, and it uses the node’s HDDs for distributed capacity storage. The controller integrates the data platform into VMware software through the use of two preinstalled VMware ESXi vSphere Installation Bundles (VIBs):

  • IO Visor: This VIB provides a network file system (NFS) mount point so that the ESXi hypervisor can access the virtual disk drives that are attached to individual virtual machines. From the hypervisor’s perspective, it is simply attached to a network file system.
  • vStorage API for Array Integration (VAAI): This storage offload API mechanism is used by vSphere to request advanced file system operations related to snapshots and cloning from the underlying storage subsystem. The controller causes these operations to occur by manipulating the metadata rather than actually copying data, providing rapid response and thus rapid deployment of new application environments.

 

Figure 1.      Cisco HyperFlex Data Optimization Flow

fig_1

As shown in Figure 1, the IO Visor intercepts workload traffic and stripes the block across available nodes in the cluster. The data then bypasses the hypervisor using VMDirectPath and is cached on the larger cache disk in one of the dedicated partitions. Replication across nodes takes place at this layer. Write blocks continue to be written to write logs until they are full, at which time they are marked as passive and destaged to disk. Data optimization processes such as deduplication and compression occur when the data is destaged from the cache and before it is written to disks.

The data platform implements a log-structured file system (LogFS) that uses a caching layer in the SSDs to accelerate read requests and write responses, and a persistence layer implemented with HDDs for capacity. The log-structured layer replicates incoming data to one or more SSDs located in different nodes before the write operation is acknowledged to the application. This process allows incoming write operations to be acknowledged quickly while protecting data from SSD or node failures. If an SSD or node fails, the replica is quickly re-created on other SSDs or nodes using the available copies of the data.

The log-structured distributed object layer also replicates data that is moved from the write cache to the capacity layer. This replicated data is likewise protected from hard-disk or node failures. A total of three data copies are available, enabling you to survive disk or node failures without risk of data loss. See the Cisco HyperFlex HX Data Platform system administration guide for a complete list of fault-tolerant configurations and settings.

VMware Horizon 6 with View

VMware Horizon with View brings the agility of cloud computing to the desktop by transforming desktops into highly available and agile services delivered from the cloud (Figure 2). View delivers virtual sessions that follow end users across devices and locations. It enables fast, secure access to corporate data across a wide range of devices, including Microsoft Windows, Mac OS, and Linux desktop computers and iOS and Android tablets.

With VMware vCenter Server, View can be used to create desktops from virtual machines that are running on VMware ESXi hosts and to deploy these desktops to end users. After a desktop is created, authorized end users can use web-based or locally installed client software to connect securely to centralized virtual desktops, back-end physical systems, or terminal servers. View uses the existing Microsoft Active Directory infrastructure for user authentication and management. See the VMware Horizon with View documentation center for more information about product enhancement and new features introduced.

Figure 2.      VMware Horizon with View Components

VMware View Storage Accelerator

VMware View Storage Accelerator is an in-memory host caching capability that uses the content-based read cache (CBRC) feature in ESXi hosts. CBRC provides a per-host RAM-based solution for View desktops, which greatly reduces the number of read I/O requests that are issued to the storage layer. It also addresses boot storms—when multiple virtual desktops are booted at the same time—which can cause a large number of read operations. CBRC is beneficial when administrators or users load applications or data frequently. Note that CBRC was used in all tests that were performed on the solution described here: Horizon with View running pooled linked-clone desktops hosted on the Cisco HyperFlex system.

Solution Architecture

Figure 3 shows the cluster topology described in this document.

Figure 3.      Cluster Topology

Reference Configuration

The hardware platform consists of nodes with dual processors and 384 GB of memory each. The processors are Intel® Xeon® processor E5-2680 v3 CPUs, each with 12 cores and operating at 2.5 GHz. The density and number of memory DIMMs selected are consistent with Intel’s recommendations for optimal performance given the number of memory channels and DIMMs supported by each CPU (Figure 4).

Figure 4.      Processor Configuration

System performance is optimized when the DIMM type and number are the same for each CPU channel. Peak memory performance is at 2133 MHz for up to two DIMMs for each CPU. Therefore, a higher-density DIMM is recommended to provide the required amount of memory while also delivering peak memory performance by not expanding to the third memory channel.

Hardware and Software Specifications

Table 2 provides cluster-based specifications. It shows the hardware required for a minimum configuration cluster with N+1 resiliency.

Table 2.       Cluster-Based Specifications

Description

Specification

Notes

Hyperconverged nodes

8 Cisco HyperFlex HX220c M4 Nodes

A cluster can consist of 4 to 8 nodes.

Fabric interconnects

2 Cisco UCS 6248UP fabric interconnects

Fabric interconnects provide policy-based stateless computing.

Layer 2 switches

2 Cisco Nexus 9372PX Switches

Optional: Deploy a pair of any 10-Gbps switches for connectivity.

Table 3 provides individual node specifications. It presents component-level details for each node in a cluster.

Table 3.       Individual Node Specifications

Description

Specification

Notes

CPU

2 Intel Xeon processor E5-2680 v3 CPUs

Memory

24 x 16-GB DIMMs

FlexFlash Secure Digital (SD) card

2 x 64-GB SD cards

Boot drives

SSD

1 x 120-GB SSD

Configured for housekeeping tasks

1 x 480-GB SSD

Configured as cache

HDD

6 x 1.2-TB 10,000-rpm 12-Gbps SAS drives

Capacity disks for each node

Hypervisor

VMware vSphere 6.0 U1b

Virtual platform for SDS

Cisco HyperFlex HX Data Platform software (SDS)

Cisco HyperFlex HX Data Platform Release 1.7.1

You can use the VMware vCenter web client to obtain an enterprisewide view. You can set up each vCenter server instance to manage multiple Cisco HyperFlex HX-Series clusters to scale with separation across clusters for greater security. You can also then extend this model by connecting multiple instances of vCenter servers using vCenter linked mode if desired. Linking multiple vCenter instances allows you to view and search resources across all instances (Figure 5).

Figure 5.      Enterprisewide View

Server Details

Table 4 shows a device-level mapping between the Cisco HyperFlex components and the underlying node hardware.

Table 4.       Device-Level Mapping

Component Function

Cisco HyperFlex HX220c

Device

Comments

Boot (ESXi)

FlexFlash SD cards 1 and 2

SD card 1 mirrored to SD card 2

Control virtual machine bootstrap

FlexFlash SD cards 1 and 2

SD card 1 mirrored to SD card 2

Control virtual machine housekeeping, data, and logs

SSD 1 in front-slot 1

(120 GB)

/var/log, /var/core, and /zookeeper

(housekeeping)

Cache layer

SSD 2 in front-slot 2

(480 GB)

Intel 3610-based high-endurance SSD for caching

Capacity layer

6 x 1.2-TB HDDs in slots 3 to 8

10,000-rpm 12-Gbps SAS HDDs

Network Layout

Figure 6 shows a virtual network interface card (vNIC), which is a virtualized hardware device presented as a vmnic to the ESXi host, and virtual switch (vSwitch) setup for networking. The goal is to provide redundant vNICs for each vSwitch and to provide parallel paths with sufficient bandwidth and separation to prevent congestion from affecting performance. Four VLANs are created: for management (routable subnet), NFS storage (jumbo frames), production workload traffic, and VMware vMotion (jumbo traffic).

Figure 6.      Network Layout

Figure 7 shows the configuration for this design.

Figure 7.      Networking Configuration

Quality-of-Service Setup in Cisco UCS Manager

Quality of service (QoS) refers to the capability of a network to provide better service to selected network traffic. The primary goal of QoS is to provide priority for specific traffic, including dedicated bandwidth and latency and improved loss characteristics. In configuring QoS, you also need to help ensure that prioritizing one type of traffic flow does not make other flows fail.

Some of the four subnets used in this design require larger frames, or jumbo frames, which are any frames with a payload of more than 1500 bytes. For the fabric interconnects to pass along these jumbo frames, the appropriate priority needs to be set in the QoS system class section in Cisco UCS Manager. Table 5 shows the settings: priority, class of service (CoS), maximum transmission unit (MTU), etc.

Table 5.       QoS Settings

Priority

Enabled

CoS

Packet Drop

Weight

MTU

Multicast Optimized

Platinum

Yes

5

No

4

9216

No

Gold

Yes

4

Yes

4

Normal

No

Silver

Yes

2

Yes

Best effort

Normal

Yes

Bronze

Yes

1

Yes

Best effort

9216

No

Best Effort

Yes

Any

Yes

Best effort

Normal

No

Fibre Channel

Yes

3

No

Best effort

Fibre Channel

The QoS policy associates vNIC templates with the appropriate QoS priority (Table 6).

Table 6.       QoS Policy Settings

QoS Policy Name

QoS Class

Burst Size

Rate

Host Control

Platinum

Platinum

10,240

Line rate

None

Gold

Gold

10,240

Line rate

None

Silver

Silver

10,240

Line rate

None

Bronze

Bronze

10,240

Line rate

None

Best Effort

Best Effort

10,240

Line rate

None

The final step is to edit the vNIC template with the QoS policy and adjust the MTU size (Table 7).

Table 7.       vNIC Template Settings

vNIC Template Name

MAC Address Pool

Fabric

MTU

QoS Policy

Other Parameters

mgmt-a

mgmt-a

A

1500

Silver

Network control policy: hyperflex-infra

mgmt-b

mgmt-b

B

storage-a

storage-a

A

9000

Gold

Network control policy: hyperflex-infra

storage-b

storage-b

B

vm-network-a

vm-network-a

A

1500

Platinum

Network control policy: hyperflex-vm

vm-network-b

vm-network-b

B

vmotion-a

vmotion-a

A

9000

Bronze

Network control policy: hyperflex-infra

vmotion-b

vmotion-b

B

Make similar changes to the corresponding vNIC settings in VMware vCenter.

Network Setup in Cisco UCS Manager

Figure 8 shows the network setup in Cisco UCS Manager.

Figure 8.      Network Setup in Cisco UCS Manager

Storage Layout

The Cisco HyperFlex HX Data Platform controller handles all read and write requests for volumes that the hypervisor accesses and thus mediates all I/O from the virtual machines. The hypervisor has a dedicated boot disk independent of the data platform.

Incoming data is distributed across all nodes in the cluster to optimize performance using the caching tier. Effective data distribution is achieved by mapping incoming data to stripe units that are stored evenly across all nodes, with the number of data replicas determined by the policies you set. When an application writes data, the data is sent to the appropriate node based on the stripe unit that includes the relevant block of information. This data distribution approach in combination with the capability to have multiple streams writing at the same time helps prevent both network and storage hot spots. It also delivers the same I/O performance regardless of virtual machine location and provides more flexibility in workload placement.

Figure 9 shows the storage layout for the Cisco HyperFlex HX Data Platform.

Figure 9.      Cisco HyperFlex HX Data Platform

●   Data write operations: For write operations, data is written to the local SSD cache and is replicated over the 10-Gbps subnet to the remote SSDs in parallel before the write operation is acknowledged. This approach eliminates the possibility of data loss due to SSD or node failure. The write operations are then staged to inexpensive, high-density HDDs for long-term storage. By using high-performance SSDs with low-cost, high-capacity HDDs, you can optimize the cost of storing and retrieving application data at full speed.

●   Data read operations: For read operations, data that is local will usually be read directly from the local SSD. If the data is not local, the data is retrieved from an SSD on the remote node. This approach allows the platform to use all SSDs for read operations, eliminating bottlenecks and delivering superior performance. Data recently read from the persistent tier is cached both in SSDs and in memory. Having the most frequently used data stored so close to the virtual machines helps make Cisco HyperFlex Systems perform very well for virtualized applications.

Thus, when you move a virtual machine to a new location, using, for instance, VMware DRS, the HX Data Platform does not require movement of the data. Movement of virtual machines thus has no impact on performance or cost.

Figure 10 shows the Cisco HyperFlex HX Data Platform.

Figure 10.    Cisco HyperFlex HX Data Platform

The HX Data Platform decouples the caching tier from the persistence tier and supports independent scaling of I/O performance and storage capacity. This flexibility sets the stage for the introduction of blades as computing-only nodes in a setup in which storage from converged nodes can be consumed. This mixed configuration allows independent scaling, which addresses the problem of stranded capacity.

In the event of a problem in the HX Data Platform controller software, data requests from the applications residing in that node are automatically routed to other controllers in the cluster.

Performance Testing

The desktop virtualization performance evaluation of the freshly installed Cisco HyperFlex HX220c M4S 8-node cluster running the general availability Release 1.7.1 software was conducted following the Cisco Virtual Desktop Infrastructure Test Protocol and Success Criteria for Desktop Virtualization using Login VSI 4.1.2. Figure 11 shows the configuration tested.

Figure 11.    Configuration Used for Performance Testing

The Cisco HyperFlex System runs in a cluster on VMware ESXi 6.0 U1b managed by the Cisco HyperFlex plug-in to the VMware vCenter web client. A 1000-seat VMware Horizon with View 6.2.2 pooled floating assigned linked-clone desktop configuration was used in each test case.

This fresh-installation testing used two test cases running the Login VSI 4.1.5 knowledge worker workload in benchmark mode:

●   48-minute Login VSI knowledge worker test in benchmark mode (end-user experience during login storm)

●   8-hour Login VSI knowledge worker test in benchmark mode (workday stability in steady state)

The tests monitored performance statistics during a virtual desktop boot storm and then waited for the systems to settle for about 20 minutes. Testing tracked ramp-up, which is the login interval for all 1000 sessions; steady state, in which all sessions are logged on and active; and logoff.

48-Minute Standard Benchmark Mode Test Results

Login VSI tests were run on 1000 linked-clone desktop virtual machines hosted on eight Cisco HyperFlex HX220c M4S servers configured in Cisco HyperFlex clusters with exceptional user performance as represented by the Login VSI Analyzer score and latency values (Figures 12 through 18).

Test result highlights include:

●   0.7 second Login VSI baseline response time

●   1.1 seconds Login VSI average response time with 1000 desktops running

●   2 seconds maximum Login VSI response time with 1000 desktops running

●   Average CPU utilization of 85 percent during steady state

●   Average of 300 GB of RAM used out of 384 GB available

●   388 MBps peak network utilization per host.

●   Aggregate of 6 milliseconds (ms) of I/O latency per cluster

●   12,500 peak I/O operations per second (IOPS) per cluster at steady state

●   250 Mbps peak throughput per cluster at steady state

Figure 12.    Login VSI Analyzer Chart for 1000 Sessions with End-User Experience (EUX) Response Time
Figure 13.    Host CPU Utilization from VMware ESXTOP: 1000 Linked Clones, Average CPU Utilization as a Percentage
Figure 14.    Host Memory Use from VMware ESXTOP: 1000 Linked Clones, Average Memory Use in MB
Figure 15.    Host Network Utilization from VMware ESXTOP: 1000 Linked Clones, Average Network Utilization in Mbps
Figure 16.    Host Storage IOPS from VMware ESXTOP: 1000 Linked Clones, Average IOPS
Figure 17.    Host Storage Read-Write Rate from VMware ESXTOP: 1000 Linked Clones, Average Read-Write Rate in MBps
Figure 18.    Performance Statistics from Web User Interface (WebUI): 1000 Linked Clones

8-Hour Benchmark Mode Test Results

Login VSI tests were run on 1000 linked clones hosted on eight Cisco HyperFlex HX220c M4S servers with exceptional user performance as represented by the Login VSI Analyzer score and latency values (Figures 19 through 25).

Test result highlights include:

●   0.7 second Login VSI baseline response time

●   1.1 seconds Login VSI average response time with 1000 desktops running

●   2.5 seconds Login VSI maximum response time with 1000 desktops running

●   Average CPU utilization of 90 percent

●   Average of 300 GB of RAM used out of 384 GB available

●   438 MBps peak network utilization per host

●   Aggregate of 8 ms of I/O latency per cluster

●   11,700 peak IOPS per cluster at steady state

●   211 Mbps peak throughput per cluster at steady state

Figure 19.    Login VSI Analyzer Chart for 1000 Sessions with EUX Response Time
Figure 20.    Host CPU Utilization from VMware ESXTOP: 1000 Linked Clones, Average CPU Utilization as a Percentage
Figure 21.    Host Memory Use from VMware ESXTOP: 1000 Linked Clones, Average Memory Use in MB
Figure 22.    Host Network Utilization from VMware ESXTOP: 1000 Linked Clones, Average Network Utilization in Mbps
Figure 23.    Host Storage IOPS from VMware ESXTOP: 1000 Linked Clones, Average IOPS
Figure 24.    Host Storage Read-Write Rate from VMware ESXTOP: 1000 Linked Clones, Average Read-Write Rate in MBps
Figure 25.    Performance Statistics from WebUI: 1000 Linked Clones

System Sizing

The reference architecture uses the sizing specifications described in this section.

Virtual Machine Test Image Build

Table 8 summarizes the virtual machine image used to provision desktop sessions in the VMware View environment for linked clones. The image conformed to testing tool standards and was optimized in accordance with the VMware View Optimization Guide for Microsoft Windows 7 and 8.

The VMware OS Optimization Tool was used to make the changes.

The reference architecture and performance tests presented in this document were run on Windows 7 optimized using the VMware OS Optimization Tool.

Table 8.       Virtual Machine Image Attributes

Attribute

Linked Clones

Desktop operating system

Microsoft Windows 7 Enterprise SP1 (32-bit)

Hardware

VMware Virtual Hardware Version 11

vCPU

2

Memory

2048 MB

Memory reserved

2048 MB

Video RAM

35 MB

3D graphics

Off

NIC

1

Virtual network adapter 1

VMXNet3 adapter

Virtual SCSI controller 0

Paravirtual

Virtual disk: VMDK 1

20 GB

Virtual disk: VMDK 2

(nonpersistent disk)

3 GB

Virtual floppy drive 1

Removed

Virtual CD/DVD drive 1

Applications

●  Login VSI 4.1.5 application installation
●  Adobe Acrobat 11
●  Adobe Flash Player 16
●  Doro PDF 1.82
●  FreeMind
●  Microsoft Internet Explorer 11
●  Microsoft Office 2010

VMware tools

Release 10.0.0.3000743

VMware View Agent

Release 6.2.2-3526061

 

Deduplication and Compression Features

You can use built-in deduplication and compression to improve storage efficiency (Figure 26).

Figure 26.    Deduplication and Compression Features

Intel Broadwell Support for Cisco HyperFlex

This section describes changes to our virtualized solution platform with eight Cisco HyperFlex HX220c M4 Nodes with the introduction of the Intel Xeon E5-2600 v4 “Broadwell” processors. This update describes the changes to the solution with a new processor architecture, and is not intended to be a comprehensive guide to every aspect of this solution. It provides guidelines for design and sizing to help you build solution platforms for up to 1000 VMware Horizon 7 virtual desktops.

Reference Configuration

The desktop virtualization performance evaluation of the Cisco HyperFlex HX220c M4S 8-node cluster running the general availability Release 1.7.1 software was conducted following the Cisco Virtual Desktop Infrastructure Test Protocol and Success Criteria for Desktop Virtualization using Login VSI 4.1.4.

A 1000-seat VMware Horizon 7 instantclone environment was provisioned by vmFork technologyHorizon 7 server. The environment was tested running Microsoft Windows 10 and Microsoft Office 2016.

Customers using Microsoft Windows 7 Enterprise SP1 (32-bit) with Microsoft Office 2010 (32-bit) desktops provisioned by Citrix Provisioning Services should expect approximately 25% higher density on the Broadwell architecture, up to 1250 desktops vs 1000 for the Haswell architecture covered earlier in this document.

This testing used the Login VSI 4.1.4.4 knowledge worker workload in benchmark mode:

●   48-minute Login VSI knowledge worker test in benchmark mode (end-user experience during login storm)

The testing monitored performance statistics during a virtual desktop boot storm and then waited for the systems to settle for about 20 minutes. Testing tracked ramp-up, which is the login interval for all 1000 sessions, steady state, in which all sessions are logged on and active and logoff.

Hardware Update

The hardware platform consists of nodes with dual processors and 512GB of memory each. The processors are Intel® Xeon® processor E5-2690v4 CPUs, each with 14 cores and operating at 2.6 GHz. The density and number of memory DIMMs selected are consistent with Intel’s recommendations for optimal performance given the number of memory channels and DIMMs supported by each CPU (Figure 27).

Figure 27.    Figure 27. Processor Configuration

System performance is optimized when the DIMM type and number are the same for each CPU channel. Peak memory performance is at 2133 MHz for up to two DIMMs channel for each CPU. Therefore, a higher-density DIMM is recommended to provide the required amount of memory while also delivering peak memory performance by not expanding to the third memory channel.

Software Update

Cisco HyperFlex release 1.7.1-14835 and Cisco UCS Manager 2.2.7 adds support for B200 M4, C220 M4 and C240 M4 blade and rack server to configure with the Intel® Xeon® Processor E5-2600 v4 series CPUs and 2400MHz memory on 6200 Series fabric interconnects.

New Hardware Support in Release 1.7.1-14835

Release 1.7.1-14835 has Installer only changes to support the following:

●   Cisco HX220c M4, HX240c M4, and B200 M4 (compute-only) shipping with the Intel Xeon Processor E5-2600 v4 series (Broadwell) CPUs on Cisco UCS 6200 Series fabric interconnects.

●   NVIDIA Tesla M6 GPU accelerator for B200 M4 (compute-only) servers

●   NVIDIA Tesla M60 GPU accelerator for C-Series Servers

New Software Features in Release 1.7.1-14835

HX Data Platform Installer is compatible with Cisco UCS Manager 2.2(7c)

●   HX Data Platform: (Please note these are HX Installer only changes) HX Data Platform software version remains at 1.7.1 and existing customers do not need to upgrade the HX Data Platform.

●   UCS Manager: Existing customers would need to upgrade to UCS Manager 2.2(7c). Please see HyperFlex Getting Started Guide for more details on upgrade steps.

Dependencies

Table 9 shows interdependencies between the hardware and versions of Cisco HX Data Platform.

Table 9.       UCSM version and HyperFlex software version dependencies.

Component

UCS Manager version

HX Data Platform Installer Version

HX Data Platform Software Version

Intel v4 Broadwell CPU

2.2(7c)

1.7.3

1.7.3

Intel v4 Broadwell CPU

2.2(7c)

1.7.1-14835

1.7.1

Table 10 shows components and their software version configured for the reference architecture in study.

Table 10.     Main components and their software version

Layer

Device

Image

Computing

Cisco HyperFlex HX220c M4 Node

Cisco UCS 6248UP 48-Port Fabric Interconnect

Release 2.2(7e)

Network

Cisco Nexus 9372 platform switch

Release 6.1(2)I3(1)

Storage

SSD: 1 x 120 GB and 1 x 480 GB SSD drives

HDD: 6 x 1.2-TB HDDs

Software

VMware vSphere ESXi

Release 6.0 U1B

VMware vCenter

Release 6.0 U1B

Cisco HyperFlex Data Platform

Release 1.7.1-14835

VMware Horizon 7

VMware Horizon 7 brings the agility of cloud computing to the desktop by transforming desktops into highly available and agile services delivered from the cloud (Figure 1). View delivers virtual sessions that follow end users across devices and locations. It enables fast, secure access to corporate data across a wide range of devices, including Microsoft Windows, Mac OS, and Linux desktop computers and iOS and Android tablets.

New features included in VMware Horizon 7:

●   Instant Clones

◦     A new type of desktop virtual machines that can be provisioned significantly faster than the traditional View Composer linked clones.

◦     A fully functional desktop can be provisioned in two seconds or less.

◦     Recreating a desktop pool with a new OS image can be accomplished in a fraction of the time it takes a View Composer desktop pool because the parent image can be prepared well ahead of the scheduled time of pool recreation.

◦     Clones are automatically rebalanced across available datastores.

◦     View storage accelerator is automatically enabled.

●   VMware Blast Extreme

◦     VMware Blast Extreme is now fully supported on the Horizon platform.

◦     Administrators can select the VMware Blast display protocol as the default or available protocol for pools, farms, and entitlements.

◦     End users can select the VMware Blast display protocol when connecting to remote desktops and applications.

◦     VMware Blast Extreme features include:

◦     TCP and UDP transport support

◦     H.264 support for the best performance across more devices

◦     Reduced device power consumption for longer battery life

◦     NVIDIA GRID acceleration for more graphical workloads per server, better performance, and a superior remote user experience

●   True SSO

◦     For VMware Identity Manager integration, True SSO streamlines the end-to-end login experience. After user logs in to VMware Identity Manager using a smart card or an RSA SecurID or RADIUS token, users are not required to also enter Active Directory credentials in order to use a remote desktop or application.

◦     Uses a short-lived Horizon virtual certificate to enable a password-free Windows login.

◦     Supports using either a native Horizon Client or HTML Access.

◦     System health status for True SSO appears in the View Administrator dashboard.

◦     Can be used in a single domain, in a single forest with multiple domains, and in a multiple-forest, multiple-domain setup.

●   Horizon 7 for Linux Desktops

◦     Support for SLED 11 SP3/SP4. However, Single Sign-on is not supported.

◦     Support for HTML Access 4.0.0 on Chrome.

◦     Support for CentOS 7.1 Linux distribution.

◦     Support to check the dependency packages unique to a Linux distribution before installing Horizon Agent.

◦     Support to use the Subnet option of /etc/vmware/viewagent-custom.conf to specify the subnet used for Linux desktop connections with multiple subnets connected.

◦     Additional Features

Note:    For performance study update on eight node HX220c cluster configured with Intel E5-2690v4 processors and 32GB @ 2400MHz DIMMs we created floating assigned automated desktop pool provisioned via Horizon 7 Instant-clone configuration.

Performance Testing

The desktop virtualization performance evaluation of the Cisco HyperFlex 1.7.1 running on eight node HX220c-M4S cluster was conducted following the Cisco Virtual Desktop Infrastructure Test Protocol and Success Criteria for Desktop Virtualization using Login VSI 4.1.5. Figure 28 shows the configuration tested.

Figure 28.    Configuration Used for Performance Testing

The Cisco HyperFlex System runs in a cluster on VMware ESXi 6.0 U1b managed by the Cisco HyperFlex plug-in to the VMware vCenter web client. A 1000-seat VMware Horizon 7 floating assignment instant-clone desktop pool running Windows 10 64bit OS with Office 2016 based configuration was tested.

Note:    As a guideline for Intel Broadwell based platform running Windows 7 and Office 2010 we saw density gain of 25% on the same platform running identical test scenario.

48-Minute Standard Benchmark Mode Test Results

LoginVSI tests were executed on 1000 instant-clone desktop VM hosted on eight Cisco HyperFlex HX220c M4S servers configured in HyperFlex cluster with exceptional user performance as represented by the Login VSI Analyzer score and latency values (Figures 29 through 35).

Test result highlights include:

●   0.7 second baseline response time

●   1 second average response time with 1000 desktops running

●   1.7 second maximum response time with 1000 desktops running

●   Average CPU utilization of 85 percent during steady state

●   Average of 300 GB of RAM used out of 512 GB available

●   560MBps peak network utilization per host.

●   Aggregate of 6 milliseconds (ms) of I/O latency per cluster

●   26500 peak I/O operations per second (IOPS) per cluster at steady state

●   400MBps peak throughput per cluster at steady state

Figure 29.    Login VSI Analyzer Chart for 1000 Sessions with End-User Experience (EUX) Response Time
Figure 30.    Host CPU Utilization from VMware ESXTOP: 1000 Instant-Clones, Average CPU Utilization as a Percentage
Figure 31.    Host Memory Use from VMware ESXTOP: 1000 Instant-Clones, Average Memory Use in MB
Figure 32.    Host Network Utilization from VMware ESXTOP: 1000 Instant-Clones, Average Network Utilization in Mbps

Note:    Figures 29 to 33 show behavior of instant clone desktop deployment in which after successful logoff completion floating assigned desktop is deleted. All charts reports the same data from 12:46 PM onwards. Once deletion task is completed Horizon will start provisioning fresh VM from parent VM reside on individual ESXi in a cluster.

Figure 33.    Performance Statistics from Web User Interface (WebUI): 1000 Instant-Clone desktops

Microsoft Windows 10 Virtual Machine Test Image Build

The reference architecture uses the sizing specifications described in this section.

Virtual Machine Test Image Build

Table 11 summarizes the virtual machine image used to provision desktop sessions in the VMware Horizon environment for instant-clones. The image conformed to testing tool standards and was optimized in accordance with the VMware View Optimization Guide for Microsoft Windows 7, 8 and 10.

The VMware OS Optimization Tool was used to make the changes.

The reference architecture and performance tests presented in this document were run on Windows 10 optimized using VMware OS Optimization Tool.

Table 11.     Virtual Machine Image Attributes

Attribute

Instant Clones

Desktop operating system

Microsoft Windows 10 Enterprise (64-bit)

Hardware

VMware Virtual Hardware Version 11

vCPU

2

Memory

2048 MB

Memory reserved

2048 MB

Video RAM

35 MB

3D graphics

Off

NIC

1

Virtual network adapter 1

VMXNet3 adapter

Virtual SCSI controller 0

Paravirtual

Virtual disk: VMDK 1

24 GB

Virtual disk: VMDK 2 (nonpersistent disk)

3 GB

Virtual floppy drive 1

Removed

Virtual CD/DVD drive 1

Applications

●  Login VSI 4.1.5 application installation
●  Adobe Acrobat 11
●  Adobe Flash Player 16
●  Doro PDF 1.82
●  FreeMind
●  Microsoft Internet Explorer
●  Microsoft Office 2016

VMware tools

Release 10.0.0.3000743

VMware View Agent

Release 7.0.1 – 3989057

 

Conclusion

This Cisco HyperFlex solution addresses urgent needs of IT by delivering a platform that is cost effective and simple to deploy and manage. The architecture and approach used provides for a flexible and high-performance system with a familiar and consistent management model from Cisco. In addition, the solution offers numerous enterprise-class data management features to deliver the next-generation hyperconverged system.

Delivering responsive, resilient, high performance VMware Horizon 7 provisioned Windows 10/7 Virtual Machines and Windows Server for hosted Apps or desktops has many advantages for desktop virtualization administrators.

Virtual desktop end-user experience, as measured by the Login VSI tool in benchmark mode, is outstanding with Intel Broadwell E5-2600 v4 processors and Cisco 2400Mhz memory.  In fact, we have set the industry standard in performance for Desktop Virtualization on a hyper-converged platform.

SOURCE: https://www.cisco.com/c/en/us/solutions/collateral/data-center-virtualization/desktop-virtualization-solutions-vmware-horizon-view/whitepaper-c11-737521.html

Leave a Reply

Your email address will not be published. Required fields are marked *