TECHNICAL WHITE PAPER – SEPTEMBER 2016
WINDOWS 10 OPTIMIZATION TEST RESULTS View in VMware Horizon 7
WINDOWS 10 OPTIMIZATION TEST RESULTS
Table of Contents Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Audience. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Horizon 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Testing Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 VMware OSOT Fling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Login VSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Optimizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Main Optimizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Additional Optimizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Benchmark Test Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 vCenter Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 About the Author. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
T E C H N I C A L W H I T E PA P E R | 2
WINDOWS 10 OPTIMIZATION TEST RESULTS
Introduction This paper presents crucial test results that demonstrate the benefits of the VMware Operating System Optimization Tool (OSOT) Fling for View desktops in VMware Horizon® 7. It compares optimized and non-optimized versions of Windows 10 and optimized Windows 7, highlighting the resources needed to upgrade to Windows 10 from Windows 7 in a virtualized environment. Additional optimizations not handled by the Optimization Fling or described in the VMware Windows Operating System Optimization Tool Guide are also included.
Audience This paper was written specifically for data center administrators and IT personnel who want to upgrade from Windows 7 to Windows 10 or to optimize Windows 10 operating systems for View virtual desktops in Horizon 7. Architects and others interested in Horizon 7 or in improving the performance of Windows desktops may also find it useful.
Horizon 7 Horizon 7 provides centralized image management for virtualized or hosted desktops and applications, which can be delivered to end users through a single platform. Desktop and application services— including RDS-hosted apps, packaged apps with VMware ThinApp®, SaaS apps, and even virtualized apps from Citrix—can all be accessed from a single, unified workspace. Horizon 7 supports both Windows and Linux-based desktops, including RHEL, Ubuntu, CentOS, and NeoKylin. This paper, however, focuses on Windows 7 and Windows 10, especially the performance differences between optimized and non-optimized versions.
Testing Details The following sections discuss the testing performed, including the VMware OSOT Fling, additional optimizations, testing methods, hardware, and results.
VMware OSOT Fling The free OSOT Fling makes it easy to apply configuration settings to desktops and servers. It includes settings to optimize the following Windows operating systems for both desktops and servers that reside in the data center and in the cloud: • Windows 7 • Windows 8 • Windows 8.1 • Windows 10 • Windows Server 2008 (including R2) • Windows Server 2012 (including R2) The OSOT Fling optimizes RDSH servers for VMware Horizon 6, Horizon 7, and VMware Horizon Air™.
T E C H N I C A L W H I T E PA P E R | 3
WINDOWS 10 OPTIMIZATION TEST RESULTS
Login VSI All performance testing documented in this paper used the Login VSI benchmarking tool, the industrystandard load-testing solution for centralized virtualized desktop environments. Login VSI measures the total response time of several specific user operations performed within a desktop workload in a scripted loop. The baseline measures the response time, in milliseconds (ms), of specific operations performed in the desktop workload. The Login VSI test suite defines several types of workloads to simulate various types of users, as shown in Table 1. WO RK LOAD
P R OFILE WEIGHT
v CPU
APPS
V IDEO
Task Worker
Light
1
2–3
No
Office Worker
Medium
1
4–6
240p
Knowledge Worker
Medium
2
4–7
360p
Power User
Heavy
2–4
5–9
720p
Table 1: Login VSI Workloads
For the tests reported in this paper, we used the Knowledge Worker workload, which simulates how resources are used in VDI environments that use two virtual CPUs in their virtual machines. The Knowledge Worker workload includes the following applications: • Microsoft Outlook • Microsoft Internet Explorer • Microsoft Word • Adobe Acrobat Reader • Microsoft PowerPoint • Microsoft Excel • Adobe Flash Player • Java 7 • FreeMind • Login VSI photo viewer • Doro PDF Writer The Login VSI Benchmark mode predefines and constrains the launch window to 2880 seconds (48 minutes). This gives a normalized baseline to compare all tests and allows for the workload to complete a full 48-minute cycle. There is only one phase to this profile. A user starts a remote desktop session on a VM and begins the workload. At the end of the workload cycle, a benchmark, or baseline, score is calculated. The lower the score, the better the result or performance.
T E C H N I C A L W H I T E PA P E R | 4
WINDOWS 10 OPTIMIZATION TEST RESULTS
M E AS URE M ENT I D
M EAS UR EMEN T AC TI O N
MEASU R EMEN T ACTION DETAILED
MEASU R ES- REL AT ED R ESOU R C E
WSLD
Start Microsoft Word and load a random document.
Start and load a local random document file from the content pool.
CPU, RAM, and I/O
NSLD
Start VSI Notepad and load a random document.
Start and load a local random text file from the content pool.
CPU and I/O
WFO
Open a file in VSI Notepad.
Ctrl+O
CPU, RAM, and I/O
NFP
Open Print in VSI Notepad.
Ctrl+P
CPU
ZHC
Compress files with high compression.
Compress a local random PST file from the content pool (5 MB).
CPU
ZNC
Compress files with no compression.
Compress a local random PST file from the content pool (5 MB).
I/O
Table 2: Actions Taken During a Knowledge Worker Test Session
We took a methodical approach to all test operations. Tests were performed in an environment where the only services and servers used were those needed by Login VSI. No core infrastructure servers, such as Active Directory and SQL servers, were in use by any other applications. After deploying all components and applications required for the Login VSI workload on each desktop, we conducted functional validation, using a very simple test methodology.
Methodology To begin, we ran a baseline Login VSI Knowledge Worker benchmark test, with the required applications, to provide a simple baseline showing performance and scale information for a traditional, non-optimized Windows 7 desktop. Next, we ran the same test on an optimized Windows 7 desktop to see the difference in baseline performance. We then ran the Login VSI Knowledge Worker test on a Windows 7 optimized desktop, Windows 10 non-optimized desktop, and Windows 10 optimized desktop. The measured operations in these benchmarks touch different subsystems, such as CPU (user and kernel), memory, disk, the OS, the application, and print. The operations typically execute quickly. When these operations consistently take a long time to execute, the system becomes saturated because of excessive queuing, and the average response times escalate. This effect is reflected in the baseline score and is also visible to end users. For each desktop, we ran the test three times to ensure consistency.
T E C H N I C A L W H I T E PA P E R | 5
WINDOWS 10 OPTIMIZATION TEST RESULTS
Hardware Each host used for testing was a Dell PowerEdge R730 with 28 cores at 2.29 GHz with 262 GB of memory, and an EMC XtremIO all-flash array for storage.
Table 3: Host Specifications
The XtremIO storage array is an all-flash system, based on a scale-out architecture that uses building blocks called X-Bricks, which can be clustered to increase performance and capacity as required. It is controlled by a standalone, dedicated Linux-based server, called the XtremIO Management Server (XMS). An XMS host, which can be either a physical or a virtual server, can manage multiple XtremIO clusters. An array continues operating if it is disconnected from the XMS, but cannot be configured or monitored.
Optimizations We used all the available optimizations in the VMware OSOT Fling. You can choose not to apply all optimizations, but your results may vary. If you do not apply all of the optimizations recommended, at least apply the main optimizations. Main Optimizations
These are some of the basic optimizations that were applied: • Adjust Windows for best performance. • Turn off all unneeded visual effects. • Set unused services to either Manual or Disabled. • Disable programs at startup. • Use High Performance Power plan. • Speed up the menu show time.
T E C H N I C A L W H I T E PA P E R | 6
WINDOWS 10 OPTIMIZATION TEST RESULTS
As shown in Figure 1, however, far more optimizations are applied by the VMware OSOT Fling.
Figure 1: Some Optimizations Applied by OSOT Fling
Additional Optimizations
Whether you choose to use the OSOT Fling or not, the following optional optimizations are strongly suggested: • Install the latest Windows updates. • Enable write caching for storage devices. • Move the page file to separate disk. • Uninstall any unneeded or unused programs. • Disable or uninstall all Internet Explorer add-ons and toolbars. • Keep the index enabled and updated (or disable the index). • Change how long notifications stay open. • Adjust privacy settings to not share. • Speed up Windows shutdown time.
T E C H N I C A L W H I T E PA P E R | 7
WINDOWS 10 OPTIMIZATION TEST RESULTS
You can set and adjust the following values for how long Windows waits for hung programs and other programs to save data and close before shutting down: ––AutoEndTasks ––HungAppTimeout ––WaitToKillAppTimeout ––WaitToKillServiceTimeout For each host, make sure the BIOS Power Management Policy settings are set for High performance.
Figure 2: BIOS Power Management Settings
Benchmark Test Results We used the Login VSI 4.0 tool to run a representative workload on virtual desktops running each of the operating system versions to get a benchmark or baseline score. VSIbase is a baseline score that reflects the response time of specific operations performed in the desktop workload when there is little or no stress on the system. A low baseline indicates a better user experience, resulting from applications responding faster in the environment. VS I BAS E S CO RE ( I N M I CROS ECON DS)
PER FOR MAN CE
0–99 ms
Excellent
800–1399 ms
Very Good
1400–1999 ms
Good
1999–9999 ms
Reasonable/Poor
Table 4: VSIbase Scoring for This Test Environment
T E C H N I C A L W H I T E PA P E R | 8
WINDOWS 10 OPTIMIZATION TEST RESULTS
For this environment, a score of 1627 was considered a good score. A score less than 1399 is a very good score compared to our baseline and shows better performance.
Optimization Results With our testing we saw an 18–25 percent increase in performance with Windows 10 optimized as compared to Windows 10 non-optimized and Windows 7 optimized, indicating better performance and a better overall user experience.
Figure 3: Graph of LSI Benchmark Scores
WI ND OWS 7 O P TI M I Z E D
W IN DOWS 10 N ON -OPTIMIZED
W IN DOWS 10 OPTIMIZED
Test run 1: 1610 ms
Test run 1: 1685 ms
Test run 1: 1311 ms
Test run 2: 1605 ms
Test run 2: 1609 ms
Test run 2: 1310 ms
Test run 3: 1624 ms
Test run 3: 1633 ms
Test run 3: 1231 ms
Table 5: Optimization Test Results
T E C H N I C A L W H I T E PA P E R | 9
WINDOWS 10 OPTIMIZATION TEST RESULTS
vCenter Data The vCenter data below also confirms that Windows 10 optimized had considerably less CPU usage, memory usage, and disk latency than Windows 10 non-optimized. CPU average usage was approximately 30 percent lower, and disk latency was lower by a little more than 50 percent. Although Windows 10 optimized and Windows 7 optimized showed similar results, the overall CPU, memory, and disk latency were lower on the Windows 10 optimized virtual machine. The Windows 7 optimized VM also had a considerably longer login time, which contributed to its higher (that is, worse) benchmark scores. The vCenter data scores are shown in the following graphs: 100
1500
75
1000
50
500
25
0 10:00 PM 10:05 PM 10:10 PM 10:15 PM 10:20 PM 10:25 PM 10:30 PM 10:35 PM 10:40 PM 10:45 PM 10:50 PM 10:55 PM Time
0
Percent
MHz
2000
Performance Chart Legend Key
Object Win7-VSI-Opt
Measurement Usage in MHz
Rollup Average
Units MHz
0
Usage in MHz
Average
MHz
Win7-VSI-Opt
Usage
Average
Percent
1
Usage in MHz
Average
MHz
Latest 24
Maximum 1764
Minimum 24
Average 478.961 239.478
9
897
9
0.4
28.83
0.4
7.832
9
792
9
190.239
Figure 4: Windows 7 Optimized CPU Utilization
T E C H N I C A L W H I T E PA P E R | 1 0
WINDOWS 10 OPTIMIZATION TEST RESULTS
5000000
3750000 Kilobytes
2500000
1250000
0
10:05 PM 10:10 PM 10:15 PM 10:20 PM 10:25 PM 10:30 PM 10:35 PM 10:40 PM 10:45 PM 10:50 PM 10:55 PM 11:00 PM Time
Performance Chart Legend Key
Object Win7-VSI-Opt
Measurement Granted
Rollup Average
Units Kilobytes
Latest 4130816
Maximum 4130816
Minimum 3960896
Average 4090785.0
Win7-VSI-Opt
Active
Average
Kilobytes
1468004
4110416
754972
1411381.2
Win7-VSI-Opt
Balloon
Average
Kilobytes
0
0
0
0
Win7-VSI-Opt
Consumed
Average
Kilobytes
4130816
4130816
3960892
4090784.5
Figure 5: Windows 7 Optimized Memory Utilization
10
7.5 Millisecond
5
2.5
0
10:05 PM 10:10 PM 10:15 PM 10:20 PM 10:25 PM 10:30 PM 10:35 PM 10:40 PM 10:45 PM 10:50 PM 10:55 PM 11:00 PM Time
Performance Chart Legend Key
Object Win7-VSI-Opt
Measurement Highest latency
Rollup Latest
Units Millisecond
Latest 0
Maximum 8
Minimum 0
Average 0.467
Figure 6: Windows 7 Optimized Disk Latency
T E C H N I C A L W H I T E PA P E R | 1 1
WINDOWS 10 OPTIMIZATION TEST RESULTS
100
4500
75
3000
50
1500
25
0 4:15 PM
4:20 PM
4:25 PM
4:30 PM
4:35 PM
4:40 PM
4:45 PM Time
4:50 PM
4:55 PM
5:00 PM
5:05 PM
5:10 PM
Percent
MHz
6000
0
Performance Chart Legend Key
Object Win10-VSI
Measurement Usage in MHz
Rollup Average
Units MHz
Latest 546
Maximum 5631
1
Usage in MHz
Average
0
Usage in MHz
Average
Win10-VSI
Usage
Average
Minimum Average 96 1026.589
MHz
267
2813
33
484.161
MHz
268
2719
55
490.117
Percent
8.93
92.05
1.57
16.784
Figure 7: Windows 10 Non-Optimized CPU Utilization
4000000
3000000 Kilobytes
2000000
1000000
0
4:20 PM
4:25 PM
4:30 PM
4:35 PM
4:40 PM
4:45 PM 4:50 PM Time
4:55 PM
5:00 PM
5:05 PM
5:10 PM
5:15 PM
Performance Chart Legend Key
Object Win10-VSI
Measurement Granted
Rollup Average
Units Kilobytes
Latest 3987440
Maximum 3987440
Minimum 1834980
Average 3419297.8
Win10-VSI
Active
Average
Kilobytes
1216348
Win10-VSI
Consumed
Average
Kilobytes
3987436
3900700
880800
1439343.4
3987436
1834980
Win10-VSI
Balloon
Average
Kilobytes
0
3419274.6
0
0
0
Figure 8: Windows 10 Non-Optimized Memory Utilization
T E C H N I C A L W H I T E PA P E R | 1 2
WINDOWS 10 OPTIMIZATION TEST RESULTS
6
4.5 Millisecond
3
1.5
0
4:20 PM
4:25 PM
4:30 PM
4:35 PM
4:40 PM
4:45 PM Time
4:50 PM
4:55 PM
5:00 PM
5:05 PM
5:10 PM
5:15 PM
Performance Chart Legend Key
Object Win10-VSI
Measurement Highest latency
Rollup Latest
Units Millisecond
Latest 0
Maximum 5
Minimum 0
Average 0.739
Figure 9: Windows 10 Non-Optimized Disk Latency
100
3000
75
2000
50
1000
25
0
5:20 PM
5:25 PM
5:30 PM
5:35 PM
5:40 PM
5:45 PM
5:50 PM Time
5:55 PM
6:00 PM
6:05 PM
6:10 PM
6:15 PM
Percent
MHz
4000
0
Performance Chart Legend Key
Object Measurement Win10-VSI-Opt Usage in MHz
Rollup Average
Units MHz
0
Usage in MHz
Average
1
Usage in MHz
Average Average
Win10-VSI-Opt Usage
Latest 593
Maximum 3598
Minimum 61
Average 677.439
MHz
299
1716
0
328.402
MHz
276
1761
0
317.575
Percent
9.7
58.82
1.01
11.076
Figure 10: Windows 10 Optimized CPU Utilization
T E C H N I C A L W H I T E PA P E R | 1 3
WINDOWS 10 OPTIMIZATION TEST RESULTS
4000000
3000000 Kilobytes
2000000
1000000
0 5:20 PM
5:25 PM
5:30 PM
5:35 PM
5:40 PM
5:45 PM
5:50 PM Time
5:55 PM
6:00 PM
6:05 PM
6:10 PM
6:15 PM
Performance Chart Legend Key
Object Measurement Win10-VSI-Opt Granted
Rollup Average
Units Kilobytes
Latest 3067904
Maximum 3489792
Minimum 1200044
Average 2745141.7
Win10-VSI-Opt Active
Average
Kilobytes
1048576
Win10-VSI-Opt Consumed
Average
Kilobytes
3067804
3900700
335544
1346369.7
3489792
1200044
Win10-VSI-Opt Balloon
Average
Kilobytes
0
2745068.8
0
0
0
Figure 11: Windows 10 Optimized Memory Utilization
6
4.5 Millisecond
3
1.5
0 5:20 PM
5:25 PM
5:30 PM
5:35 PM
5:40 PM
5:45 PM
5:50 PM Time
Units Millisecond
Latest 2
5:55 PM
6:00 PM
6:05 PM
6:10 PM
6:15 PM
Performance Chart Legend Key
Object Measurement Win10-VSI-Opt Highest latency
Rollup Latest
Maximum 4
Minimum 0
Average 0.339
Figure 12: Windows 10 Optimized Disk Latency
T E C H N I C A L W H I T E PA P E R | 1 4
WINDOWS 10 OPTIMIZATION TEST RESULTS
Conclusion Windows 10 virtual desktops showed a considerable improvement in performance—around 30 percent—when optimized with the OSOT Fling. Non-optimized Windows 10 desktops showed a similar performance improvement compared to optimized Window 7 desktops, so the contrast between Windows 7, whether optimized or not, and Windows 10 optimized with OSOT is truly dramatic. The OSOT Fling also reduced the amount of memory and storage I/O used by each desktop. A reduction of this magnitude can improve resource utilization—or resource recovery—by enabling each host to support a larger number of virtual desktops. We did not calculate or predict the cost or resource savings because these must vary for each implementation, but a performance improvement in the reported range should be reason enough to use the OSOT Fling.
T E C H N I C A L W H I T E PA P E R | 1 5
WINDOWS 10 OPTIMIZATION TEST RESULTS
About the Author Chris White is an Architect on the VMware End-User-Computing Technical-Marketing Center of Excellence team. He wishes to thank Gary Sloane, VMware Consulting Editor, for many suggestions and contributions to this paper.
T E C H N I C A L W H I T E PA P E R | 1 6
VMware, Inc. 3401 Hillview Avenue Palo Alto CA 94304 USA Tel 877-486-9273 Fax 650-427-5001 www.vmware.com
Copyright © 2016 VMware, Inc. All rights reserved. This product is protected by U.S. and international copyright and intellectual property laws. VMware products are covered by one or more patents listed at http://www.vmware.com/go/patents. VMware is a registered trademark or trademark of VMware, Inc. in the United States and/or other jurisdictions. All other marks and names mentioned herein may be trademarks of their respective companies. Item No: 4687-WP-APPVOLUMESREVIEWERSGUIDE-USLET-20160915-WEB 9/16