Download VMware vSphere vs. Microsoft Hyper-V: Cost Comparison - Savings with VMware and more Study notes Technology in PDF only on Docsity!
TOTAL COST COMPARISON:
VMWARE VSPHERE VS. MICROSOFT HYPER-V
APRIL 2012
A PRINCIPLED TECHNOLOGIES TEST REPORT
Commissioned by VMware, Inc.
Total cost of ownership (TCO) is the ultimate measure to compare IT infrastructure platforms, as it incorporates the purchase and support costs of the platform along with ongoing operational and management expenses. The operational efficiency built into your software stack can greatly affect your bottom line—once you have procured and implemented your platform, operational costs for administration and maintenance can easily balloon. A solution that streamlines and automates routine maintenance tasks can increase uptime and save an organization time and money. In this study, we use results from the VMware Cost-Per-Application calculator and examine the operational expenses of the two platforms using five scenarios to provide a hypothetical TCO comparison. In our labs at Principled Technologies, we compared the automated administration capabilities of two common virtualization platforms, VMware vSphere® 5 and Microsoft Windows Server® 2008 R2 SP1 Hyper-V™, in several scenarios. It took significantly less time to complete common administrative tasks with the VMware solution, potentially reducing post-acquisition operational expense costs in the five operational tasks we tested by as much as 91 percent over a two-year period compared to the Microsoft solution.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 2
When we combine the operational cost savings VMware vSphere provides with the capital expenses the VMware Cost-Per-Application calculator predicts, we find that VMware virtualization platforms can provide substantially lower two-year total cost of ownership compared to Microsoft platforms.
Figure 1: In the operational scenarios we tested, VMware had 91 percent lower operational costs over a two-year period.
$0 VMware^ Microsoft
$5,
$10,
$15,
$20,
$25,
$30,
$35,
$40,
$45,
Two-year evaluated operating expenses
SELECTING A COMPLETE SOLUTION
When choosing a virtualization platform, considering all costs—both acquisition- related and operational—is essential. An organization must account for not only the cost of acquiring licenses and software, but also the cost associated with the time a system administrator will devote to maintenance and management tasks within each environment. Since system administrator time can be more valuable when used on strategic IT initiatives that deliver a competitive edge for their organization instead of routine maintenance, it is always beneficial to reduce operational administrative costs. As these operational costs add up over time, they can become a significant portion of overall costs for a data center. We discuss both acquisition and operational costs below. For acquisition estimates, we used the VMware Virtualization Cost-Per-Application Calculator on VMware’s Web site at http://www.vmware.com/technology/whyvmware/calculator/.
Acquisition costs
As verified by Principled Technologies’ 2011 testing,^1 VMware vSphere offers significant advantages that can lead to higher VM density than Microsoft Hyper-V.
(^1) http://www.principledtechnologies.com/clients/reports/VMware/vsphere5density0811.pdf
91% lower operational costs with VMware vSphere
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 4
Figure 2 : Using VMware products can lower your operational cost by as much as $37,540 over the course of a two-year period compared to comparable Microsoft offerings.
$
$5,
$10,
$15,
$20,
$25,
$30,
$35,
$40,
Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov
US dollars
Cumulative savings over two years using the VMware
solution
Performing non- disruptive disaster recovery testing
Provisioning new hosts
Isolating a storage- intensive VM
Adding new volumes and redistributing VM storage
Shifting virtual machine workloads for host maintenance
Shifting virtual machine workloads for host maintenance Firmware upgrades, BIOS updates, and hardware replacements often require short periods of server downtime. To perform this routine maintenance, an administrator must first offload the virtual machines running on those servers to other servers to keep the infrastructure running. This time to migrate VMs from source to destination servers requires valuable hands-on time from the administrator; the faster these migrations happen, the better. Figure 3 depicts the live migration process.
Figure 3 : VM live migration time is critical during a server maintenance event.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 5
To test this scenario for both VMware and Microsoft, we placed six VMs, each with 10 GB of RAM, on each server in our three-server cluster and ran a medium database workload on each of the 18 VMs. We then measured the time it took one server in the cluster to enter maintenance mode, evacuate all its VMs to the two remaining servers, and then migrate the VMs back to the original server. We performed these tests using both the VMware solution and the Microsoft solution. We found that the solution running VMware vSphere 5 reduced the time to complete the shifting of the VM workloads by 79 percent over the Microsoft solution. Figures 4 and 5 show the time it took to complete each task needed to perform physical maintenance on a server. We provide further details in Appendix C.
Figure 4 : It took 79 percent less time to shift the VM workloads using the VMware solution than it did with the Microsoft solution. Lower numbers are better.
VMware Microsoft 0
2
4
6
8
10
12
Minutes
Time to shift VM workloads
Task VMware solution Microsoft solution Time to fully migrate all VMs off one node and enter maintenance mode 01:06^ 07: Time to exit maintenance mode 00:01 00: Time to migrate VMs back 01:09 02: Total without boot 02:16 11: Figure 5 : Times, in minutes:seconds, to complete the live migration relating to performing physical maintenance on one server.
Adding new volumes and redistributing VM storage
If your business is growing, the increasing numbers of VMs and data in your environment mean that you will need new storage. System admins must frequently add new storage capacity, which requires them to redistribute existing VM storage to new storage or to re-provision existing storage. Using the available features in each platform,
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 7
each solution to migrate VM storage. Figure 8 and Appendix D provide a breakdown of each task we performed and the time required for completion. We did not measure the time necessary to implement the new tray of storage, as it was the same for both platforms. Nor did we measure the actual storage migration time as we assume administrators would let this operation run automatically.
Figure 7 : It took 95 percent less time to add a new volume and redistribute VM storage using the VMware solution than it did with the Microsoft solution. Lower numbers are better.
VMware (^) Microsoft 0
20
40
60
80
100
120
Minutes
Time to migrate VM storage
VMware solution Microsoft solution Task Time Task Time
- Plan for the brief but inevitable downtime with Quick Storage Migration. We assume 15 minutes of coordination time per VM, and a density of six VMs on the affected volume to be migrated.
- On a host, rescan the iSCSI Software adapter for the new LUN on the new storage tier.
0:02:10 2. On each host, connect to the new LUN using iSCSI initiator. We assume three hosts. 0:01:
- Add the new LUN as a datastore to the cluster. 0:01: 3. Using disk management on one of the hosts, create a new simple volume using the new LUN.
Bring the LUN online on each host. 0:02:
Add the new datastore to the preexisting datastore cluster. 0:00:
Using failover clustering services on the management server, add the disk to the cluster and add it to cluster shared volumes.
- Assess administrator time necessary to manually calculate how many migrations are necessary to balance LUN capacity using the new storage tier. We assume 1 minute per VM, and six VMs on the affected volume.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 8
VMware solution Microsoft solution
- Click “Run Storage DRS” to start the redistribution of the VMs using the new storage tier.
- Using SCVMM and quick storage migration, queue each quick storage migration using the built-in wizard.
Total 0:04:23 Total 1 : 43 : 06 Figure 8 : Times, in hours:minutes:seconds, to complete the tasks relating to adding a new datastore and redistributing VM storage. (=estimated)*
Isolating a storage-intensive VM
Both VMware and Microsoft virtualization solutions implement some degree of resource management when it comes to CPU and RAM. However, when a particular user’s VMs overwhelm storage I/O resources, IT staff must isolate this “noisy neighbor” in order to distribute resources properly for other users. For VMware, this isolation process involves enabling storage I/O control and capping the VM IOPS within the vCenter Server console. As was the case with the previous storage scenario, Hyper-V has no equivalent feature. For Hyper-V to fully isolate the VM, the VM’s virtual disks must be offloaded to different physical storage. Figure 9 shows how VMware Storage I/O control works.
Figure 9 : VMware Storage I/O control easily isolates and caps VMs’ storage bandwidth.
We isolated and redistributed resources from the noisy neighbor using both solutions, and found that it took 97 percent less time to do so using the VMware solution compared to the Microsoft solution (see Figure 10). VMware vSphere Storage I/O Control was able to quickly isolate the user, where Microsoft’s manual isolation approach took significantly longer. We provide the detailed steps we used in Figure 11 and in Appendix E.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 10
VMware solution Microsoft solution online and format on each host in the cluster.
- In Failover Clustering Services, add the new disks as a cluster disk(s). 0:00:
- Add the disk(s) to cluster shared volumes. 0:00:
- Using SCVMM, move the noisy VM(s) to the new disk with the quick storage migration feature.
Total 0:02:05 Total 1:34: Figure 11 : Times, in hours:minutes:seconds, to complete the tasks relating to redistributing resource from a noisy neighbor VM.
Provisioning new hosts
Provisioning new hosts in a data center environment is a constant requirement if your business is growing, or even if your business is simply refreshing your hardware. Each solution has automated tools to accomplish the provisioning task. In our testing, we set up both platforms’ automated solutions: for VMware we used VMware vSphere Auto Deploy (see Figure 12), and for Microsoft Hyper-V we used System Center Configuration Manager 2007 R3 bare metal deployment task sequence.
Figure 12 : VMware Autodeploy quickly deploys new diskless hosts.
Using VMware Auto Deploy provisioned new hosts more quickly than using Microsoft SCCM 2007 R3—by up to 78 percent—and without the use of onboard storage (see Figure 13). We provide the detailed steps we followed in Figure 14 and Appendix F.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 11
Figure 13 : It took 78 percent less time to provision new hosts using the VMware solution than it did with the Microsoft solution. Lower numbers are better.
VMware Microsoft 0
1
2
3
4
5
6
7
8
Minutes
Time to provision new hosts
VMware solution Microsoft solution Task Time Task Time
- Click Apply Host Profile. 0:00:05 1. Enter license and log into the domain. 0:00:
- Answer profile questions. 0:01:03 2. Connect LUNs via iSCSI Initiator. 0:01:
- Wait until host is configured and ready. 0:01:45 3. Bring disks online via Disk management. 0:00:
- Create four new virtual networks for Hyper-V. 0:0^2 :^12
- Join host to the cluster. 0:02: Total 0:0 2 : 53 Total 0:07: Figure 14 : Times, in hours:minutes:seconds, to complete the tasks relating to provisioning new hosts.
Performing non-disruptive disaster recovery testing
We set out to test a non-disruptive disaster recovery plan, where each step of the process causes no downtime, retargeting of production workloads, or production networking changes. For VMware, we used VMware Site Recovery Manager, and for Microsoft we used two distinct site clusters and a manual runbook procedure. We opted not to use a geographically stretched Hyper-V failover cluster, because their distance limitations can make them unsuitable for some use cases and there is no way to perform disaster recovery testing scenarios without disrupting or altering the production workload. In our testing, we measured the time it took to perform a complete non- disruptive disaster recovery test using VMware Site Recovery Manager, then measured or approximated the equivalent actions using the Microsoft solution. For our time calculation scenarios, we assume the organization has five SAN systems and 1,000 VMs,
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 13
VMware solution Microsoft solution
On each host, online the disks.^6 0 : 10 : 37
For each volume, attach to the cluster hosts.^7 0:^2 0:
Run prepared scripts for VM power on and IP addressing. Perform DR testing.^8 0:^3 0:^0 0*
In vCenter Server, within the SRM plug-in, right-click your recovery plan and choose "Cleanup"
0:00:10 8.^ Run prepared scripts to power down VMs. 9 0:0 1 : 0 0*
- Cleanly remove volumes from DR cluster.^10 0:^11 :
- For each volume on each host, offline the disks, and disconnect the disks.^11 0:^01 :^10
- Clean up and revert storage configuration from DR test.^12 0:^11 :
- Revert DNS or WAN for normal operation.^13 0:10:00*
- Unpause SAN replication.^14 0: 00 : 50 Total 1 :00:20 Total 12 : 09 : 07 Figure 1 6 : Times, in hours:minutes:seconds, to complete the tasks relating to provisioning new hosts. (=estimated)*
snapshots for DR testing. We assume 10 volumes per SAN, and two DR SANs; therefore, we multiplied our original time (0:01:06)
6 times 20. We assume 75 of our 1,000 VMs are tier 1 protected VMs. We also assume a host density for Microsoft of 12 VMs per host, which
7 amounts to seven hosts (75/12=6.25, which requires seven hosts). Therefore, we multiplied our original time (0:01:31) by seven. 8 We assume 10 volumes per SAN, and two DR SANs; therefore, we multiplied our original time (0:01:00) by 20. 9 We assume a flat 30-minute cost for this process. 10 We assume a flat 1-minute cost for this process. 11 We assume 10 volumes per SAN, and two DR SANs; therefore, we multiplied our original time (0:00:34) by 20. We assume seven hosts (see footnote 6), sharing two volumes, but each only connecting to one volume. Therefore, we multiplied
12 our original time (0:00:10) by seven. This time will differ by SAN vendor. Our manual process on the Dell EqualLogic storage in our lab was to mimic the automated process that VMware performed. We manually removed the writeable snapshots on the storage, and then demoted volume to a replica set for DR replication. We assume 10 volumes per SAN, and two DR SANs; therefore, we multiplied our original time
13 (0:00:34) by 20. Estimated time to approximate networking staff adjusting configuration on networking hardware. We assume a flat 10-minute
14 cost for this process. We assume two of the five SANs in our sample organization are tier 1 DR SANs that must be paused during the DR test. Therefore, we multiplied our original “unpause” hand timing step (0:00:25) by two.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 14
CALCULATING TWO-YEAR OPERATIONAL COSTS FOR THE SCENARIOS
To calculate the management operational costs of the two solutions, we timed how long it took to perform the tasks in each of our five management scenarios with VMware and Hyper-V. We were able to perform each set of tasks faster with VMware. Figure 17 shows our tested times and the time savings for VMware vs. Hyper-V on the five test scenarios.
Scenario VMware solution^ Microsoft solution^ Savings solution^ with^ VMware
Scenario 1: Shifting virtual machines workloads for host maintenance 0:02:16^ 0:11:05^ 0:08: Scenario 2: Adding new volumes and redistributing VM storage 0:04:23^1 :^43 :^06 1 :^38 :^43 Scenario 3: Isolating a storage-intensive VM 0:02:05 1:34:32 1:32: Scenario 4: Provisioning new hosts 0:02:53 0:07:21 0:04: Scenario 5: Performing non-disruptive disaster recovery testing 1:00:20^12 :^09 :^07 11 :^08 :^47 Figure 17 : Time savings in hours:minutes:seconds for VMware compared to Hyper-V on five test scenarios. Times and savings are for one iteration of each scenario on our tested server. To illustrate how these time savings can affect an organization’s bottom line, we assumed an example environment consisting of 1,000 VMs, with a VM density of 15 VMs per server for VMware vSphere servers, and 12 VMs per server for Microsoft Hyper-V servers. We then calculated the cost savings for an enterprise that chooses VMware vSphere over Microsoft Hyper-V and must repeat many of these scenarios through a typical two-year period. We assumed the tasks would be carried out by a senior system administrator and calculated costs based on that individual’s salary plus benefits.^15 Each minute of that Senior System Administrator’s time is valued at $1.02.
Figure 18 shows the times and time savings in the previous figure multiplied by $1.02.
Scenario VMware cost per iteration^ Microsoft iteration^ cost per
Scenario 1: Shifting virtual machines workloads for host maintenance $2.32^ $11. Scenario 2: Adding new volumes and redistributing VM storage $4.47 $105. Scenario 3: Isolating a storage-intensive VM $2.12 $96. Scenario 4: Provisioning new hosts $2.94 $7. Scenario 5: Performing non-disruptive disaster recovery testing $61.54 $ 743. 70 Figure 18 : Cost savings for VMware for one iteration of each scenario.
(^15) The average national base salary for a senior system administration was $88,599 and total compensation was $126,662 according to salary.com on March 5, 2012. Total compensation includes base salary, employer contributions for bonuses, Social Security, 401k and 401b, disability, healthcare, and pension, and paid time off. We calculated the average cost per minute for a Senior Systems Administrator at that salary at $1.02 based on 52 forty-hour weeks.
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 16
round (84 servers * .33) + round (84 servers * 0.1) = 37 events 37 events * 2 years = 74 events
Savings per two year period = (74 * Microsoft cost) – (60 * VMware cost)
Performing non-disruptive disaster recovery testing
We assume a monthly test of disaster recovery, for 24 events per two-year period. Figure 19 shows the estimated number of events and the subsequent savings per two-year period that a company would realize when choosing VMware and managing these scenarios; that value is the product of the estimated events value multiplied by the savings per iteration value in the previous figure.
Scenario Total events per two-year period^ Savings period^ per^ two-year
Scenario 1: Shifting virtual machines workloads for host maintenance (^) Microsoft:VMware:^568244 $ 6 , 444. 52
Scenario 2: Adding new volumes and redistributing VM storage 120 $ 12 , 082. 80 Scenario 3: Isolating a storage-intensive VM 24 $ 2 , 263. 20
Scenario 4: Provisioning new hosts (^) Microsoft:VMware:^6740 $ 378. 60
Scenario 5: Performing non-disruptive disaster recovery testing 24 $ 16 , 371. 84 Total savings $^37 ,^540.^96 Figure 19 : Estimated operational cost savings based on these scenarios when using VMware vs Microsoft with 1000 VMs over a two year period.
CALCULATING ACQUISITION AND CAPITAL COSTS
We used the VMware Cost-Per-Application Calculator to calculate the acquisition costs of virtualization platforms needed to support a 1,000-VM data center. We used the following as inputs to the calculator: 1,000 VMs, “Typical” workload profile, “Server B” configuration, iSCSI storage, VMware Enterprise Plus edition, use of physical management servers, and average electricity and real estate costs. The VMware Cost- Per-Application Calculator normally factors in a 50 percent VM density advantage for vSphere over Hyper-V, but we use a more conservative 25 percent advantage for VMware (12 VMs per host for Microsoft, 15 VMs per host for VMware). With those assumptions, the VMware Cost-Per-Application Calculator finds that the VMware platform requires 67 vSphere hosts and two vCenter management servers, while the Microsoft platform requires 84 Hyper-V hosts and 11 System Center and SQL Server management servers (based on Microsoft’s documented best practices, see the VMware Cost-Per-Application Calculator methodology paper^16 for references). Additionally, we factor in VMware Site Recovery Manager Standard Edition acquisition cost and two
(^16) http://www.vmware.com/go/costperapp-calc-methods
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 17
years of support for our 75 protected VMs at a cost of $20,768. The calculated costs of hardware (servers, networking, and storage), software (virtualization, management, OS licenses, VMware vCenter Site Recovery Manager) and data center infrastructure with two years of support are as follows: VMware: $2,300, Microsoft: $2,278,
CALCULATING TOTAL COST OF OWNERSHIP
We calculated the two-year total cost of ownership as the sum of the platform acquisition costs generated by the VMware Cost-Per-Application Calculator and the operational costs of the five scenarios we evaluated for our hypothetical 1000-VM data center (see Figure 20). VMware solution Microsoft solution 2-year CAPEX (hardware, software, and support costs) $2,300,768^ $2,278, 2-year operating expenses (from five evaluated scenarios) $3,503^ $41, 2-year TCO $2,304,271 $2,319, Figure 20 : Two-year total cost of ownership for the two solutions.
The results show that VMware’s lower operational costs can lead to a lower TCO for the VMware platform compared to Microsoft, when considering the five scenarios we tested. However, these five scenarios are only a small subset of the typical operational requirements of an organization, and other studies of cross-industry IT spending show that annual operational expenses are over two times capital expenses.^17 This means the impact of operational cost savings for platform technologies such as virtualization may be multiplied well beyond the totals for the five common tasks we include in this analysis. Therefore, organizations may find that additional features of VMware vSphere 5—such as a single unified management interface in vCenter, hot-add CPU for guest VMs, VM-to-host and VM-to-VM affinity capabilities, and VM storage tier placement automation—could lead to further operational time savings.
WHAT WE TESTED
About VMware vSphere 5
vSphere 5 is the latest virtualization platform from VMware. vSphere 5 allows companies to virtualize their server, storage, and networking resources, achieving significant consolidation ratios, all while gaining significant management time savings as we demonstrate in this paper. To learn more about VMware vSphere 5, visit http://www.vmware.com/products/vsphere/overview.html.
(^17) http://storage.networksasia.net/content/migrating-cloud-beware-prickly-financial-situations
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 19
APPENDIX A – SERVER AND STORAGE CONFIGURATION INFORMATION
Figures 21 and 22 present configuration information for our test servers and storage.
System 3 x Dell PowerEdge R710 servers Power supplies Total number 2 Vendor and model number Dell Inc. N870P-S Wattage of each (W) 870 Cooling fans Total number 5 Vendor and model number Nidec UltraFlo™ RK385-A Dimensions (h x w) of each 2.5 x2. Volts 12 Amps 1. General Number of processor packages 2 Number of cores per processor 6 Number of hardware threads per core 2 CPU Vendor Intel® Name Xeon® Model number X Stepping B Socket type FCLGA 1366 Core frequency (GHz) 2. Bus frequency 6.4 GT/s L1 cache 32 KB + 32 KB (per core) L2 cache 6 x 256 KB (per core) L3 cache 12 MB (shared) Platform Vendor and model number Dell PowerEdge R Motherboard model number OYDJK BIOS name and version Dell Inc. 6.0. BIOS settings Default Memory module(s) Total RAM in system (GB) 96 Vendor and model number M393B1K70BH1-CH Type PC3- Speed (MHz) 1, Speed running in the system (MHz) 1, Timing/Latency (tCL-tRCD-tRP-tRASmin) 9-9-9- Size (GB) 8 Number of RAM module(s) 12 Chip organization Double-sided Rank Dual
Management cost comparison: VMware vSphere vs. A Principled Technologies test report 20
System 3 x Dell PowerEdge R710 servers Microsoft OS Name Windows Server 2008 R2 SP Build number 7601 File system NTFS Kernel ACPI x64-based PC Language English VMware OS Name VMware vSphere 5.0. Build number 469512 File system VMFS Kernel 5.0. Language English Graphics Vendor and model number Matrox® MGA-G200ew Graphics memory (MB) 8 RAID controller Vendor and model number PERC 6/i Firmware version 6.3.0- 0001 Cache size (MB) 256 Hard drives Vendor and model number Dell ST9146852SS Number of drives 4 Size (GB) 146 RPM 15, Type SAS Onboard Ethernet adapter Vendor and model number Broadcom® NetXtreme® II BCM5709 Gigabit Ethernet Type Integrated 10Gb Fibre adapter for vMotion scenario Vendor and model number Intel Ethernet Server Adapter X520-SR Type Discrete Quad-port Ethernet adapter for Storage I/O Control scenario Vendor and model number Intel PRO/1000 Quad Port LP SVR Adapter Type Discrete Optical drive(s) Vendor and model number TEAC DV28SV Type DVD-ROM USB ports Number 6 Type 2. Figure 21 : Detailed configuration information for our test servers.
