Already a member?
Sign in
Hyper-V Performance on R900 with Six-Core Intel Xeon Whitepaper
Information Technology organizations are finding that combining today’s industry standard servers based on multi-core processors with server virtualization can create highly efficient solutions. A series of tests were run comparing a Dell™
PowerEdge™ R900 server using new Intel six-core processors against previous generation quad-core processor servers. Microsoft Windows Server 2008’s server virtualization technology Hyper-V was used as the basis with many Virtual
Machines running on each of the servers tested. The results of these tests showed that the Dell PowerEdge with Intel six-core processors was able to outperform quad-core processor based servers, with increased performance per watt efficiency.
Performance Advantage of Dell PowerEdge R900 with Six-Core Intel Processors Running Hyper-V - This is a link to the complete whitepaper in PDF form. This page on DellTechcenter provides the highlights from the paper and a place for comments and questions about the paper.
Storage
The PowerEdge R900 and HP ProLiant DL585 G2 were connected to a dedicated Ethernet based iSCSI storage area network (SAN) with a dual-port Intel gigabit PCI Express NIC. The storage utilized was four Dell EqualLogic™ PS5000XV arrays each with 16 146 GB, 15,000 rpm disks, for a total of 64 disks. Dell PowerConnect 5448 gigabit Ethernet switches were used to for the iSCSI SAN. The EqualLogic Host Integration Kit’s Remote Configuration was used to configure the iSCSI settings such that only the iSCSI subnet and NICs would be used for iSCSI traffic. The iSCSI software initiator included with Windows Server 2008 was used to connect via two NIC ports to the EqualLogic storage.
A RAID-10 storage pool with all four EqualLogic PS5000XV arrays was configured and eight 200 GB volumes were created in this storage pool. With EqualLogic storage arrays the volumes are spread across the member arrays automatically to optimize performance. During this testing these volumes were the only active volumes, so they were spread evenly across the member arrays. Four volumes were assigned to the R900 and four to the DL585. VMs were created and evenly divided across the volumes.
VM Testing and Configuration
To compare the relative performance of the PowerEdge R900 and the HP DL585 G2 servers, a set of Windows Server 2008 virtual machines was created. An initial VM was created and then installed with x64 Windows Server 2008, and x64 SQL Server 2005 with service pack 2. The Hyper-V Integration Services were also installed in the VM which has drivers optimized for Hyper-V VMs. The medium size Dell DVD Store (http://www.delltechcenter.com/page/DVD+Store ) database of approximately 1GB was then created using the scripts included with the DVD Store kit. The DVD Store database simulates the database back end of a simple Web-based storefront. The database size is as setup for this test is small (approximately 1 GB), and representative of a database used for development or testing.
The initial VM was then replicated by using sysprep to first prepare the virtual disk for replication and then doing a simple file copy of the virtual hard disk. The virtual hard disks were evenly distributed across the LUNs assigned to the server on the EqualLogic arrays. To complete the VM replication, new VMs were created and configured to use the new virtual hard disks. Table 3 shows the configuration for the VMs in the test environment.
Performance and Power Consumption. Performance with a virtualized server can be measured in two components. The first is a sizing or capacity which is the number of VMs that can be supported. This immediately leads to the second component, which is the aggregate performance that those VMs are able to achieve. The performance results from the testing are presented both in number of VMs and the sum of the associated performance, Orders Per Minute (OPM) of those VMs.
The power consumption of systems has become a real issue for customers and needs to be considered as part of the overall server performance. An important factor when measuring power consumption is the configuration and load level of the server. The power numbers recorded and reported here are heavily dependent on the amount of RAM, number of PCI adapters, number of internal disks, and the amount of load on the system. All of the configuration options were kept as equal as possible between the systems for this reason.
The difference in the number of VMs and the OPM performance metric indicated the relative difference in performance. The test team calculated the performance results for the SQL Server 2005 VMs by totaling the OPM from all the VMs running in the test environment. Table 4 summarizes the performance results including the power consumption for the three servers tested.
Table 4: Workload performance results for each server in the test environment
These results show that the Dell R900 with six-core Intel Xeon 7450 processors provided a 27% performance advantage over the HP DL585 G2 with AMD Opteron quad-core processors and a 6% advantage in performance per watt. A second key finding is that by simply upgrading the processors in the R900 the performance increased by approximately 18% while power consumption decreased by 11% resulting in a 33% improvement in performance per watt.
The complete whitepaper is available in PDF form. Please use this page to ask questions / discuss.
Microsoft Virtualization Team Blog - Hyper-V Performance Scales Well in 24-Core Dell Server
Virtualization Frontier - Hyper-V Performance with 24-Core R900
Virtualization.info - Benchmarks: Hyper-V performance on Dell R900 with Quad-Core and Six-Core Intel Xeon
Gabes Virtual World - Unbelievable Hyper-V Performance on Dell R900
PowerEdge™ R900 server using new Intel six-core processors against previous generation quad-core processor servers. Microsoft Windows Server 2008’s server virtualization technology Hyper-V was used as the basis with many Virtual
Machines running on each of the servers tested. The results of these tests showed that the Dell PowerEdge with Intel six-core processors was able to outperform quad-core processor based servers, with increased performance per watt efficiency.
Performance Advantage of Dell PowerEdge R900 with Six-Core Intel Processors Running Hyper-V - This is a link to the complete whitepaper in PDF form. This page on DellTechcenter provides the highlights from the paper and a place for comments and questions about the paper.
Hardware
Servers| HP ProLiant DL585 G2 | PowerEdge R900 Quad-Core | PowerEdge R900 Six-Core | |
| Virtualization software | Microsoft Windows Server 2008 Enterprise with Hyper-V 1.0 | Microsoft Windows Server 2008 Enterprise with Hyper-V 1.0 | Microsoft Windows Server 2008 Enterprise with Hyper-V 1.0 |
| Processor | Four quad-core AMD Opteron 2.4 GHz 8358 SE processors with 512L L2 and 2 MB L3 cache | Four quad-core Intel X7350 Xeon processors at 2.93 GHz with 4MB L2 Cache Per Core Pair | Four six-core Intel E7450 Xeon processors at 2.4 GHz with 3 MB L2 Cache per Core Pair and 12MB L3 Shared Cache |
| HyperTransport / Frontside bus | 1 GHz | 1,066 MHz | 1,066 MHz |
| Memory | 128 GB (32x 667 MHz PC2-5300 DDR2 4GB DIMMs) | 128 GB (32x 667 MHz fully buffered 4GB DIMMs) | 128 GB (32x 667 MHz fully buffered 4GB DIMMs) |
| Internal disks | Two Serial Attached SCSI (SAS) 73 GB, 15,000 rpm drives | Two Serial Attached SCSI (SAS) 146 GB, 15,000 rpm drives | Two Serial Attached SCSI (SAS) 146 GB, 15,000 rpm drives |
| Network interface card (NIC) | Four 10/100/1,000 Mbps internal NICs | Four 10/100/1,000 Mbps internal NICs | Four 10/100/1,000 Mbps internal NICs |
| Disk controller | Smart Array P400 | PERC 6/i | PERC 6/i |
Storage
The PowerEdge R900 and HP ProLiant DL585 G2 were connected to a dedicated Ethernet based iSCSI storage area network (SAN) with a dual-port Intel gigabit PCI Express NIC. The storage utilized was four Dell EqualLogic™ PS5000XV arrays each with 16 146 GB, 15,000 rpm disks, for a total of 64 disks. Dell PowerConnect 5448 gigabit Ethernet switches were used to for the iSCSI SAN. The EqualLogic Host Integration Kit’s Remote Configuration was used to configure the iSCSI settings such that only the iSCSI subnet and NICs would be used for iSCSI traffic. The iSCSI software initiator included with Windows Server 2008 was used to connect via two NIC ports to the EqualLogic storage.
A RAID-10 storage pool with all four EqualLogic PS5000XV arrays was configured and eight 200 GB volumes were created in this storage pool. With EqualLogic storage arrays the volumes are spread across the member arrays automatically to optimize performance. During this testing these volumes were the only active volumes, so they were spread evenly across the member arrays. Four volumes were assigned to the R900 and four to the DL585. VMs were created and evenly divided across the volumes.
| Dell/EqualLogic PS5000XV | |
| iSCSI Connections | 3x 1 Gb/s per Array |
| Arrays | 4x Dual Controller PS5000XV Arrays Each with 16x 146 GB 15,000 rpm disks (Total of 64 Disks) |
| Storage Pool | 1 RAID-10 (With all four PS5000XV Arrays as Members) |
| Volumes (LUNs) | Eight 200 GB (four assigned to each server) |
Testing Configuration
In order to configure the systems to run Hyper-V both test servers were installed with Windows Server 2008 x64 Enterprise Edition. The Hyper-V role was then added to enable the hypervisor. The version of Hyper-V that shipped with Windows Server 2008 was a beta so it was necessary to then apply the Hyper-V update for Windows Server 2008 patch, available from Microsoft’s website, which upgrades Hyper-V to the final version.VM Testing and Configuration
To compare the relative performance of the PowerEdge R900 and the HP DL585 G2 servers, a set of Windows Server 2008 virtual machines was created. An initial VM was created and then installed with x64 Windows Server 2008, and x64 SQL Server 2005 with service pack 2. The Hyper-V Integration Services were also installed in the VM which has drivers optimized for Hyper-V VMs. The medium size Dell DVD Store (http://www.delltechcenter.com/page/DVD+Store ) database of approximately 1GB was then created using the scripts included with the DVD Store kit. The DVD Store database simulates the database back end of a simple Web-based storefront. The database size is as setup for this test is small (approximately 1 GB), and representative of a database used for development or testing.
The initial VM was then replicated by using sysprep to first prepare the virtual disk for replication and then doing a simple file copy of the virtual hard disk. The virtual hard disks were evenly distributed across the LUNs assigned to the server on the EqualLogic arrays. To complete the VM replication, new VMs were created and configured to use the new virtual hard disks. Table 3 shows the configuration for the VMs in the test environment.
| Operating System | Workload | Memory | Disk | Number of Virtual NICs | Number of virtual processors |
| Windows Server 2008 x64 | Microsoft SQL Server 2005 x64 | 2048 MB | 18 GB | 1 | 1 |
Performance and Power Consumption Testing Results
The VMs were run on the servers in successive tests, adding VMs in each round as described the previous section, until total CPU utilization on the server reached 80%. A power meter attached to the servers measured the power consumption during these tests. Combining both the performance and power numbers allowed for a final calculation of performance per watt.Performance and Power Consumption. Performance with a virtualized server can be measured in two components. The first is a sizing or capacity which is the number of VMs that can be supported. This immediately leads to the second component, which is the aggregate performance that those VMs are able to achieve. The performance results from the testing are presented both in number of VMs and the sum of the associated performance, Orders Per Minute (OPM) of those VMs.
The power consumption of systems has become a real issue for customers and needs to be considered as part of the overall server performance. An important factor when measuring power consumption is the configuration and load level of the server. The power numbers recorded and reported here are heavily dependent on the amount of RAM, number of PCI adapters, number of internal disks, and the amount of load on the system. All of the configuration options were kept as equal as possible between the systems for this reason.
The difference in the number of VMs and the OPM performance metric indicated the relative difference in performance. The test team calculated the performance results for the SQL Server 2005 VMs by totaling the OPM from all the VMs running in the test environment. Table 4 summarizes the performance results including the power consumption for the three servers tested.
| Number of VMs | Total Server CPU Utilization | Performance in Orders Per Minute (OPM) | Dell v. HP Performance Advantage % | Avg Power Consumption (W) | Performance Per Watt (OPM/W) | Dell v. HP OPM/W Advantage % | |
| HP DL585 G2 Quad-Core | 26 | 80 % | 58,355 | 739 | 81.0 | ||
| Dell R900 Quad-Core | 30 | 80 % | 62,983 | 8 % | 971 | 64.7 | - 20% |
| Dell R900 Six-Core | 40 | 80 % | 74,084 | 27 % | 864 | 85.7 | 6 % |
These results show that the Dell R900 with six-core Intel Xeon 7450 processors provided a 27% performance advantage over the HP DL585 G2 with AMD Opteron quad-core processors and a 6% advantage in performance per watt. A second key finding is that by simply upgrading the processors in the R900 the performance increased by approximately 18% while power consumption decreased by 11% resulting in a 33% improvement in performance per watt.
The complete whitepaper is available in PDF form. Please use this page to ask questions / discuss.
Related Blog Posts
Microsoft Virtualization Team Blog - Hyper-V Performance Scales Well in 24-Core Dell Server
Virtualization Frontier - Hyper-V Performance with 24-Core R900
Virtualization.info - Benchmarks: Hyper-V performance on Dell R900 with Quad-Core and Six-Core Intel Xeon
Gabes Virtual World - Unbelievable Hyper-V Performance on Dell R900
scott_hanson |
Latest page update: made by scott_hanson
, Oct 23 2008, 11:37 AM EDT
(about this update
About This Update
Edited by scott_hanson
1 word added 1 word deleted view changes - complete history) |
|
Keyword tags:
6-core
dunnington
dvd store
hyper-v
intel
microsoft
performance
six-core
More Info: links to this page
|
