VMWare ESX(i) VS. Microsoft Hyper-V. This comparison table, albeit bias, is BRUTAL. In some cases the VMWare writer doesn’t even soften the punch with some corporate dialect. The writer might as well say “You suck. We are better. So there!” I like it!!
- Hyper-V only supports Novell as a guest OS. No Linux.
- Hyper-V’s third-party hardware drives not written and reviewed for virtualization
- 4 processors ONLY Windows 2008 Server guest OSs
- No “native” NIC teaming
- No overcommitted on memory resources
| Features |
VMware ESX/ESXi 3.5 |
Microsoft Hyper-V RTM |
Hypervisors Comparison |
| Thin and Purpose-Built Architecture |
| Small disk footprint |
|
|
Hyper-V is fully dependent on a general-purpose OS(Windows 2008). Microsoft’s best-case disk footprint is 2.6GB (Hyper-V + Server Core), which is more than 80x times larger than ESXi’s (32MB). Hyper-V + Windows Server 2008 is even larger with a footprint of ~10GB. The large size of both options creates an easier target for security attacks and increasing performance overhead. Also, patches to Windows Server 2008 / Server Core will affect Hyper-V guests. VMware ESXi on the other hand is a fully functional hypervisor in a 32MB disk footprint, which reduces the risk of downtime and increases reliability. |
| OS independence |
|
|
Hyper-V depends on Windows 2008 and is consequently biased towards Windows guests when it comes to making full use of the available hardware resources:
- Hyper-V support for non-Windows guest OSs is limited to only Novell SUSE
- Hyper-V support for 4-ways vSMP is limited to Windows Server 2008. For Windows 2003, Hyper-V only supports up to 2-way vSMP
VMware ESX/ESXi is optimized for virtualization, offers thebroadest OS support (10 Windows versions, 16 Linux versions, 2 Novell Netware versions and 2 Sun Solaris versions) and allows all supported guest OSs to optimally utilize available resources (4-way vSMP is available on all supported server OSs) |
| Hardened drivers |
|
|
Windows 2008 is a general purpose OS that relies on generic device drivers built by third parties that are not designed for virtualization. This fact dramatically increases the likelihood of performance problems, downtime related to driver patches, and system failures, even though the drivers go through Microsoft’s certification process. Windows drivers already go through Microsoft certification today and driver problems still exist. On the other hand, VMware works closely with IHVs to test and optimize their drivers for use with ESX/ESXi to ensure, overall reliability and performance when running concurrent virtualized workloads. |
| Resource Efficiency |
| Advanced memory management |
|
|
System memory management is in most cases the dominant factor that determines the number of concurrent VMs that can run on a physical host. Hyper-V cannot overcommit host memory because it does not support features such as transparent memory page sharing and memory ballooning to reclaim idle physical memory. Consequently Hyper-V delivers lower virtual machine density, and low ratios of server consolidation compared to VMware ESX/ESXi. VMware ESX/ESXi fully supports memory overcommit, enabling high rates of VM density while delivering consistent performance. VMware customers regularly use memory overcommit in their production datacenters at rates of 1.5X to 2.5X without any noticeable impact to end-users. |
| Advanced storage management |
|
|
Hyper-V uses the NTFS file system for storage. NTFS is not a clustered file system. Therefore, Hyper-V does not provide native support for shared virtual disks and live migration. Hyper-V’s lack of a clustered file system means that with Microsoft quick migration (based on host clustering) all VMs on a LUN will be migrated at the same time. To avoid this unwanted result, a user must assign one VM per LUN, which creates a storage nightmare. The only other option is to purchase a third party clustered file system at additional licensing and support cost. VMware ESX/ESXi comes with VMFS, a clustered file system designed specifically for virtualization. VMFS fully supports live migration, and enables multiple VMs to share a single LUN and still migrate/failover individual VMs. |
| Advanced networking management |
|
|
Hyper-V does not support native physical NIC teaming, instead requiring 3rd party drivers from IHVs (Intel, Broadcom). This limitation prevents Hyper-V from providing out-of-the-box, hardware independent protection from NIC failure. Additionally, Hyper-V drivers for supported NICs are not virtualization aware, cannot be controlled from within the hypervisor, and must be loaded in the parent partition. VMware ESX/ESXi fully supports NIC teaming, resulting in a simple, comprehensive out of the box solution for NIC failover. |
| Performance Acceleration |
| I/O scalability |
|
|
Hyper-V uses an “indirect driver” model that keeps all device drivers in a Windows 2008 instance running in the Parent Partition of Hyper-V and relies on the general purpose OS to handle the I/O traffic. In Hyper-V’s model, all I/O requests from guest OSs must travel through Windows Server 2008 (or Server Core). This design negatively affects the efficiency of the process scheduler to manage I/O requests and results in higher latency. This impact is even more accentuated with virtual SMP as the number of processes handled by the scheduler must manage grows. The outcome is scalability issues when running many concurrent virtual machines under load, resulting in low consolidation ratios. VMware ESX/ESXi utilizes a direct driver model with a scheduler built specifically to handle the demands of multiple, high workloads. This scalability results in higher VM density ratios with while delivering consistent, high performance. |
| Distributed resource allocation |
|
|
Hyper-V lacks many capabilities to optimally distribute hardware resources., based on real time workload conditions. Unlike ESX/ESXi, Hyper-V does not support:
- Network traffic shaping for load balancing
- Storage I/O prioritization to ensures that critical VMs receive priority access to storage devices
- Memory swap prioritization (memory ballooning)
VMware ESX/ESXi has a proven solution for hardware resource distribution, including full support for network traffic shaping, I/O prioritization, and memory ballooning for memory management and prioritization. |
| Support for performance enhancing technologies |
|
|
Hyper-V lacks support of several performance enhancing technologies such as:
- AMD RVI (nested page tables), included in the latest release of the AMD Barcelona CPU
- Large memory pages, which allows improved performance when virtualizing databases
VMware ESX/ESXi supports both nested page tables and large memory pages and is therefore capable of superior performance even in the most demanding environments. Mission critical and intensive applications such as an Oracle database will perform better in a VMware virtual environment. |
| Linux paravirtualization support |
|
|
Hyper-V supports the “old way” of doing Linux para-virtualization, where IT admins maintained two kernels, one for physical and one for virtual machines. This extra maintenance increased complexity and management costs. On the other hand, VMware ESX/ESXi supports Linux para-virtualization through paravirt_ops / VMI (Virtual Machine Interface) – the new industry accepted way to para-virtualize Linux. This new way does not require IT admins to maintain and support separate kernels for physical and virtual machines – it’s all in one kernel. Currently, SLES 10, and Ubuntu 7 and 8 all support paravirt-ops/VMI. |
| Advanced Virtualization Features |
| Virtual security technology |
|
|
Hyper-V does not provide dedicated capabilities for security vendors to extend their security capabilities to virtualized environments. VMsafe, an upcoming feature of VMware ESX/ESXi, will enable a rich ecosystem of third-party security solutions for virtualized environments. VMsafe is a virtual security technology that provides fine-grained visibility over virtual machine resources, making it possible to monitor every aspect of system execution and stop previously undetectable viruses, rootkits and malware before they can infect a system. |