vNUMA介绍
http://virtualbarker.com/
vSphere VMware Performance
With every release of vSphere the overhead of running an application on the vSphere virtualized platform is reduced by the new performance improving features. Typical virtualization overhead for applications is less than 10%. Many of these features not only improve performance of the virtualized application itself, but also allow for higher consolidation ratios. Understanding these features and taking advantage of them in your environment helps guarantee the highest level of success in your virtualized deployment. Table 6 provides details on vSphere VMware performance.
Table 6 vSPhere VMware performance
NUMA Support
ESX/ESXi uses a NUMA load-balancer to assign a home node to a virtual machine. Because memory for the virtual machine is allocated from the home node, memory access is local and provides the best performance possible. Even applications that do not directly support NUMA benefit from this feature.
See The CPU Scheduler in VMware ESXi 5.1:
http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.1.pdf
Transparent page sharing
Virtual machines running similar operating systems and applications typically have identical sets of memory content. Page sharing allows the hypervisor to reclaim the redundant copies and keep only one copy, which frees up the total host memory consumption. If most of your application virtual machines run the same operating system and application binaries then total memory usage can be reduced to increase consolidation ratios.
See Understanding Memory Resource Management in VMware ESXi 5.1: http://www.vmware.com/files/pdf/perf-vsphere-memory_management.pdf
Memory ballooning
By using a balloon driver loaded in the guest operating system, the hypervisor can reclaim host physical memory if memory resources are under contention. This is done with little to no impact to the performance of the application.
See Understanding Memory Resource Management in VMware ESXi 5.1: http://www.vmware.com/files/pdf/perf-vsphere-memory_management.pdf
Memory compression
Before a virtual machine resorts to host swapping, due to memory over commitment the pages elected to be swapped attempt to be compressed. If the pages can be compressed and stored in a compression cache, located in main memory, the next access to the page causes a page decompression as opposed to a disk swap out operation, which can be an order of magnitude faster.
See Understanding Memory Resource Management in VMware ESXi 5.1:
http://www.vmware.com/files/pdf/perf-vsphere-memory_management.pdf
Large memory page support
An application that can benefit from large pages on native systems, such as MS SQL, can potentially achieve a similar performance improvement on a virtual machine backed with large memory pages. Enabling large pages increases the memory page size from 4KB to 2MB.
See Performance Best Practices for VMware vSphere 5.1:http://www.vmware.com/pdf/Perf_Best_Practices_vSphere5.1.pdf
and see Performance and Scalability of Microsoft SQL Server on VMware vSphere 4:
http://www.vmware.com/files/pdf/perf_vsphere_sql_scalability.pdf