Properly Sizing VRAM
vRAM and pRAM
For purposes of clarity, vRAM is used to refer to the amount of virtual RAM assigned a virtual machine. pRAM is the amount of physcial RAM on a server. For a particular host, it is possible to allocate more vRAM to the powered on VMs on that host than there is actual physical RAM. However, on average, the ratio of vRAM to pRAM is about .7 according the VKernel Virtualization Management Index.
The key is to provide each VM with just enough vRAM to hit performance goals, and to pack each host with the maximum number of VMs permitted by the amount of pRAM. This assumes of course, that your applications are RAM limited. By doing this approach, you would:
- Reduce vSphere licensing costs
- Reduce server CAPEX and OPEX
Understanding VMware vRAM Licensing
vRAM licensing is used to determine the amount of vSphere licenses required. Each physical CPU socket is allocated a certain amount of vRAM. The vRAM allotment can be pooled across servers and across vCenters. It is believed that VMware implemented this type of model to prevent revenue loss caused by servers doubling in power every 18 months with ever higher core densities.
VMware's initial announcement had low VRAM entitlements. A subsequent announcement on August 3, 2011 increased vRAM allotments and put a maximum vRAM consumption per VM to support large VMs.
Supporting materials:
- VMware's vSphere Pricing Document
- VMware's initial blog entry clarifying the pricing change
- VMware's revised pricing blog entry announcing the revised limits
- Good summary of changes on Computerworld
- Article from SearchServerVirtualization explaining the fresh revisions "VMware Blinks on vSphere 5 Licensing"
White Papers
Understand VMware Memory Metrics
Before attempting to optimize an environment for vRAM, it is important to understand the key memory metrics in VMware that should be monitored. "Memory Management Metrics for VMware Environments" is a white paper that provides information on the key VMware memory metrics you should use for VM (virtual machine) memory allocation sizing and VM performance issue monitoring, as well as recommendations on how to make these assessments when large memory pages are in use.
vRAM Sizing Considerations White Paper
Balancing VM memory sizing to both reduce vSphere licensing and maximize virtual machine performance can be a difficult task. VKernel has written a white paper with noted vExpert David Davis, called "VM Memory (vRAM) Sizing Considerations" for virtualization administrators to gain a deeper understanding of VMware's memory over-commitment capabilities and which VMware system metrics to analyze to make VM memory sizing less painful.
Free vRAM Assessment Tool
CapacityVIEW collects CPU, memory and storage allocations and utilization levels. It then analyzes this data to provide you with desktop capacity alerts concerning:
- Amount of allocated vRAM
- Virtual machine I/O Latency
- Under-provisioned memory, CPU, storage
- Low capacity for new deployments
- Over-provisioned memory, CPU, storage
Reducing vRAM Allocation
To reduce vRAM allocation, virtualization administrators should:
- Rightsize vRAM allocation to each VM based on the amount of memory a VM actual needs
- Find and eliminate Zombie VMs
- Implement chargeback or showback to assist with rightsizing and also reduce VM sprawl
VKernel's vOPS Server Standard is available as a 30-day trial to assist with Rightsizing, Zombie VM hunting and chargeback.
