HOST SIZING AND CONSOLIDATION PLANNING
Host sizing consists of determining the total of all the workloads to be consolidated (CPU, RAM, Disk I/O, Network I/O, and so on) as well as the largest individual workloads to be consolidated. Next, a standard host server architecture (or architectures) is designed and benchmarked to determine the real world capacity of each host server architecture. The capacity of the individual standard host server is divided into the workload to be consolidated to determine the number of host servers required. This is done individually for processor, RAM, Disk I/O, Network I/O, and so on. This number is used to determine the overall consolidation ratio by comparing it to the original number of candidate servers.
The end result of the sizing methodology is one or more standard host server architectures and the required number of servers required of each architecture in order to virtualize the complete candidate workload.
In most cases, the amount of RAM in the host server is the factor that first constrains the number of guests the server is able to host. Additionally, the amount of RAM allocated to an individual guest is usually the constraining factor for guest OS performance. Fortunately, the price of RAM and the maximum amount of RAM supported by commodity servers has improved significantly in the last few years. As such, this guidance recommends host servers with a large amount of RAM and the allocation of at least 2 GB of RAM for production guest virtual machines. While it is certainly possible to achieve much higher consolidation ratios, the previous guidance will ensure high performing virtual machines.
Consolidation Candidate Workload Analysis
During the Discovery and Assessment phases, a set of physical servers that are consolidation candidates is identified. The consolidation candidates are analyzed over a period of time to determine the average and peak utilization of CPU, RAM, Disk I/O, Network I/O, and so on. This workload analysis is a critical part of the host sizing process because it determines the total workload that must be supported on the host architecture as well as the largest individual workloads that are going to be consolidated to ensure that they are within the physical limits of a single virtual machine (4 CPU Core, 64 GB of RAM, and so on). If a single workload exceeds the limits of a single virtual machine, it should not be migrated to a virtual machine without an architecture change to scale it out to additional virtual machines.