eBooks/eGuides

End User Computing - A Hybrid Multicloud Approach

Issue link: https://insights.oneneck.com/i/1437687

Contents of this Issue

Navigation

Page 31 of 36

32 Host utilization limits A couple of other host and virtualization cluster recommendations are to never exceed 80% host utilization and always size your cluster for N+1. The 80% host utilization is not just for app/desktop virtualization deployments, it's a recommendation that applies to any workload running on a hypervisor. If you are running your hosts past the 80% mark you have very little room for peaks and may also not have enough resource overhead to account for a host failure, depending on the size of your cluster. Always size your cluster for N+1 The second item of figuring for N+1 in your cluster sizing is to ensure that there are enough resources in your cluster to account for a single host failure, to ensure that all VMs can keep running and failed ones will restart without issues. A single host failure is the most common level of resiliency; there is a small set of customers that require N+2 to account for higher SLA requirements. CPU ratios The final item on the compute sizing topic is the CPU ratio, which focuses on the number of virtual CPUs to physical CPUs (vCPU:pCPU). This ratio is very important because if one goes too high with this ratio, it will reach a point where a CPU scheduling issue will arise, dramatically affecting performance and user experience. When a CPU scheduling issue happens on vSphere hosts, the amount of CPU-ready time increases and this lets one know that the scheduler is having trouble getting all of the vCPUs scheduled onto pCPUs. This means that the vCPU will have to wait, even though it's ready. The CPU ratio is very different for the various types of workloads that are virtualized on VMware clusters. Typically, server and database workloads have a much smaller ratio, while VDI workloads are able to have a higher ratio. The use of vCPUs is not a linear calculation, meaning that one can build a host that has a higher consolidation ratio if all VMs have only a single vCPU. When many VMs have two or more vCPUs, this will affect the calculations. It's not as easy as dividing by two to account for twice as many vCPUs. Figure 6 represents a range that has proven to work with real customer deployments. Manufacturers that do synthetic testing may show higher ratios. One should be careful with these, as they do not always apply to real-world designs.

Articles in this issue

Archives of this issue

view archives of eBooks/eGuides - End User Computing - A Hybrid Multicloud Approach