VMware Validated Design for Cloud Providers (for vCloud Director 9.7) - Reference Architecture
In this update of the Reference Architecture, the foundation document has changed from the Certified Reference Design which we used for the last version, to the new VMware Validated Design for Cloud Providers. To get the most from this poster you should read the full Scale and Performance Guide. This version of the Reference Architecture poster is based on the VVD for vCloud Director v9.7 and the Bill of Materials sidebar shows the validated versions of the components tested in this guide.
The Reference Architecture poster shows a simplified model of connectivity between the various components. The service or product specific Reference Architectures in this section provide more detailed connectivity topologies and requirements for their particular areas. This poster should be considered an “overview” to help establish how the reference architecture depicted would impact infrastructure design within a Cloud Provider’s broader Datacenter estate.
The Scale and Performance guide establishes the resources needed to ensure a given level of performance for solutions up to a specified scale. Some elements of the resulting guidelines are fixed, while some are left to the Cloud Provider to meet in their own way. For example, one of the attributes of the Scale Profile (Profile B in this version) establishes the maximum number of configured VMs supported within the solution at 15,000. The vCenter Server Appliance deployed in “medium” size (24GB RAM, 8 vCPU and 400GB storage) can support 4000 VMs, so, in order to manage 15,000 VMs we’ll need four Workload vCenter appliances (in addition to the separate Management one). Having multiple vCenters managed by a single vCD increases both the resilience of the overall solution, and the number of parallel tasks which the vCD solution can potentially have running at once. Our best practice is to have one vCD cell per managed vCenter plus at least one more for resilience (N+1). This means that for Scale Profile “B”, Cloud Providers should deploy 5 vCD cells.
The number of ESXi hosts which are required is addressed differently. Once again there is a formula to derive the appropriate number, but unlike the vCenter example above, the guide is not prescriptive about the specification of the individual hosts a Cloud Provider should use. Instead, we calculate the total (powered on) workload which the Scaling Profile seeks to support, and then estimate the number of hosts each Cloud Provider will need based on their specific hosts’ hardware profile to support that workload.
Scaling Profile B supports 9000 powered on VMs with an average of 1 vCPU per VM. This gives us 9000 vCPUs to support. The calculation then used to estimate the number of hosts incorporates the following elements:
- The number of CPU sockets in each host.
- The number of Cores on each socket.
- The effective Hyper-threading (HT) ratio expected per core.
- The Cloud Provider’s maximum vCPU to pCPU contention ratio.
The VVD Scale and Performance guide goes into more detail about this calculation and the example hosts we used in the validation lab. The guide also provides details of the Management components both required “Core” and optional. In a similar manner to the Workload compute capacity, the guide does not specify how many hosts a Cloud Provider should deploy within the vCD management cluster. Instead it provides a total number of vCPUs, RAM and storage which are required for the selected components. A note on the poster shows the total, but this may vary based upon the optional components selected.
For a more detailed explanation of the Scaling and Performance Guide, as well as full details of the performance achieved, please see the full VMware Validated Design for Cloud Providers - Scale and Performance Guide available from the VMware Cloud Solutions.