Scalability Overview
This section provides an overview of the scalability options available for VMware Integrated OpenStack 2.0 and VMware NSX 6.2 and later environments. The general scalability is divided into two separate options. These two options provide methods for granular or bigger scale steps to grow the environment naturally and elastically. Also, the environment, based on this design, supports automatic scalability. However, this is an optional addition and the necessary configuration and automation steps to provide automatic scalability are not part of this design. They can be added as an “add on” by leveraging the management cluster software as installed in this design.
Scalability of one VMware Integrated OpenStack solution installation will be 10,000 VMs1 and 100 ESXi hosts
OpenStack API concurrency will be six per vCenter Server.
This OpenStack Cloud environment is based on having three major components in place:
The management cluster ‒ Housing all needed software components to run, manage and maintain the OpenStack environment.
The edge cluster ‒ Housing all required components to maintain virtual to physical network connectivity (NSX Edge gateways).
The compute cluster(s) ‒ Providing resources for the workloads generated by consumers of the VMware Integrated OpenStack Solution environment.
Solution Scale-Up
This method will provide further resources by adding additional hardware to the environment. The described solution can scale all three VMware Integrated OpenStack Solution components. This section covers all three clusters but mainly focuses on the scalability of the compute clusters since they are providing resources to the end user.
The maximum scale-up numbers for available options are provided to give an overview of the general scalability of the system. Consider that scaling of ESXi hosts always requires other hardware additions (network ports, power, cooling) that are not described in this section. This design covers only the vSphere and VMware Integrated OpenStack related components; such as rack mount servers or blades.
Table 4. Scale-Up of the Management Cluster
| Attribute | Specification |
|---|---|
| Minimum number of ESXi hosts | 4 |
| Minimum increment of ESXi hosts for scaling | 1 |
| Maximum number of ESXi hosts | 641 |
The current management cluster is sized to support a maximum number of compute and edge resources, without further scaling. The four ESXi hosts currently specified shall be powerful enough to handle all needed software appliances for running the VMware Integrated OpenStack Solution environment. So, a general scale-up of the management cluster is not expected.
Table 5. Scale-Up of the Edge Cluster
| Attribute | Specification |
|---|---|
| Min Number of ESXi Hosts | 3 |
| Min increment of ESXi Hosts for scaling | 2 |
| Max Number of ESXi Hosts | 641 |
The edge cluster might need to scale up based on the number of created virtual networks. Normally an edge host can easily handle 32 NSX Edge appliances (without over commitment), which provides support of up to 32 virtual networks. Since the edge cluster is design for high availability, a 50/50 split is considered by this design. This means that no more than 50 percent of the available capacity shall be used to support an HA switch in case of a host failure. This cluster must always be scaled up by adding two ESXi hosts.
Table 6. Scale-up of the Compute Clusters
| Attribute | Specification |
|---|---|
| Minimum number of ESXi hosts | 3 |
| Minimum increment of ESXi hosts for scaling up | 1 |
| Maximum number of ESXi hosts | 641 |
1The maximum number of ESXi hosts per cluster is also described in the VMware vSphere 6 maximums guide, which can be found at Configuration Maximums.
This is not official guidance, but from previous engagements we have noticed that it is important to understand the expected velocity of parallel provisioning events because that will impact the performance to deliver the application deployments (as the tasks are depending on VMware vCenter to handle concurrent tasks).
The compute cluster is the main VMware Integrated OpenStack solution resource provider and can be scaled on a per single increment basis. This provides maximum flexibility for the chosen design and infrastructure model.
A compute cluster equals an availability zone in this design. This is based on the assumption that a rack can only fit a single vSphere cluster (fully scaled). See more on this in Section 9.1, VMware Integrated OpenStack Architecture.
Solution Scale-Out
If the upscaling of a single VMware Integrated OpenStack Solution environment is not providing enough resources, the next method to consider is to scale on a “per VIO environment” basis. This means that scale-up is done by simply adding more VMware Integrated OpenStack solution environments, each providing the same scale-up options as described earlier.
There is theoretically no limit on the number of VMware Integrated OpenStack solution environments you can add. However, it is important to note that there might be a point when the solution to add additional VMware Integrated OpenStack solution environments is not suitable anymore from a management or operations point of view.
As mentioned earlier, this section was only meant to provide an overview about the scaling mechanics of this design, and not meant to provide the complete story on scalability on its own. Following sections of this document cover other important decisions and configuration details to maintain this scalability.
1. This number reflects the maximum number of VMs supported by this configuration. The actual VM count will vary based on CPU and memory sizes of deployed VMs, so it might also be less. ↩