11.8. Over-Provisioning and Service Offering Limits
(Supported for XenServer, KVM, and VMware)
CPU and memory (RAM) over-provisioning factors can be set for each cluster to change the number of VMs that can run on each host in the cluster. This helps optimize the use of resources. By increasing the over-provisioning ratio, more resource capacity will be used. If the ratio is set to 1, no over-provisioning is done.
The administrator can also set global default over-provisioning ratios in the cpu.overprovisioning.factor and mem.overprovisioning.factor global configuration variables. The default value of these variables is 1: over-provisioning is turned off by default.
Over-provisioning ratios are dynamically substituted in CloudStack's capacity calculations. For example:
Capacity = 2 GB
Over-provisioning factor = 2
Capacity after over-provisioning = 4 GB
With this configuration, suppose you deploy 3 VMs of 1 GB each:
Used = 3 GB
Free = 1 GB
The administrator can specify a memory over-provisioning ratio, and can specify both CPU and memory over-provisioning ratios on a per-cluster basis.
In any given cloud, the optimum number of VMs for each host is affected by such things as the hypervisor, storage, and hardware configuration. These may be different for each cluster in the same cloud. A single global over-provisioning setting can not provide the best utilization for all the different clusters in the cloud. It has to be set for the lowest common denominator. The per-cluster setting provides a finer granularity for better utilization of resources, no matter where the CloudStack placement algorithm decides to place a VM.
The overprovisioning settings can be used along with dedicated resources (assigning a specific cluster to an account) to effectively offer different levels of service to different accounts. For example, an account paying for a more expensive level of service could be assigned to a dedicated cluster with an over-provisioning ratio of 1, and a lower-paying account to a cluster with a ratio of 2.
When a new host is added to a cluster, CloudStack will assume the host has the capability to perform the CPU and RAM over-provisioning which is configured for that cluster. It is up to the administrator to be sure the host is actually suitable for the level of over-provisioning which has been set.
11.8.1. Limitations on Over-Provisioning in XenServer and KVM
In XenServer, due to a constraint of this hypervisor, you can not use an over-provisioning factor greater than 4.
The KVM hypervisor can not manage memory allocation to VMs dynamically. CloudStack sets the minimum and maximum amount of memory that a VM can use. The hypervisor adjusts the memory within the set limits based on the memory contention.
11.8.2. Requirements for Over-Provisioning
Several prerequisites are required in order for over-provisioning to function properly. The feature is dependent on the OS type, hypervisor capabilities, and certain scripts. It is the administrator's responsibility to ensure that these requirements are met.
All VMs should have a balloon driver installed in them. The hypervisor communicates with the balloon driver to free up and make the memory available to a VM.
The balloon driver can be found as a part of xen pv or PVHVM drivers. The xen pvhvm drivers are included in upstream linux kernels 2.6.36+.
The balloon driver can be found as a part of the VMware tools. All the VMs that are deployed in a over-provisioned cluster should have the VMware tools installed.
All VMs are required to support the virtio drivers. These drivers are installed in all Linux kernel versions 2.6.25 and greater. The administrator must set CONFIG_VIRTIO_BALLOON=y in the virtio configuration.
11.8.2.2. Hypervisor capabilities
The hypervisor must be capable of using the memory ballooning.
The DMC (Dynamic Memory Control) capability of the hypervisor should be enabled. Only XenServer Advanced and above versions have this feature.
Memory ballooning is supported by default.
11.8.3. Setting Over-Provisioning Ratios
There are two ways the root admin can set CPU and RAM over-provisioning ratios. First, the global configuration settings cpu.overprovisioning.factor and mem.overprovisioning.factor will be applied when a new cluster is created. Later, the ratios can be modified for an existing cluster.
Only VMs deployed after the change are affected by the new setting. If you want VMs deployed before the change to adopt the new over-provisioning ratio, you must stop and restart the VMs. When this is done, CloudStack recalculates or scales the used and reserved capacities based on the new over-provisioning ratios, to ensure that CloudStack is correctly tracking the amount of free capacity.
It is safer not to deploy additional new VMs while the capacity recalculation is underway, in case the new values for available capacity are not high enough to accommodate the new VMs. Just wait for the new used/available values to become available, to be sure there is room for all the new VMs you want.
To change the over-provisioning ratios for an existing cluster:
Log in as administrator to the CloudStack UI.
In the left navigation bar, click Infrastructure.
Under Clusters, click View All.
Select the cluster you want to work with, and click the Edit button.
Fill in your desired over-provisioning multipliers in the fields CPU overcommit ratio and RAM overcommit ratio. The value which is intially shown in these fields is the default value inherited from the global configuration settings.
In XenServer, due to a constraint of this hypervisor, you can not use an over-provisioning factor greater than 4.
11.8.4. Service Offering Limits and Over-Provisioning
Service offering limits (e.g. 1 GHz, 1 core) are strictly enforced for core count. For example, a guest with a service offering of one core will have only one core available to it regardless of other activity on the Host.
Service offering limits for gigahertz are enforced only in the presence of contention for CPU resources. For example, suppose that a guest was created with a service offering of 1 GHz on a Host that has 2 GHz cores, and that guest is the only guest running on the Host. The guest will have the full 2 GHz available to it. When multiple guests are attempting to use the CPU a weighting factor is used to schedule CPU resources. The weight is based on the clock speed in the service offering. Guests receive a CPU allocation that is proportionate to the GHz in the service offering. For example, a guest created from a 2 GHz service offering will receive twice the CPU allocation as a guest created from a 1 GHz service offering. CloudStack does not perform memory over-provisioning.
