Using the CPU entitlement plug-in

Page last updated:

This topic describes how Cloud Foundry administrators can use the CPU entitlement plug-in to help them manage their CPU resources.

CPU Entitlement

To watch an overview of the topic, see CPU Entitlements in Cloud Foundry on YouTube.

CPU entitlement describes the percentage of host CPU a particular app instance is entitled to use. The CPU entitlement plug-in shows that apps have a CPU performance of 100% when they are using exactly the CPU they are entitled. Apps have a CPU performance of less than 100% when their usage is less than their entitlement and greater than 100% when they are above their entitlement.

In Cloud Foundry, apps have CPU entitlements that are proportional to their allocated memory by default. For example, an app with access to 256MB of memory on a 512MB machine has access to half of the memory on the machine and is also entitled to half of the CPU of that machine.

You can change the exact mapping of memory to CPU by configuring the Garden BOSH release property called experimental_cpu_entitlement_per_share_in_percent. This property changes the CPU percentage entitled to a container per CPU share. For example, a value of 0.3 can mean that each app has access to 0.3% of the total CPU per share.

By default, the total shares available on a host is equal to the amount of memory on the host, an optimal value for experimental_cpu_entitlement_per_share_in_percent is 100% divided by the amount of total memory on the host. For example, a host with 1024 MB of memory can have an optimal value of 100 / 1024, which is 0.0977% per share. An 256 MB app would have access to 76.8% of the total CPU. In this case, if there is only a single core machine, an operator has over committed the amount of CPU available.

The way you configure experimental_cpu_entitlement_per_share_in_percent can create three possible states for the apps in your Cloud Foundry deployment:

• Undercommitted - apps’ CPU entitlements are guaranteed minimums, but some host CPU might not be used

• Optimal - apps’ CPU entitlements are guaranteed minimums

• Overcommitted - apps are not guaranteed to have access to their entitlement

The following table contains examples of optimal values for machines with different total memory and number of CPU cores:

CPU Entitlement plug-in

The metric absolute_entitlement shows an app’s CPU usage relative to its entitlement.

To retrieve absolute_entitlement metrics for all instances of an app:

1. Install the CPU Entitlement Plug-in from the cpu-entitlement-plug-in repository on GitHub.

2. Run:

   cf cpu-entitlement APP-NAME
   

   Where APP-NAME is the name of the app.
   
   This command returns absolute_entitlement metrics for all instances of the app, similar to the following example:

   Showing CPU usage against entitlement for app dora-example in org example-org / space example-org-staging as dora@example.com ...
   ​
        avg usage   curr usage
   #0   1.62%       1.66%
   #1   2.93%       3.09%
   #2   2.51%       2.62%
   

After you run the cf cpu-entitlement command, you see the following values: * avg usage * curr usage.

The average usage is used to split the app into two groups, good and bad. Good apps have an average CPU usage that is below 100% of their CPU entitlements, and bad apps have an average CPU usage that is over 100% of their CPU entitlements. In the receding example, all values for the average CPU usage are under 100% of their CPU entitlements.

This partial is used for both VMware Tanzu Application Service for VMs and Cloud Foundry docs.

Don’t reference VMware Tanzu Application Service for VMs or BOSH configurations here.

Spare CPU resources and throttling

If there are three cores on the Diego Cell to which your app is deployed, 300% CPU can be distributed between all the apps on the Diego Cell. This is the percentage that the cf app command displays. The metrics depend on factors such as the capacity of the Diego Cell and the total number of apps on it that are not visible to the user. This can make it difficult for users and operators to balance CPU resources.

An app always receives 100% of its CPU entitlement, no matter what other apps are on the Diego Cell. If there are spare resources on the machine, an app can consume over 100% CPU.

Spare resources are distributed on the Diego Cell when the experimental_cpu_throttling property is set to false. If an app consumes less than its entitlement the majority of time and needs CPU resources over its entitlement, it receives the maximum amount of CPU resources from the spare CPU. but without the CPU entitlement feature that depends on the other apps running on the same Diego Cell. If the other apps constantly use high CPU, they restrict other apps to occasionally use more CPU when needed. Even though the other apps on the same Diego Cell usually use much more than their CPU entitlements, when an app that usually consumes less than its CPU entitlement needs extra CPU, the spare capacity is distributed evenly between that app and all the other apps. Because of this, that app can never spike over a certain amount of CPU. To occasionally provide the bandwidth, operators might over provision resources for Diego Cells.

When the experimental_cpu_throttling property is set to false, apps are allowed to temporarily exceed their entitlement again, but apps that have been using less than their entitlement over a longer period of time are prioritized. Apps are never forced below their entitlement.

Both good and bad apps always get their entitlement. If you are using the CPU Entitlement plug in, the difference is in how the spare CPU gets distributed between them. All spare CPU is given to the good app that needs it, and the bad apps are throttled to only the amount of CPU to which they are entitled.

If the CPU spike of a good app is for a long period of time, it eventually becomes a bad app and therefore is throttled to 100% if a different good app needs the CPU. This way, resources get distributed more fairly between apps.