Best practices for shared reservations

To maximize resource usage and reduce management overhead, you can share reservations of Compute Engine resources across multiple projects in your Google Cloud organization. A shared reservation allows up to 100 projects in the same organization to create Compute Engine instances by using the same reserved capacity.

Shared reservations offer the following benefits:

  • You maximize the use of reserved capacity.

  • You lower costs by minimizing idle resources.

  • You manage a single, shared reservation rather than many.

This document explains the best practices for creating and managing shared reservations. For more information about the different methods to reserve Compute Engine resources, see Choose a reservation type.

Best practices for shared reservations

When you create a shared reservation, or Compute Engine creates one to fulfill a future reservation request, we recommend the following best practices to help you manage and use the reservation:

  • Have the owner project create the shared reservation but not consume it.

    • To distribute costs among projects, we recommend that only consumer projects, the projects you share a reservation with, consume the reservation. Only use the owner project, the project in which the reservation exists, to create shared reservations.
    • When you create a shared on-demand reservation or future reservation, the owner project must have sufficient quota for the total resources to reserve. Then, after Compute Engine creates the reservation, the owner project must have quota for any reserved resources that it wants to consume. For more information, see Additional quota requirements for shared reservations.

  • Minimize the number of projects in your organization that can create shared reservations.

    • You can view only the reservations that exist in your project. Thus, you can only view a shared reservation in its owner project. Having only a few owner projects in your organization helps you monitor and manage your shared reservations. To control which projects can create shared reservations, see Manage shared reservation creation.
    • Share a reservation only with a few projects to better manage the quota of your reserved resources.

  • Minimize the number of separate shared reservations with identical compute instance properties.

    • An organization can have up to 100 shared reservations for each unique combination of compute instance properties. As a result, if you minimize the number of shared reservations with identical compute instance properties, you mitigate this limit.
    • Having fewer shared reservations improves manageability.

  • Only share reservations between projects with the same Cloud Billing account.

    • Limit each shared reservation to consumer projects that have the same Cloud Billing account as the owner project. This approach helps you track reservation consumption and associated billing.
  • For future reservation requests, carefully review the total count of compute instances that you request.
    • If you are creating a future reservation request, then verify that you request a total count of compute instances that includes all of the following:
      • All matching reserved compute instances that will already exist at the future date.
      • All matching unreserved compute instances that will already exist at the future date.
      • Any matching, unused on-demand reservations that will already exist at the future date.
      • The increase in usage that you want to reserve at the future date.

      For example, suppose that you need 10 additional compute instances at a future date and you'll already have the following resources at that future date:

      • 40 matching reserved compute instances
      • 50 matching unreserved compute instances

      or

      • 40 matching reserved compute instances
      • 50 matching unused on-demand reservations

      Because your existing usage at the future date already adds up to 90 matching compute instances and reservations, and you need an additional ten compute instances, you must specify a total count of 100 in your future reservation request.

      For more information, see Count and provision reserved resources.

What's next