GPUs are scarce computing resources, and on-demand acquisition involves uncertainty that may interrupt or delay critical workloads. Alibaba Cloud Container Service (ACS) provides two resource reservation modes to deliver deterministic GPU capacity for your applications.
GPU Pod capacity reservation (Pod-level Reservation)
-
How it works: Pod capacity reservation is a workload-oriented, standardized capacity reservation. You specify Pod specifications (such as 2×A10 GPU, 16 vCPU, 32 GiB memory) and the number of Pods to reserve (such as 12). The platform reserves computing capacity that accommodates exactly these 12 Pods with the specified configuration.
-
Determinism provided: This mode delivers "workload capacity determinism". Whenever you initiate a creation request, the system guarantees capacity for your specified 12 Pods. This simplifies capacity planning because you do not need to worry about underlying node specifications or resource fragmentation—only your application Pod requirements.
-
Scenarios:
-
Homogeneous workloads: Ideal when your application (such as large-scale distributed training or online inference) consists of many Pods with identical specifications.
-
Simplified operations: Suitable when you want to delegate underlying resource planning to the platform and focus only on application-level capacity needs.
-
GPU-HPN capacity reservation (Node-level Reservation)
-
How it works: This mode reserves and locks dedicated GPU computing node capacity in the underlying ACS resource pool. These resources are exclusively allocated to your account, ensuring that active hardware is always available to host new GPU Pods. This prevents Pod scheduling failures (Pending status) caused by resource pool constraints.
-
Determinism provided: This mode delivers "physical resource determinism". When you need to scale out, the underlying infrastructure (GPU nodes) is guaranteed to be available. You decide how to schedule and combine Pods of different specifications on these nodes (known as "bin packing").
-
Scenarios:
-
Heterogeneous workloads: Provides maximum flexibility when you need to run GPU Pods with various specifications in the same resource pool.
-
Fine-grained resource control: Suitable when you want to use custom scheduling policies (such as Taints/Tolerations, Node Affinity) to control Pod placement precisely for performance optimization or resource isolation.
-
Summary and comparison
|
Attribute |
GPU Pod capacity reservation (Pod-level) |
GPU-HPN capacity reservation (Node-level) |
|
Reservation object |
Number of Pods with specific specifications. |
Underlying GPU computing node capacity. |
|
Reservation granularity |
Logical workload (such as 12 Pods with 1A10GPU8C16G). |
Physical node resources (such as 2 P16EN nodes). |
|
Guarantee level |
Workload capacity determinism. |
Physical node resource determinism. |
|
Flexibility |
Lower (bound to specific Pod specifications). |
Very high (can run Pods with flexible specifications). |
|
Management complexity |
Low (platform handles resource matching). |
Higher (requires response to node management events). |
|
Selection recommendation |
|
Medium to large-scale applications with complex and variable specifications. |
Choose the reservation mode that matches your workload profile and determinism requirements to mitigate GPU resource acquisition risks and keep your AI applications running reliably.