Container Compute Service:GPU-HPN capacity reservation

Background information

ACS clusters do not require you to manage nodes. However, to maintain compatibility with native Kubernetes, virtual nodes are still visible in the cluster. Virtual nodes provide a large capacity of computing resources, which gives ACS clusters high elasticity to handle sudden increases in service traffic. The default virtual nodes are generated based on the vSwitchIds in the acs-profile ConfigMap and do not consume any computing resources.

About GPU-HPN capacity reservation

GPU-HPN capacity reservations currently support only the subscription billing method. After a successful reservation, the resources remain available throughout the subscription period. If you reserve multiple instances at once, you can associate them with multiple clusters simultaneously. Reserved nodes can be allocated as pod instances of different quantities and specifications. For example, in scenarios such as large model training and fine-tuning that require an HPN for distributed communication between GPU cards, a GPU-HPN capacity reservation allows for billing deductions based on whole GPU cards. The number of cards is determined by the pod specification, with no constraints on CPU and memory sizes.

Billing and limits

Capacity reservations are billed by the second and follow pay-as-you-go standards. The billing cycle for a capacity reservation is as follows:

Billing start: Billing starts when the reservation is successfully created and its status changes to active.

Billing end: Billing stops after the reserved capacity expires and is automatically released.

Create a GPU-HPN capacity reservation

1. Log on to the ACS console. In the left-side navigation pane, click Capacity Reservation.

2. On the Capacity Reservation page, click Create GPU-HPN Capacity Reservation and enter the following information.

   Parameter	Description
   Region	Select the region where the resource reservation is located.
   Zone	Select the zone where the resource reservation is located.
   Quantity	Currently, only Instance is supported.
   Category	Currently, only GPU is supported.
   HPN Zone	Select the HPN Zone where the resource reservation is located.
   Node Type	Refer to the options displayed in the console.
   Subscription Duration	1 Month, 1 Year, 3 Years.
   Down Payment (%)	0% down payment. Use the service immediately after purchase.
   Discount Information	Varies by subscription duration. A 1-month subscription has No Discount. A 1-year subscription has a Discount for 12-month Subscription. A 3-year subscription has a Discount for 36-month Subscription.
   Billing Cycle	Monthly.
   Installments	Depends on the subscription duration. For example, a 1-month subscription has 1 Installment, and a 1-year subscription has 12 Installments.
   Quantity	The number of instances to purchase.

   After you complete the configuration, click Buy Now. On the Confirm Order page, click Pay. On the payment page, click Subscribe to complete the purchase.

Associate a cluster

1. Log on to the ACS console. In the left-side navigation pane, click Capacity Reservation.

2. On the Capacity Reservation page, find the capacity reservation that you want to associate with a cluster. In the status bar of the reservation, click Associate Cluster.

   Important

   Only ACS clusters, ACK managed clusters, ACK One registered clusters, and ACK One distributed workflow Argo clusters can be associated. Other cluster types are not supported.

3. In the Resource Association dialog box, select or enter the cluster ID and the number of instances to associate. Then, click OK.

   Important

   In the Resource Association dialog box, the Associate Cluster drop-down list shows only the IDs of ACS clusters and ACK managed clusters. For ACK One registered clusters and ACK One distributed workflow Argo clusters, you must enter the cluster ID directly.

   

Query allocated resources in a GPU-HPN capacity reservation

A GPU-HPN capacity reservation is represented as a Kubernetes node in the cluster. You can use the kubectl tool to view the allocated resources of the node.

1. Run the following command to view the GPU-HPN node by its label.

   kubectl get node -l alibabacloud.com/node-type=reserved

   Expected output:

   NAME                                      STATUS   ROLES   AGE   VERSION
   cn-wulanchabu-c.cr-rkccqmu0xz8rea1*****   Ready    agent   20m   v1.28.3-aliyun

2. Run the following command to query the allocated resources in the GPU-HPN capacity reservation. Replace the node name in the command with the GPU-HPN node name that you obtained in the previous step.

   kubectl describe node cn-wulanchabu-c.cr-rkccqmu0xz8rea1***** | grep Allocated -A 10

   Expected output:

   Allocated resources:
     (Total limits may be over 100 percent, i.e., overcommitted.)
     Resource              Requests    Limits
     --------              --------    ------
     cpu                   16 (8%)     16 (8%)
     memory                128Gi (7%)  128Gi (7%)
     ephemeral-storage     30Gi (0%)   30Gi (0%)
     nvidia.com/gpu        1           1
   Events:                 <none>

   Lines 5 to 8 show the allocated amount and the allocation rate for CPU, memory, ephemeral storage, and GPUs, respectively.