Manage resource quotas

Basic Information: Includes resource quota name, ID, and associated workspace.

• Click the edit icon to update Resource Quota Name, associated workspace, and Tag.

• After associating a workspace, click the workspace name to go to its details page. Then use the resource quota for AI development in that workspace.

Resource Information:

• Resource group information: Includes the resource group to which the resource quota belongs, as well as its parent and child resource quotas. For a diagram of the parent-child relationship, see the Create parent-child quotas section. Click a resource name to navigate to its details page.

• GPU Default Driver: Allows you to configure a default driver to meet custom requirements for GPU types, drivers, SDKs, and container images.

Network Information: Displays limits on network resources, including VPCs, security groups, and NAT Gateways. This information defines the resource quota's scope at the network layer.

Scheduling Information: Displays scheduling information for the resource quota and allows you to update its configuration.

• Scheduling Policy: Select a suitable scheduling policy based on its underlying principles to improve dequeuing efficiency and compute resource utilization. For configuration details, see Scheduling policies.

• Child-level PreemptionandSelf-level Preemption: When resources are limited, queued jobs on the current resource quota can preempt running jobs on child-level or self-level quotas. For configuration details, see Preemption policies.

• Idle Sharing: Enabled by default. This allows idle jobs to use resources from the current quota and its child quotas.

Resource Change History: View a history of create, scale, and delete operations. The record shows the change type, initiator, status, and target instance type.

Event History: View events related to the quota, such as node scaling, enabling/disabling scheduling, and attaching/detaching nodes. The history includes the event's trigger time, initiator, and content.

Advanced Information:

• Enable Local Cache: For Intelligent Computing Lingjun resource quotas, you can enable this feature to cache data close to compute nodes, which improves read performance. Enabling this feature consumes some resources on the node, so ensure that you have sufficient resources. For more information, see Intelligent Computing Lingjun cache acceleration.

  • Storage service: Supports local cache for OSS, General-purpose NAS, and BMCPFS.

  • Supports RDMA: Cache-accelerated reads support RDMA (tail), providing higher communication bandwidth to meet data read performance requirements.

• ENI warm-up: Pre-provisions elastic network interfaces (ENIs) for nodes within a resource quota to accelerate job startup. Note that enabling this feature consumes ENI resources, so ensure that your VPC subnet has a sufficient number of available IP addresses. For a node allocated to a child resource quota, the configuration of the lowest-level quota takes precedence.

Node details: Includes Node Specification, the Dedicated Resource Group Name/ID, resource In Use and Total amounts (GPU Type, Number of GPUs, and CPU Cores), the zone (AZ), the high-speed network interconnect zone (HZ), and the Tasks and Instances created on the node.

• Job and instance details: In the Tasks and Instances columns for a target node, click the corresponding number to view job and instance details.

• Filter nodes: Filter nodes by node status or order status, or sort by resource amount.

• hz (high-speed network interconnect zone): In Intelligent Computing Lingjun scenarios, hz indicates the high-speed network zone where the underlying compute resources reside. Resources with the same hz identifier can communicate over a high-speed network.

Node Status: Supported statuses:

• Ready: Compute node is available.

• Not Ready: Compute node is being initialized.

• Scheduling Disabled: Node cannot be scheduled. Possible causes:

  • Stopped by User: A user manually disabled scheduling on the node.

  • Expired: Node subscription has expired.

  • Recovering: Node is in the recovery process. If jobs are running on this node, stop them promptly to avoid interrupting recovery.

  • Unknown: Cause is unknown. Contact your account manager for assistance.

Manage nodes:

• Stop/Start scheduling on a node: In the Actions column for the target node, click Stop Scheduling or Start Scheduling to pause or resume resource scheduling on that node.

• Clear a node: In the Actions column for the target node, click Clear Node to terminate all jobs (including DSW, DLC, or EAS jobs) on that node.

• Download node list: Click the icon to export the node list from the current page and download it to your local machine.

Filter jobs: Filter by Type or Job Status.

View details: Click a job name, resource quota name, or workspace name to go to the corresponding details page.

Filter jobs in the current quota: Enable the View Current Resource Quota switch to display only jobs created within this quota.

Filter users of current quota: Enable the View Current Resource Quota Users switch to display only users who submitted jobs directly to this quota, not its child quotas.

View job details: In the Number of Tasks column for a user, click Details to view the jobs submitted by that user. On the job list page, click a job name to view its details.

Displays a GPU computing power heat map (real-time utilization), resource levels, and job status distribution.

View monitoring information from two perspectives: quota and node. This includes metrics for CPU, memory, disk, network, and GPU. For details on the metrics and how to use CloudMonitor and ARMS to view data, configure alerts, and subscribe to metrics, see Resource quota monitoring and alerting.

Resource View: Displays allocation of CPU, memory, and GPU resources for the current quota and its child quotas.

Job View: Displays the total number of jobs created in the current quota and its child quotas, with a breakdown by status.