The requirements involve two complementary capabilities: automatically adjusting inference-serving capacity and prioritizing GPU allocation between competing workloads.
KServe autoscaling, represented by option B, enables inference deployments to adjust their replica counts according to configured scaling policies and demand. This helps the environment allocate inference-serving resources efficiently as traffic changes.
HPE AI Essentials provides model deployment and autoscaling configuration capabilities through its integrated inference-serving infrastructure. Administrators can define supported scaling thresholds, replica limits, and resource requirements.
HPE AI Essentials scheduler priorities, represented by option D, establish the relative importance of different workload categories. Production inference workloads can be assigned higher priority than experimental notebooks.
When GPU resources become constrained, the scheduling framework uses these priority settings to determine allocation and preemption decisions. Higher-priority inference workloads can therefore take precedence over lower-priority notebook workloads, subject to the scheduler's supported preemption behavior.
Option A is not the primary mechanism for demand-based inference scaling or GPU priority management. Option C concerns controlled application-version rollouts and traffic distribution rather than automatic resource allocation.
The combined use of KServe autoscaling and scheduler priorities addresses both the changing inference demand and the requirement to prioritize production workloads.
Official HPE Reference: HPE AI Essentials Software — GPU Resource Management; GPU Scheduling Workload Scenarios; MLIS Object Model Reference Schema — Deployment AutoScaling.
HPE AI Essentials Software 1.7.x Documentation
Submit