Apache YuniKorn

Apache YuniKorn is a lightweight, universal resource scheduler designed for container orchestrator systems like Kubernetes. It provides fine-grained resource sharing and efficient scheduling for batch, data, and machine learning workloads in multi-tenant, cloud-native environments. YuniKorn replaces the default Kubernetes scheduler with a more powerful alternative that supports hierarchical queues, resource fairness, and application-aware scheduling.

Features

• Application-aware scheduling with context of users, apps, and queues
• Hierarchical resource queues for fine-grained resource control
• Gang scheduling to reserve and allocate resources for entire applications
• Job ordering policies including FIFO, Fair, Priority, and StateAware
• Pluggable node sorting policies and preemption support
• Centralized UI for monitoring cluster, queue, and application resource usage
• ARM and x86 architecture support with prebuilt images

Capabilities

• Full compatibility with Kubernetes workloads
• Transparent integration with existing Kubernetes applications
• Elastic resource management with cloud autoscaler support
• Efficient bin-packing and proactive up-scaling for cost optimization
• Centralized dashboard for resource planning and monitoring
• Configurable reservation behavior and multi-gang scheduling

Benefits

• Improved scheduling predictability and fairness across tenants
• Enhanced resource utilization and reduced fragmentation
• Simplified client-side operations with queue-based resource management
• Cost savings through optimized cloud resource usage
• Better control over resource quotas and priorities
• Scalable and flexible deployment for both on-prem and cloud environments