Adaptive Serverless Platform
- Home >
- Product >
- CLOUD-NATIVE ARCHITECTURE >
- Adaptive Serverless Platform
Qubole delivers self-service machine learning and analytics with its Adaptive Serverless Platform architecture. It automatically provisions, manages and optimizes cloud resources balancing cost, workloads, and performance requirements.
Automated Cluster Lifecycle Management
Qubole’s automated cluster lifecycle management protects you from incurring cloud compute costs associated with idle clusters. Qubole automatically shuts down a cluster without the risk of data loss when there are no more active jobs, thus, you only pay for the compute you use.
Cluster Lifecycle Management Savings (CLCM)
Qubole terminates the cluster when there is no activity
Workload Aware Autoscaling Savings (WAAS)
Nodes are added on-demand to cluster only when the workload requires more resource capacity
Workload-Aware Autoscaling
Unlike vanilla autoscaling, Qubole’s workload-aware autoscaling upscales, downscales, and rebalances clusters with a complete context of the workload, SLA, and priority for each job running on the cluster.
Qubole’s architectural approach to autoscaling results in significant benefits in reliability, cost, and responsiveness for all applications processing large datasets.
Intelligent Spot Management
Qubole’s Intelligent Spot Management allows organizations to optimize the use of AWS Spot instances, resulting in cost savings of up to 80% on AWS compute costs. Spot instances require careful management when bursty data workloads need to scale at a moment’s notice. Qubole provides policy-based automation of Spot instance management to balance performance, cost, and SLA requirements.
Spot Shopper Savings
Save up to 80% with Spot instances
Heterogeneous Cluster Configuration
Qubole’s Heterogeneous Cluster Configuration for on-demand and Spot nodes allows you to pick the most cost-effective combination for your job. Qubole enables you to configure heterogeneous clusters by mixing nodes of multiple instance types, delivering much greater data processing efficiency.
Homogeneous clusters are not optimal for bursty big data workloads, as the availability of instances varies considerably and could result in significant delays.
Aggressive Cluster Downscaling
Qubole’s Aggressive Cluster Downscaling prevents cost overruns by shutting down idle clusters. Qubole uses intelligent self-learning algorithms such as Smart Victim Selection, Graceful Downscaling, and Container Packing to balance workloads across active nodes and decommission idle ones without the risk of data loss.
Without Aggressive Downscaling
Node Utilization vs. Time
With Agressive Downscaling
Node Utilization vs. Time
