Ephemeral Storage

Cloudera Operational Database now supports NVMe (Non-volatile Memory Express) based cache that significantly improves the performance when you deploy Cloudera Operational Database with object storage.

Accessing cloud storage is generally slower than block storage, as data in such deployments is not local to HBase. Our benchmarks show that Cloudera Operational Database over cloud storage performs, on average, five times slower than Cloudera Operational Database clusters using HDFS on block storage with HDD. To address this penalty, Cloudera Operational Database uses ephemeral storage (NVMe disks available with all major cloud providers) to deploy RegionServers with file base bucket cache on the ephemeral volumes. With all data cached locally in the ephemeral disks, Cloudera Operational Database on cloud storage can achieve performance parity compared to Cloudera Operational Database with HDFS on block storage, but at a lower cost.

Once you enable the entitlement for your tenant, Cloudera Operational Database automatically configures the newly created cluster for this feature. It also allows the auto-scale functionality based on the available cache space. For more information, see Auto-scaling in public cloud environments. No additional configuration is needed.

AWS🔗

Cloudera Operational Database supports a 1.6 TB NVMe based cache that significantly improves the performance when you deploy Cloudera Operational Database with S3.

Azure🔗

The NVMe-based cache is now enabled on Azure too. It uses a high-performance AMD instance type with a 2 TB NVMe disk for the worker nodes.

GCP🔗

Cloudera Operational Database supports a 3 TB NVMe-based cache that uses local SSD for ephemeral storage on GCP. This significantly enhances the performance when you deploy Cloudera Operational Database on GCP.