Comparing GlusterFS performance as a persistent storage layer for containerized applications on bare metal vs. PaaS deployments

By Amrik Jalif, Head of Storage UK&I, Red Hat

Container technology offers significant improvements in application density and deployment time. Containers have the versatility and power to do to virtualized environments what virtualization did to “a single server for every app.” Density goes up from 20 virtual machines per node to the equivalent of 50. Managing storage in a containerized world should be easy and automated. By containerizing the storage layer, developers have much more control over their environments.

In previous posts, we’ve outlined how Red Hat Storage can offer significant benefits to enterprises looking to deploy applications in containerized and PaaS environments, such as OpenShift Enterprise. We now have benchmark studies that compare performance between accessing GlusterFS from applications running on bare-metal x86 servers vs. from application containers in a PaaS environment, as in in OpenShift Enterprise.

We used the Flexible IO (FIO) tool in distributed mode as the workload generator for random and sequential workloads. We scaled OpenShift pods starting from 5 to 500, varying the file size and number of jobs in each pod while the dataset size was kept constant at 400GB. For a true “apples to apples” comparison, we ran distributed FIO on 6 OpenShift pods and on 6 bare-metal clients and scaled the pods from 5 to 1000 to push the limits on performance.

Continue reading “Comparing GlusterFS performance as a persistent storage layer for containerized applications on bare metal vs. PaaS deployments”