Overall, our performance testing revealed several areas that you need to consider before deploying OneFS clusters of APEX File Storage for AWS, so that the clusters can meet your organization’s performance requirements. The following are the key findings:
- Instance type: For sequential read and write workloads, scaling up the instance size from m6idn.8xlarge to m6idn.12xlarge, provides you scaling performance.
- Cluster size: Clusters with more nodes can serve higher throughput workloads. Throughput performance grows when adding nodes to the cluster.
- Aggregated EBS volume throughput per node: When determining the specific configuration of a cluster, it is a best practice to make the aggregated EBS volume bandwidth the same as the instance-level EBS bandwidth limit. The instance aggregated EBS bandwidth is calculated using the following formula:
Aggregated EBS bandwidth = (Number of volumes) * (Defined throughput per volume)
For example, the m6idn.12xlarge EBS bandwidth limit is 37.5 Gbps, which is 4687 MB/sec. If this node contains 12 EBS volumes, it is best practice to set the max throughput of each EBS volume to be 391 MB/sec (391 * 12 = 4692).
- Cluster volume type: gp3 cluster can effectively balance performance and cost for both sequential reads/writes and metadata intensive workloads. While st1 cluster can effectively balance storage capacity and cost, it is suitable for archive workloads.