Appendix A. Additional information
A.1. Configuration guidance
The following configuration guidance is intended to provide a framework for creating Hyperconverged Infrastructure environments. This guidance is not intended to provide definitive configuration parameters for every Red Hat OpenStack Platform installation. Contact the Red Hat Customer Experience and Engagement team for specific guidance and suggestions that fit your specific environment.
A.1.1. Cluster sizing and scale out
The Red Hat Ceph Storage Hardware Guide provides recommendations for IOPS optimized, throughput optimized, and cost and capacity optimized Ceph deployment scenarios. Follow the recommendation that best represents your deployment scenario and add the NICs, CPUs, and RAM required to support the Compute workload.
An optimal, small footprint configuration consists of seven nodes. Unless you have a requirement for IOPS optimized performance in your environment and you are using all flash storage, the throughput optimized deployment scenario should be used.
Three node Ceph Storage cluster configurations are possible. In this configuration, you should:
- use all flash storage.
-
set the
replica_countparameter to 3 in theceph.conffile. -
set the
min_sizeparameter to 2 in theceph.conffile.
If a node leaves service in this configuration, IOPS continue. To retain 3 copies of the data, replication to the third node is queued until it returns to service. Data is then backfilled to the third node.
HCI configurations of up to 64 nodes have been tested. Some HCI environment examples have been documented up to 128 nodes. Large clusters such as these can be considered with a Support Exception and Consulting Services engagement. Contact the Red Hat Customer Experience and Engagement team for guidance.
A deployment with two NUMA nodes can host a latency sensitive Compute workload on one NUMA node and Ceph OSDs services on the other. If there are network interfaces on both nodes, and the disk controllers are on node 0, use a network interface on node 0 for the Storage network and host the Ceph OSD workload on node 0. Host the Compute workload on node 1 and configure it to use the network interfaces on node 1. When acquiring hardware for your deployment, be mindful of which NICs will use which nodes and attempt to split them between storage and workload.
A.1.2. Capacity planning and sizing
The throughput optimized Ceph solution defined in the Red Hat Ceph Storage Hardware Guide provides a balanced solution for most deployments that do not require optimization for IOPS. In addition to the configuration guidelines provided with the solution, note the following when creating your environment:
- The allotment of 5 GB of RAM per OSD ensures OSDs have sufficient operational memory. Ensure your hardware can support this requirement.
- CPU speed should match the storage medium in use. The advantages of faster storage mediums such as SSDs can be negated by CPUs too slow to support them. Similarly, a fast CPU can be more efficiently used by faster storage mediums. Balance CPU and storage medium speed so that neither becomes a bottleneck for the other.
