Storage requirements

You must understand the storage requirements for StorageGRID Webscale nodes.

StorageGRID Webscale nodes require three logical categories of storage:

You must use RAID-backed block devices (LUNs/volumes) for all storage categories. Non-redundant disks, SSDs, or JBODs are not supported. You can use shared or local RAID storage for any of the storage categories; however, if you want to use StorageGRID Webscale’s node migration capability, you must store both system metadata and object data on shared storage. (For more information about node migration, see "Node migration requirements" and the Recovery and Maintenance Guide.)

The following table shows the minimum storage space required for each type of node in each storage category.

Type of node Container pool System metadata Object data
Admin 100 GB 390 GB 0 GB
API Gateway 100 GB 90 GB 0 GB
Storage 100 GB 90 GB 4,000 GB
Archive 100 GB 90 GB 0 GB

Use the table to determine the minimum amount of storage that you will need to provide on each host in each category, based on the number and type of StorageGRID Webscale nodes you plan to run on that host. For example, if you plan to run one Admin Node, one API Gateway Node, and one Storage Node on a single host, you must provide a minimum of 300 GB of container pool storage, 570 GB of system metadata storage, and 4,000 GB of object data storage on that host.

The following table shows the number and size of block storage volumes (LUNs) that must be made available on an individual host, where appropriate, as a function of the number and type of nodes that will be deployed on that host.

Note: The numbers are for each host, not for the entire grid.
LUN purpose Storage category Number of LUNs Minimum size/LUN
Docker storage pool Container pool 1 NTotal × 100 GB/node
Per-node /var/local volume System metadata NTotal 90 GB
Admin Node audit logs System metadata NAdmin 200 GB
Admin Node tables System metadata NAdmin 100 GB
Storage Nodes Object data NStorage × R 4,000 GB

(see "Storage requirements for Storage Nodes")

In the table:
Use the table to determine the number and size of the block storage volumes you must assign to each host to support the desired StorageGRID Webscale nodes. For example, suppose you plan to run one Admin Node, one API Gateway Node, and one Storage Node on a single host. In this scenario, you must provide a minimum of seven block storage volumes to the host, as follows:

Storage requirements for Storage Nodes

When assigning space to Storage Nodes, be aware that StorageGRID Webscale reserves 2 TB of space on each Storage Node for the Cassandra database. This database stores all object metadata as well as certain configuration data.

Cassandra data is stored on the first storage volume (rangedb0) of each Storage Node. Any additional capacity in this volume is used for object storage.
  • Normally, you should use at least three storage volumes for each Storage Node. Each storage volume should be 4 TB or larger.

  • If you use only one storage volume for each Storage Node, you must assign it at least 4 TB of space.
    Note: If you use only one storage volume for the Storage Node and you assign 2 TB or less to that volume, the Storage Node immediately enters the Storage Read-Only state on startup and stores object metadata only.
  • If you use more than one storage volume for each Storage Node, you must assign at least 2 TB to rangedb0; however, 4 TB is recommended. Using 4 TB ensures that rangedb0 has adequate capacity for object metadata and that it can also be used for object data.

Performance requirements

The performance of the volumes used for the container pool, system metadata, and object metadata (rangedb0) significantly impacts the overall performance of the system. You should use performance-tier (10K SAS or SSD) storage for these volumes to ensure that they provide adequate disk performance in terms of latency, input/output operations per second (IOPS), and throughput. You can use capacity-tier (NL-SAS/SATA) storage for the persistent storage of object data.

The volumes used for the container pool, system metadata, and object data must have write-back caching enabled. The cache must be on a protected or persistent media.