Architecture

This page explains how an Exaba deployment is put together: the node roles, drives, topology, and networking you plan and provide. Exaba installs and configures the cluster on top.

Nodes

An Exaba cluster is built from two node roles:

  • Compute nodes (C-nodes). Stateless nodes that serve the S3 client endpoints and handle erasure coding and data placement. Add C-nodes to scale S3 serving and keep compute from bottlenecking; aggregate throughput is typically bounded by network and drive bandwidth.
  • Storage nodes (D-nodes). Nodes that hold the drives and present them to the cluster. Add D-nodes to increase capacity.

The two roles can run converged (compute and storage on the same servers, simplest to operate and well suited to backup and capacity workloads) or disaggregated (compute and storage on separate servers, so each scales to the workload).

Drives

Each storage node uses two kinds of drive:

  • Metadata drives: fast NVMe drives (a minimum of two per node) that hold object metadata and the data staging area.
  • Data drives: hold the object data. Typically HDDs, which give the lowest cost per TB; use NVMe for read-latency-sensitive workloads (hot, small-object, AI/ML reads). With enough HDDs and staging headroom, write throughput is comparable across media.

For minimum and reference hardware specifications, see Hardware Sizing and the Compatibility Matrix.

Highly available vs. Edge

  • Highly available (HA) cluster. Starts at three nodes, with five or more recommended, running active/active with no single point of failure; the cluster survives node, drive, and rack failures. It scales out online by adding nodes and drives, with multi-cluster federation for larger footprints. This is the standard production deployment.
  • Edge / single node. A single node for a remote or branch site, or for evaluation, without cluster redundancy.

See Positioning for the full set of deployment models, including OEM and air-gapped.

Networking

Plan three separate networks:

  • Client network: external S3 API traffic from your applications, typically behind Exaba’s bundled load balancer.
  • Data network: a high-bandwidth, low-latency internal fabric carrying storage traffic between nodes over NVMe-oF, on RDMA (RoCEv2 or iWARP) or TCP.
  • Management network: out-of-band hardware management (BMC).

What you provide, what Exaba provides

  • You provide: commodity x86_64, ARM, or Power servers, their drives, and the network fabric.
  • Exaba provides: the software, and installs, configures, and (for connected clusters) monitors the deployment.