RADOS

RADOS
Name	RADOS
Type	Distributed object store
Developer	Red Hat
First release	2010
Stable release	15.2.0 (example)
Written in	C++
License	LGPLv2.1
Website	Ceph Project

RADOS

RADOS is a distributed object storage system designed to provide reliable, scalable block, object, and file storage for large-scale deployments. It operates as the foundational storage layer for projects such as Ceph and integrates with orchestration and virtualization projects like OpenStack, Kubernetes, and libvirt. Originally developed by contributors from Red Hat and academic collaborators, RADOS implements cluster-wide placement, replication, and recovery mechanisms to serve infrastructure needs in environments influenced by designs from Google File System, Amazon S3, and academic research such as Paxos and Raft-inspired consensus work.

Overview

RADOS presents a symmetric, peer-to-peer architecture in which each storage node participates as an equal member of a cluster, avoiding centralized metadata servers common to systems like GlusterFS or traditional Network File System. Clients interact with RADOS via libraries and gateways such as librados, rbd, and rgw, enabling integration with projects similar to CephFS and object APIs compatible with Amazon S3 and OpenStack Swift. The system emphasizes self-healing through automated rebalancing and recovery procedures, and supports erasure coding and replication strategies comparable to those used in large deployments by organizations like Facebook and Netflix.

Architecture and Components

RADOS clusters are composed of OSD daemons running on storage nodes, monitors responsible for cluster membership and mapping, and metadata services when layered with file systems such as CephFS. OSDs (Object Storage Daemons) manage physical devices and implement the internal object store; monitors (ceph-mon) maintain cluster maps and election processes similar to consensus services like etcd and Zookeeper. Gateways like radosgw provide HTTP REST interfaces compatible with Amazon S3 and OpenStack Swift, while client libraries such as librados and librbd expose APIs used by projects like QEMU, Docker, and Kubernetes CSI. Tools for orchestration and deployment integrate with systems like Ansible, SaltStack, and orchestration platforms including OpenShift.

Data Placement and Striping

RADOS places objects using a deterministic placement algorithm, the CRUSH algorithm, which maps objects to OSDs based on cluster topology and rules rather than centralized lookups. CRUSH enables administrators to define failure domains such as datacenter, rack, host, and disk—terms relevant to deployments in environments run by organizations like Google and Microsoft Azure—and supports data distribution strategies including replication and erasure coding. Striping across placement groups and backends allows workloads from storage consumers like KVM instances, CephFS clients, and S3-style objects to achieve balanced utilization, mimicking principles used in systems such as Hadoop Distributed File System for large-file distribution.

Reliability, Replication, and Failure Recovery

RADOS ensures durability through configurable replication factors or erasure coding pools, offering trade-offs between storage overhead and resilience patterns analogous to techniques adopted by RAID arrays and hyperscalers. Monitors coordinate membership changes and trigger reweighting and rebalancing when OSDs or hosts fail, while OSDs perform scrubbing and backfilling to maintain data integrity. Recovery and consensus mechanisms draw conceptual parallels with distributed agreement protocols developed in academic circles such as Paxos and Raft, and are implemented to handle split-brain and network-partition scenarios encountered in large deployments by enterprises like IBM and Dell EMC.

Performance and Scalability

RADOS scales horizontally by adding OSD nodes and expanding placement groups, similar to scaling strategies in Apache Cassandra and Cassandra Ring-based designs. Performance tuning involves adjusting parameters for journaling, cache tiers, and network fabrics such as InfiniBand or 10/25/40/100 Gigabit Ethernet used in datacenters run by companies like NVIDIA and Intel. Features like client-side caching, SSD journaling, and hybrid tiers enable low-latency access patterns required by virtualization stacks including OpenStack Nova and container runtimes. Benchmarks for block storage (via rbd), object throughput (via radosgw), and filesystem workloads (via CephFS) are commonly compared to alternatives such as LVM and ZFS.

Use Cases and Integrations

RADOS underpins use cases spanning cloud infrastructure, archival storage, and media delivery; notable integrations include block storage for OpenStack Cinder, image storage for OpenStack Glance, and container storage via Kubernetes CSI drivers. It is used in production by telecommunications carriers, research institutions, and cloud providers alongside orchestration tools like Terraform and monitoring stacks such as Prometheus and Grafana. Gateways and adapters enable compatibility with object storage clients built for Amazon S3 and OpenStack Swift, making RADOS suitable for web-scale object hosting, backup systems in enterprises like HP Enterprise, and high-throughput analytics workloads relying on frameworks like Apache Spark.

Development, APIs, and Administration

Development occurs within community repositories linked to projects such as Ceph and coordinated by contributors from companies like Red Hat and academic partners. Primary APIs include librados, librbd, and radosgw S3/Swift endpoints; administrative tooling consists of cephadm, ceph-deploy, and orchestration integrations for platforms like Kubernetes and OpenShift. Operational tasks—health checks, configuration tuning, and upgrade orchestration—leverage monitoring integrations with systems like Nagios and alerting via PagerDuty in enterprise operations. Continuous development tracks feature additions for erasure coding, snapshotting, and encryption compatible with standards utilized by vendors including Seagate and Western Digital.

Category:Distributed file systems