The Data Center File System

Distributed File System

Parallel POSIX file system volumes, that can be mounted to any number of hosts and include:

• File locking, incl. range locking and unlock-to-lock consistency
• Shared ACLs across Linux, NFS, Windows, OS X, and S3 – incl. NFSv4 ACLs
• Extended attributes (xattrs)
• Automatic locking for applications that are not cluster aware
• Metadata snapshots
• Striping: GB/s for individual files
• Block size configurable down to file level (512 B to 2 MB)

Unif​ied Access

Unified multi-interface access with common data, metadata, access control, ACLs, and locks:

• Native user-land client for Windows, Linux, and macOS (automatic fault-tolerance and parallel performance).
• Simultaneous access to the same file from all interfaces
• NFS v3/v4
• SMB via Windows Server or Samba
• High-performance, stateless S3 incl. web browser-based client
• HDFS (Hadoop Filesystem)

File, Block, and Object Workloads within a Single System

High-performance storage for block workloads (VMs and databases):

• High-concurrency random block IO with sub-millisecond latency.
• High random IOPS and throughput on the same file system.
• Native Qemu integration

Service provider-grade object storage with S3 interface:

• Extensive S3 feature coverage
• Authentication via LDAP, Keystone, and password files
• Stateless, scalable architecture

Master small-file workloads with ultra-low latency

100% Software for Diverse Deployment Models

• Runs on any x86 server, dedicated or converged. All major Linux distributions supported, no kernel components required.
• Connect storage to a public cloud (AWS, Google Cloud Platform, Microsoft Azure).
• No RAID necessary. All redundancy is done in software, across servers. No NVRAM necessary.
• SAS/SATA HDDs and SSDs, NVMe, SMR drives
• Fast network (10GbE+ recommended)
• Storage cluster or hyperconverged, can be spread across rooms or geographically separated datacenters, all in one rack or a federation of clusters.

Linear Scalability and Performance

• Direct communication from client to storage servers enables linear scalability.
• Throughput and IOPS scale linearly with hardware resources.
• Single file IO can scale to entire cluster bandwidth.
• Sub-millisecond latency
• Any number of clients or servers (starting with 4).
• Any number of volumes.
• Any number of tenants.
• GB/s on single files.

Data Safety, Redundancy, Availability, and Integrity in Software

• No single point of failure and split-brain safe:
• Non-disruptive updates with zero downtime. Mixed version deployments possible.
• Automatic failover in software, no extra HA setup or cost
• Hardware failures do not threaten data consistency
• Redundancy in software eliminates RAID:
• Replication: quorum-based replication for strong data consistency; the entire replication stack is split-brain safe.
• Erasure coding: space-efficient alternative to replication for sequential writes and high-throughput workloads.
• Asynchronous replication through volume mirroring for better disaster recovery
• End-to-end checksums protect data and metadata in transit and at rest

Security

• IP-based and X.509-based access control
• Integrated certificate management
• Management authentication via LDAP, OpenStack Keystone or integrated user management
• S3 authentication via LDAP, OpenStack Keystone or password files
• Multi-tenancy on the logical and physical level
• At-rest data encryption
• NFSv4 ACLs