XtreemFS implements striping of files over several storage servers for increased I/O bandwidth and capacity. This is similar to RAID0. Paralell I/O ensures that read/write operations are executed in parallel on the storage servers to exploit the full bandwidth.

XtreemFS implements POSIX interfaces and semantics. It has the same "look and feel" as a local file system or NFS. Applications do not need to be modified to take advantage of advanced features like replication or striping. Since release 1.1 this includes also support for locking (fcntl locks).

Read-only replication allows you to create copies of files on several OSDs. Since the files are immutable (read-only) there is no overhead for coordinating the replicas. We also support partial replicas, only the data requested by a client is fetched from other replicas. For a demonstration see our screencast.

Automatic "replicate-on-close" ensures that you can replicate files from applications that require read/write access to files.

With XtreemFS all replicas are initially empty and can be used immediately by applications. Depending on the strategy the replica is filled up automatically or on-demand as applications read data. This helps you to reduce start-up times, network bandwidth and disk usage.

XtreemFS can work with plain TCP connections or over SSL. XtreemFS supports X.509 certificates for user and service authentication. Our plug-in architecture allows you to easily create your own X.509 cert handlers to adapt XtreemFS to your security infrastructure.

XtreemFS clients and servers can be distributed world-wide. Several features will make your life easier when working over the internet such as caching in the client and file replication. SSL and X.509 support means that you can use XtreemFS over the internet without a VPN. Simply mount XtreemFS from your laptop to acccess your files no matter where you are.

Objects on the storage devices can be checksummed to detect data corruption of file content automatically. This will be extended in future versions to full end-to-end checksums.

XtreemFS is extensible. You can modify the behaviour of XtreemFS to better integrate with e.g. your security infrastructure. For more information see our developers guide.

XtreemFS supports various ways to automatically select OSDs and replicas close to the client creating or opening a file. We support Vivaldi (dynamic network coordinates), Datacenter Maps (static configuration based on IP networks) and FQDN match (distance based on match of hostname of client and OSDs).

With asynchronous MRC backups you can easily create a consistent snapshot of your metadata without interrupting operations. The MRC will do the backup in background and continue to process client requests.

The XtreemFS client can cache metadata to speed up interactive use, e.g. from your shell. This feature is particularly useful if you mount volumes over the WAN or slow links like DSL.

Starting with release 1.2 XtreemFS includes a web-based admin console. It gives you an overview of the system and makes admin tasks, like changing policies, even easier.

We have developed an experimental FileSystem driver for Hadoop. This client is written purely in Java and does not require the FUSE client to be mounted. Like the regular XtreemFS client, this driver does support parallel read and write operations.

Replication of the metadata server (MRC) and directory service (DIR) makes XtreemFS fully fault-tolerant. The replication will be master/slave based and the automatic failover will ensure that a new master is elected once the current master crashed or gets disconnected.

Full read/write support for replicated files complements the read-only replication. This replication mode allows you to replicate files with regular POSIX semantics. For technical details see also our publications.

Consistent backups allow you to create a backup of your volume based on a snapshot, even when the volume is modified. Your backup will only contain consistent versions of files.

XtreemFS will support versioned files and asynchronous metadata snapshots. Together these two techniques allow you to create snapshots of a volume instantaniously.

We plan to add services which automatically replace replicas when an OSD or MRC fails. In addition, we investigate methods to include automatic replica creation and deletion, e.g. based on file popularity.