Rackspace Us, Inc.

Intel Corporation

Nvidia Corporation

Swiftstack, Inc.

Clay Gerrard

Gregory Lee Holt

Michael Barton

David Patrick Goetz

John A Dickinson

Samuel N Merritt

Paul E Luse

Paul Luse

Charles B Thier

Methods and apparatus to dynamically assign and relocate object fragments in distributed storage systems are disclosed. In some examples, an apparatus to compile fragments of an object includes a fragment compiler to: compile an object from fragments stored in storage nodes, respective ones of the fragments corresponding to (a) a node index of storage identifiers representative of the storage nodes and (b) a fragment index of fragment identifiers associated with the respective ones of the fragments of the object, respective ones of the fragment identifiers being representative of a sequential order of the fragments of the object, the respective ones of the fragment identifiers to be associated with the respective ones of the storage identifiers to enable verification of storage locations of the respective ones of the fragments of the object relative to respective storage nodes; request a first one of the fragments from a first one of the respective storage nodes; determine if a first fragment index assigned to the first one of the fragments matches a first node index assigned to the first one of the fragments; when the first fragment index matches the first node index, compile the first one of the fragments into the object based on the first node index; and when the first fragment index does not match the first node index, compile the first one of the fragments into the object based on the first fragment index.

Methods and apparatus to assign indices and relocate object fragments in distributed storage systems

A technique is applied to eventually converge on a single data storage strategy for any set of object data which had an inconsistent data storage strategy applied during storage while there was a network partition. This state could occur in instances of a highly available distributed object storage system which can store objects according to multiple data storage strategies. Upon the healing of a network partition, the technique discovers if multiple data storage strategies were applied to the object data stored during a network partition, deterministically identifies which data storage strategy represents the correct strategy, for example based on the log of state transitions requested by the client according to the API contract, and ensures that this strategy is consistently applied to all object data in the collection.

Reconciler for a distributed storage system

Methods and apparatus to dynamically assign and relocate object fragments in distributed storage systems are disclosed. In some examples, the methods and apparatus encode an object with error correction coding to separate the object into fragments, create a first index indicative of storage nodes where the fragments of the object are to be stored, encode a second index into identifiers of the fragments of the object, the second index based on the first index, and store the fragments of the object and the corresponding second index encoded identifiers in the storage nodes based on the first index.

An improved scalable object storage system allows multiple clusters to work together. In one embodiment, a trust and federation relationship is established between a first cluster and a second cluster. This is done by designating a first cluster as a trust root. The trust root receives contact from another cluster, and the two clusters exchange cryptographic credentials. The two clusters mutually authenticate each other based upon the credentials, and optionally relative to a third information service, and establish a service connection. Services from the remote cluster are registered as being available to the cluster designated as the trust root. Multi-cluster gateways can also be designated as the trust root, and joined clusters can be mutually untrusting. Two one-way trust and federation relationships can be set up to form a trusted bidirectional channel.

Cluster federation and trust in a cloud environment

In one embodiment, there is a multi-cluster synchronization system between two or more clusters. The multi-cluster synchronization system uses variable compression to optimize the transfer of information between the clusters. Compression is used not only to minimize the total number of bytes sent between the two clusters, but to dynamically vary the size of the objects sent across the wire to optimize for higher throughput after considering packet loss, TCP windows, and block sizes. This includes both the packaging of multiple small files together into one larger compressed file, saving on TCP and header overhead, but also the chunking of large files into multiple smaller files that are less likely to have difficulties due to intermittent network congestion or errors. A further embodiment uses forward error correction to maximize the chances that the remote end will be able to correctly reconstitute the transmission.

Higher efficiency storage replication using compression

An improved scalable object storage system includes methods and systems allowing multiple clusters to work together. Users working with a first cluster, or with a multi-cluster gateway, can ask for services and have the request or data transparently proxied to a second cluster. This gives transparent cross-cluster replication, as well as multi-cluster compute or storage farms based upon spot availability or various provisioning policies. Vendors providing a cloud storage 'frontend' can provide multiple backends simultaneously. In one embodiment, a multi-cluster gateway can have a two, three, or higher-level ring that transparently matches an incoming request with the correct cluster. In the ring, a request is first mapped to an abstract 'partition' based on a consistent hash function, and then one or more constrained mappings map the partition number to an actual resource. In another embodiment, the multi-cluster gateway is a dumb gateway, and the rings are located only at the cluster level.

Virtual multi-cluster clouds

An intercluster repository synchronizer and method for synchronizing objects are disclosed. An example intercluster repository synchronizer includes an information processing system, including a processor, computer-readable medium, and network device. The intercluster repository synchronizer includes a structured information repository on the computer-readable medium. The structured information repository contains a plurality of records corresponding to a selected group of stored information objects. The intercluster repository synchronizer further includes a synchronization indicator that stores an address associated with a remote replication target. The intercluster repository synchronizer also includes a replicator, operable to send a message using the network device to the replication target responsive to changes in the structured information repository, and further operable to receive a message that a plurality of stored information objects have been duplicated at the remote replication target. The duplicated stored information objects are selected based on a shared metadata indicator stored in the structured information repository.

Intercluster repository synchronizer and method of synchronizing objects using a synchronization indicator and shared metadata

Cluster federation and trust

An improved scalable object storage system includes methods and systems allowing multiple clusters to work together. In one embodiment, there is a multi-cluster synchronization system between two or more clusters. The multi-cluster synchronization system uses variable compression to optimize the transfer of information between the clusters. Compression is used not only to minimize the total number of bytes sent between the two clusters, but to dynamically vary the size of the objects sent across the wire to optimize for higher throughput after considering packet loss, TCP windows, and block sizes. This includes both the packaging of multiple small files together into one larger compressed file, saving on TCP and header overhead, but also the chunking of large files into multiple smaller files that are less likely to have difficulties due to intermittent network congestion or errors.

An improved scalable object storage system allows multiple clusters to work together. Users working with a first cluster, or with a multi-cluster gateway, can ask for services and have the request or data transparently proxied to a second cluster. This gives transparent cross-cluster replication, as well as multi-cluster compute or storage farms based upon spot availability or various provisioning policies. Vendors providing a cloud storage 'frontend' can provide multiple backends simultaneously. In one embodiment, a multi-cluster gateway can have a two, three, or higher-level ring that transparently matches an incoming request with the correct cluster. In the ring, a request is first mapped to an abstract 'partition' based on a consistent hash function, and then one or more constrained mappings map the partition number to an actual resource. In another embodiment, the multi-cluster gateway is a dumb gateway, and the rings are located only at the cluster level.

Growing community of inventors

San Francisco, CA, United States of America

Clay Gerrard