Nexenta Systems, Inc.

Robert E Novak

Caitlin Bestler

Alexander Aizman

The present disclosure provides methods and systems for multicast collaborative erasure encoding and methods and systems for distributed parity protection. One embodiment relates to a method of multicast collaborative erasure encoding of a chunk stored in a distributed object storage cluster. A roll-call request is multicast to every storage server in a negotiating group for the chunk. Roll-call inventory responses are generated and multicast by every storage server in the negotiating group. The roll-call inventory responses are collected by every storage server in the negotiating group from other storage servers in the negotiating group to form a set of roll-call inventory responses. A logical evaluation of the set of roll-call inventory responses may then be performed by every storage server in the negotiating group. Other embodiments, aspects and features are also disclosed.

Multicast collaborative erasure encoding and distributed parity protection

The present disclosure provides systems and methods for Key-Value-Tuple-encoded (KVT-encoded) object storage. In accordance with an embodiment of the invention, chunks of objects, storing payload data or metadata, are persistently stored by servers using KVT-encoded storage. Furthermore, the KVT encoding may be applied to advantageously re-structure content within a distributed object storage cluster, especially for object storage systems that allow payload references to be cached extensively. For such systems, it is of considerable value to honor (use as valid) existing chunk references after the underlying content has been re-structured. Further, the KVT encoding taught herein is completely compatible with a fully-distributed object storage cluster. One embodiment relates to a storage server that includes a persistent storage module that provides a key-value application programming interface that encodes multiple key-value-tuple (KVT) entries for one chunk. Other embodiments, aspects and features are also disclosed.

Methods and systems for key-value-tuple-encoded storage

The present disclosure provides an innovative technique for parallel transparent restructuring of immutable content in a distributed object storage system. The content restructuring technique disclosed herein may be implemented with parallel operations by multiple storage servers in a live system. The restructuring is transparent in that the original CHIT may still be used to retrieve and validate the original content, such that a client requesting the original content need not be aware that the original content has been restructured. Furthermore, the restructuring is performed while maintaining support for pre-existing immutable metadata. Other embodiments, aspects and features are also disclosed. The content remains immutable from the perspective of the end user, but still may be advantageously restructured to optimize storage efficiency.

Parallel transparent restructuring of immutable content in a distributed object storage system

The present disclosure also provides systems and methods for sharding objects stored in a distributed storage system. In accordance with one embodiment disclosed herein, a key sharding technique is used. Key sharding is an advantageously efficient technique when dealing with an object containing a collection of key-value records. In accordance with an embodiment of the invention, referenced chunks identified by the key shards may each store a subset of the collection of the key-value records, and the key-value records in the subset have key hashes that have a range of matching bits in common. One embodiment disclosed herein provides a method of performing a delta edit of a named object stored in a distributed storage system in which a payload of the named object is stored in key shards. Other embodiments, aspects and features are also disclosed.

Methods and systems for key sharding of objects stored in distributed storage system

The present disclosure provides systems and methods for sharding objects stored in a distributed storage system. Such sharding may be advantageously utilized for an object that stores a collection of key-value records and for otherwise encoded objects. The object sharding techniques disclosed herein advantageously enable a read-modify-write process at the storage server level. One embodiment disclosed herein provides a method of creating a new chunk by modifying a payload of an existing chunk at the storage server level in a distributed object storage system. The method includes validating the new chunk at the gateway server. Other embodiments, aspects and features are also disclosed.

Read-modify-write processing of chunks at the storage server level in a distributed object storage system

The present invention relates to object storage systems that support hierarchical directories within a namespace manifest stored as an object. The namespace manifest is stored as namespace manifest shards that are determined based on a partial key derived from the full name of the referenced version manifest in each directory entry. Each storage server maintains a local transaction log that keeps track of changes to the namespace manifest in response to put transactions. The namespace manifest can be updated after each put transaction is acknowledged, which minimizes the number of write operations required for a put transaction. Updates to the namespace manifest are batched, distributed, and processed concurrently and in parallel by the storage servers that store the corresponding shards. This reduces the latency that would otherwise exist in the object storage cluster as a result of a put transaction. Updates to namespace manifest shards optionally occur in a batch process using MapReduce techniques.

Object storage system with local transaction logs, a distributed namespace, and optimized support for user directories

Embodiments disclosed herein provide a scalable multicast transport. The multicast transport protocol provides effectively reliable multicast delivery while avoiding the overhead associated with point-to-point protocols. Additional embodiments disclosed herein relate to a scalable object storage system that uses a multicast transport. The object storage system assigns responsibility for providing storage services for a chunk to a negotiating group of storage servers in the cluster using a shared and distributed hash allocation table. The object storage system dynamically determines a rendezvous group of storage servers in the cluster to store the chunk using the multicast transport. Other embodiments, aspects and features are also disclosed.

Scalable object storage using multicast transport

Embodiments disclosed herein provide advantageous methods and systems that use multicast communications via unreliable datagrams sent on a protected traffic class. These methods and systems provide effectively reliable multicast delivery while avoiding the overhead associated with point-to-point protocols. Rather than an exponential scaling of point-to-point connections (with expensive setup and teardown of the connections), the traffic from one server is bounded by linear scaling of multicast groups. In addition, the multicast rendezvous disclosed herein creates an edge-managed flow control that accounts for the dynamic state of the storage servers in the cluster, without needing centralized control, management or maintenance of state. This traffic shaping avoids the loss of data due to congestion during sustained oversubscription. Other embodiments, aspects and features are also disclosed.

Growing community of inventors

Union City, CA, United States of America

Robert E Novak