Amazon Technologies, Inc.

Sameer Choudhary

Nikhil Shah

Vishwas Narendra

James Zuber

Phillip H Pruett, Iv

Bryan T Wilkinson

Tyler Michael Jung

Douglas Stewart Laurence

Ernest S Cohen

Arvinth Ravi

Fusheng Yuan

Kaiwen Qu

Ning Liao

Ruiwen Zhao

Josef Schiefer

Michael Banfield

Nishant Jain

Slavcho Georgiev Slavchev

Suchindra Yogendra Agarwal

Abhishek Kannan

John Pender

Christopher Andrew Stephens

Prathiban Mohanasundaram

Hardik Navin Nagda

Karishma Chawla

Karmveer Veer Singh

Vijayasarathy Kannan

Nicholas Jacob Essenburg

Liming Ye

Janko Jerinic

Rajesh Sampath

Pierre Vigneras

Milos Faro

David Grant Emeny-Smith

Wilson Li

Tianxing Dun

Ozan Okumusoglu

Alexander Koshelev

Mehak Mehta

Addison Joseph Burns

Preetham Kowshik

RaghuKishore Balivada

Harishkumar Katagal

Systems and methods are provided for using request rate data to determine when and how to split storage partitions in a distributed storage system. Data objects in a given partition are stored and managed using a set of computing resources, such as caches, a witness service, and persistent storage nodes. When the request rate for the partition is too great, then the partition may be split in order to distribute the request load over two sets of computing resources. Some requests that would be able to be handled by the cache are passed through to the persistent storage nodes, which can track the requests at a granular level. Based on this data, specific partition splitting decisions can be made.

Data item and witness service partitioning in a distributed storage system

Systems and methods are provided for efficiently maintaining a transaction sequence witness service to ensure that requests to read data provide consistent results across all storage nodes in a distributed system. Each storage node that stores data in response to a particular update event may first update the transaction sequence witness service with the sequence number for the update event. Thus, any other storage node that services a subsequent request for the data will be able to see, via the transaction sequence witness service, the sequence number that must be available on the storage node to respond to the request.

Data event management for monotonic read consistency in a distributed storage system

Systems and methods are provided for efficiently maintaining a transaction sequence witness service to ensure that requests to read data provide consistent results across all storage nodes in a distributed system. When a storage node receives a request for a data object, the node may obtain&#x2014;from transaction sequence witness service&#x2014;the latest sequence number associated with the object. If the sequence number indicates the object has been modified in the period of time since the currently stored version was stored, the node can obtain the current version from persistent storage.

Witness service for ensuring data consistency in a distributed storage system

Systems and methods are provided for managing the order of data written to a transaction log in a distributed storage system. In a system with multiple nodes, if sequencers are naively generated without taking into consideration inconsistencies among the different nodes generating the sequencers, then the sequencers may not increase for each data transaction. To alleviate this problem, the node committing the transaction to the transaction log may perform a consistent read and verifies that the sequencer advances. If the sequencer does not advance, the node can perform a context-dependent operation such as adjusting the sequencer, acknowledging the write without committing it to the transaction log, or rejecting the write altogether.

Transaction management for monotonic write consistency in a distributed storage system

Systems and methods are provided for using a distributed cache architecture with different methods to load balance requests depending upon whether a requested data item is a freely-requested item (e.g., a 'hot key'). The cache may be implemented as a consistent hash ring, and most keys may be assigned to particular node based on a consistent hash. For hot key requests, the requests may be distributed among a subset of nodes rather than being assigned to a specific node using consistent hashing. When a witness service is used to ensure that cached data is fresh, verification requests for data regarding hot keys may be batched to avoid overloading the witness service with hot key requests.

Multi-tier cache for a distributed storage system

Unstructured data items are stored in an item collection. An access request comprising an input interpretation rule and a filtering criterion is directed to at least a portion of the item collection. Records are extracted from an unstructured data item using the input interpretation rule. A response to the access request is obtained, based at least partly in a determination that one of the records satisfies the filtering criterion.

Server-side filtering of unstructured data items at object storage services

Unstructured data items are stored at an object storage service. A filtering criterion to be used to generate a result set for an access request is determined. A test that can be used to determine, without completing parsing of a record identified in an unstructured data item, whether the record satisfies the filtering criterion is identified. Parsing of a particular record is abandoned in response to determining, using the test, that the record satisfies the filtering criterion. A response to the access request is determined using a subset of records that satisfy the filtering criterion.

Filtering pipeline optimizations for unstructured data

Growing community of inventors

Seattle, WA, United States of America

Sameer Choudhary