Microsoft Technology Licensing, LLC

Knut Magne Risvik

Trishul A Chilimbi

Karthik Kalyanaraman

John Grant Bennett

Michael Joseph Hopcroft

Vishesh Parikh

Jan O Pedersen

Chad P Walters

Alexander Shamis

Yutaka Suzue

Hui Shen

Richendra Khanna

Chiranjeeb Buragohain

Dushyanth Narayanan

Miguel Oom Temudo De Castro

Ming-Chuan Wu

Aleksandar Dragojevic

Christopher Walter Anderson

Matthew Renzelmann

Henry Setiawan

Mike Hopcroft

Timothy Tan

Methods and computer storage media are provided for generating entries for documents in a forward index. A document and its document identification are received, in addition to static features that are query-independent. The document is parsed into tokens to form a token stream corresponding to the document. Relevant data used to calculate rankings of document is identified and a position of the data is determined. The entry is then generated from the document identification, the token stream of the document, the static features, and the positional information of the relevant data. The entry is stored in the forward index.

Efficient forward ranking in a search engine

A distributed graph database that enables scaling and efficient processing is described. The distributed graph database can, for example, scale up to petabytes of data to enable transactional processing of graph data with low latency and low processing overhead. The distributed graph database can include a cluster of devices and a remote direct memory access (RDMA)-based communication layer to perform low latency messaging between devices of the cluster of devices. Additionally, the distributed graph database can include a shared memory layer that provides one or more data structures, a transaction layer to facilitate query processing, and a graph database layer stored in computer-readable media and executed on a processor to implement a graph data model. In at least one example, the graph data model can be mapped to the one or more data structures.

Distributed graph database

An 'RDMA-Based Memory Allocator' applies remote direct memory access (RDMA) messaging to provide fast lock-free memory allocations and de-allocations for shared memory distributed across multiple servers in an RDMA-based network. Alternately, in various implementations, the RDMA-Based Memory Allocator combines RDMA messaging and remote procedure call (RPC) requests to provide fast lock-free memory allocations and de-allocations for shared memory distributed across multiple servers in an RDMA-based network. In either case, any of the networked servers can act as either or both a client for requesting (or releasing) memory allocations and a host for hosting a portion of the distributed memory. Further, any server (including the requesting client) may act as the host for the distributed memory being allocated or de-allocated by any client via RDMA messaging. Advantageously, being lock-free improves overall performance of memory access between networked computers by reducing overall system latency.

Distributed lock-free RDMA-based memory allocation and de-allocation

A 'Distributed Storage Controller' applies an RDMA-based memory allocator to implement a distributed thread-safe and lock-free storage system in shared memory distributed across multiple networked computers. The Distributed Storage Controller enables any number of client applications, processes, subroutines, etc., on any number of networked computers to concurrently perform self-directed lock-free allocations, de-allocations, reads, writes, etc., on the shared memory via sequences of one-way RDMA messages (e.g., RDMA reads, RDMA writes, and RDMA atomic compare and swap (CAS) operations) without requiring CPU locks. As such, the CPUs of computing devices hosting the shared memory do not need to be notified of RDMA-based reads, writes or CAS operations on that memory. Consequently, CPU utilization for computing devices hosting any portion of the shared memory is minimal, thereby enabling the Distributed Storage Controller to scale to very large numbers of concurrent accesses by very large numbers of networked computing devices.

Low latency RDMA-based distributed storage

A search index includes tiered posting lists. Each posting list in the search index corresponds with a different atom and includes a list of documents containing the particular document. Additionally, a rank is stored with each document listed in a posting list for a given atom representing the relevance of the atom to the context of each document. At least some of the posting lists in the search index are tiered. A tiered posting list is divided into a number of tiers with the tiers being ordered by document while each tier is internally ordered by document. Employing tiered posting lists within the search index allows a search engine to evaluate search queries in a manner that allows for a number of efficiencies and precise stopping.

Tiering of posting lists in search engine index

Methods and systems are provided for using a hybrid-distribution system to identify relevant documents based on a search query. A group of documents is assigned to a particular segment. The group of documents is indexed both by atom and by document to form a reverse index and a forward index. Both indexes are divided amongst each node in that segment so that each node is responsible for storing and accessing a different portion of both the reverse and forward indexes. The reverse index portion is accessed on each of a first set of nodes to identify a first set of documents that is relevant to a particular search query. Document identifications associated with the first set of documents are used to identify a second set of nodes that access their forward index portions to limit the number of relevant documents to a second set of documents.

Hybrid-distribution model for search engine indexes

Methods are provided for populating search indexes with atoms identified in documents. Documents that are to be indexed are identified, and for each document, atoms are identified and are categorized as unigrams, n-grams, and n-tuples. A list of atom/document pairs is generated such that an information metric can be computed for each pair. An information metric represents a ranking of the atom in relation to the particular document. Based on the information metric, some atom/document pairs are discarded and others are indexed.

Selection of atoms for search engine retrieval

A preliminary segment root and a final segment root are selected for each segment. Each time a search query is received, a set of nodes in each segment that will be used to resolve the search query is identified. A preliminary segment root is selected from the set of nodes. Based on statistical data from each node in the set of nodes indicating each node's capability to act as a final segment root that assembles query-execution data, the preliminary segment root algorithmically selects the final segment root. The other nodes in the set of nodes are notified regarding the identity of the final segment root.

Dynamic query master agent for query execution

Methods and computer storage media are provided for generating an algorithm used to provide preliminary rankings to candidate documents. A final ranking function that provides final rankings for documents is analyzed to identify potential preliminary ranking features, such as static ranking features that are query independent and dynamic atom-isolated components that are related to a single atom. Preliminary ranking features are selected from the potential preliminary ranking features based on many factors. Using these selected features, an algorithm is generated to provide a preliminary ranking to the candidate documents before the most relevant documents are passed to the final ranking stage.

Decomposable ranking for efficient precomputing

A priority hash index provides efficient lookup of posting lists for search query terms. The priority hash index is a data structure in which hash values for terms are distributed across multiple storage devices based on importance of the terms and access speeds of the storage devices. Terms are grouped into search lists with each search list including a storage location on each storage device. When a search query is received, a term is identified and hashed to a location on the first storage device and to generate a unique hash value for the term. The locations on the storage device for the term's search list are sequentially read until the hash value for the term is located to access the posting list for the term.

Priority hash index

Search results are identified and returned in response to search queries by evaluating and pruning candidate documents in multiple stages. The process employs a search index that indexes atoms found in documents and pre-computed scores for document/atom pairs. When a search query is received, atoms are identified from the search query and a reformulated query is generated based on the identified atoms. The reformulated query is used to identify matching documents, and a preliminary score is generated for matching documents using a simplified scoring function and pre-computed scores in the search index. Documents are pruned based on preliminary scores, and the remaining documents are evaluated using a final ranking algorithm that provides a final set of ranked documents, which is used to generate search results to return in response to the search query.

Matching funnel for large document index

Decomposable ranking for efficient precomputing that selects preliminary ranking features comprising static ranking features and dynamic atom-isolated components

Growing community of inventors

Mo I Rana, Norway

Knut Magne Risvik