Abstract: 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.

Abstract: A “Key-Value Manager” applies an RDMA-based memory allocator to implement a distributed and lock-free B-Tree based key-value store. In various implementations, the Key-Value Manager is implemented as a standalone library running concurrently on large numbers of networked computing devices. The RDMA-based memory allocator of the Key-Value Manager allocates and accesses the memory of the key-value store via various sequences of RDMA messages (e.g., RDMA reads, RDMA writes, and RDMA atomic compare and swap (CAS) operations). As such, CPUs of computing devices hosting the memory of the key-value store do not need to be notified of or involved in RDMA-based reads and writes to that memory. Consequently, CPU utilization for any computing device that hosts any portion of the key-value store is minimal, thereby enabling the Key-Value Manager to scale to very large numbers of concurrent accesses by very large numbers of computing devices.

Abstract: A performance investigation tool (PIT) is described herein for investigating the performance of a distributed processing system (DPS). The PIT operates by first receiving input information that describes a graph processing task to be executed using a plurality of computing units. The PIT then determines, based on the input information, at least one time-based performance measure that describes the performance of a DPS that is capable of performing the graphical task. More specifically, the PIT can operate in a manual mode to explore the behavior of a specified DPS, or in an automatic mode to find an optimal DPS from within a search space of candidate DPSs. A configuration system may then be used to construct a selected DPS, using the plurality of computing units. In one case, the graph processing task involves training a deep neural network model having a plurality of layers.

Abstract: 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.

Abstract: An “RDMA-Based RPC Request System” combines the concepts of RPC and RDMA in a way that can be implemented on commodity networking communications hardware, e.g., RDMA-enabled network interface controllers (NICs) in an Ethernet-based network. In various implementations, the RDMA-Based RPC Request System enables fast lock-free and thread-safe execution of RPC requests between different computers in a network via RDMA-based messages. In other words, the RDMA-Based RPC Request System combines RDMA messaging and RPC requests to enable fast RPC requests via a sequence of RDMA messages transmitted over Ethernet using commodity NICs between networked computers in a data center or other network environment. This RDMA message process is both lock-free and thread-safe.

Abstract: Described herein are various technologies pertaining to shapewriting. A touch-sensitive input panel comprises a plurality of keys, where each key in the plurality of keys is representative of a respective plurality of characters. A user can generate a trace over the touch-sensitive input panel, wherein the trace passes over keys desirably selected by the user. A sequence of characters, such as a word, is decoded based upon the trace, and is output to a display or a speaker.

Abstract: 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.

Abstract: A “Key-Value Manager” applies an RDMA-based memory allocator to implement a distributed and lock-free B-Tree based key-value store. In various implementations, the Key-Value Manager is implemented as a standalone library running concurrently on large numbers of networked computing devices. The RDMA-based memory allocator of the Key-Value Manager allocates and accesses the memory of the key-value store via various sequences of RDMA messages (e.g., RDMA reads, RDMA writes, and RDMA atomic compare and swap (CAS) operations). As such, CPUs of computing devices hosting the memory of the key-value store do not need to be notified of or involved in RDMA-based reads and writes to that memory. Consequently, CPU utilization for any computing device that hosts any portion of the key-value store is minimal, thereby enabling the Key-Value Manager to scale to very large numbers of concurrent accesses by very large numbers of computing devices.

Abstract: 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.

Abstract: An “RDMA-Based RPC Request System” combines the concepts of RPC and RDMA in a way that can be implemented on commodity networking communications hardware, e.g., RDMA-enabled network interface controllers (NICs) in an Ethernet-based network. In various implementations, the RDMA-Based RPC Request System enables fast lock-free and thread-safe execution of RPC requests between different computers in a network via RDMA-based messages. In other words, the RDMA-Based RPC Request System combines RDMA messaging and RPC requests to enable fast RPC requests via a sequence of RDMA messages transmitted over Ethernet using commodity NICs between networked computers in a data center or other network environment. This RDMA message process is both lock-free and thread-safe.

Abstract: A performance investigation tool (PIT) is described herein for investigating the performance of a distributed processing system (DPS). The PIT operates by first receiving input information that describes a graph processing task to be executed using a plurality of computing units. The PIT then determines, based on the input information, at least one time-based performance measure that describes the performance of a DPS that is capable of performing the graphical task. More specifically, the PIT can operate in a manual mode to explore the behavior of a specified DPS, or in an automatic mode to find an optimal DPS from within a search space of candidate DPSs. A configuration system may then be used to construct a selected DPS, using the plurality of computing units. In one case, the graph processing task involves training a deep neural network model having a plurality of layers.

Abstract: Training large neural network models by providing training input to model training machines organized as multiple replicas that asynchronously update a shared model via a global parameter server is described herein. In at least one embodiment, a system including a model module storing a portion of a model and a deep learning training module that communicates with the model module are configured for asynchronously sending updates to shared parameters associated with the model. The techniques herein describe receiving and processing a batch of data items to calculate updates. Replicas of training machines communicate asynchronously with a global parameter server to provide updates to a shared model and return updated weight values. The model may be modified to reflect the updated weight values. The techniques described herein include computation and communication optimizations that improve system efficiency and scaling of large neural networks.

Abstract: The subject disclosure is directed towards a technology by which the accuracy of context-based information provided by at least one data source for received context data is increased. Correctness information received in association with usage of looked up context-based information is logged. The correctness information may be processed to increase the overall accuracy by correcting a data source, and/or by creating a blended data source that includes the most likely accurate portions (segments) from among multiple data sources as determined via the correctness information.

Abstract: Described herein are various technologies pertaining to shapewriting. A touch-sensitive input panel comprises a plurality of keys, where each key in the plurality of keys is representative of a respective plurality of characters. A user can generate a trace over the touch-sensitive input panel, wherein the trace passes over keys desirably selected by the user. A sequence of characters, such as a word, is decoded based upon the trace, and is output to a display or a speaker.

Abstract: The subject disclosure is directed towards a technology by which the accuracy of context-based information provided by at least one data source for received context data is increased. Correctness information received in association with usage of looked up context-based information is logged. The correctness information may be processed to increase the overall accuracy by correcting a data source, and/or by creating a blended data source that includes the most likely accurate portions (segments) from among multiple data sources as determined via the correctness information.

Abstract: Described is extracting structured information from web pages for use in directly answering queries with data items from the structured data. Users' post-search browsing behaviors (search trails) are treated as implicit labels as to the relevance between web content and user queries, and are used to determine wrappers for extracting structured information. In one implementation, a system identifies websites from web search logs, builds wrappers from users' search trails, filters out bad wrappers (from inconsistent user clicks), and combines structured information from different web sites, e.g., for each query.

Abstract: The sharing of language models is disclosed. In one embodiment, a language model service is provided that is shareable among handlers for input devices. The service includes a pre-processing mode of operation, and a correction mode of operation. In the former mode, the language model service is designed to receive a range within a document from a handler for an input device, and in response provide advice regarding text under consideration by the handler to insert within the document at the range, based on the context of the document within the range. In the latter mode, the language model service is designed to supervise correction over a range of text within a document, in which a number of different handlers for a number of different input devices were initially responsible for insertion of the text, such that the service solicits suggestions from the handlers, and based thereon determines text corrections. The corrections can then be made by the handlers, or by the service.

Abstract: An improved system and method for providing context information of executable code to an input method is provided. Advanced text input methods may be made aware of the type of text expected to be received as input so that input methods may achieve a higher accuracy in recognition of text input. Generic interfaces provide a framework for supporting application authoring platforms to allow application developers to easily specify context information to the system and have it reliably forwarded to the correct input methods. Additionally, a context tagging tool may associate specific text input fields of an existing application with an input scope without modifying the application itself. The context tagging tool may create a manifest that contains tags associating the specific text input fields with an input scope. Any advanced input methods use by the application may be updated with instructions for accessing the context information stored in the manifest.

Abstract: An improved system and method for providing context information of executable code to an input method is provided. Advanced text input methods may be made aware of the type of text expected to be received as input so that input methods may achieve a higher accuracy in recognition of text input. Generic interfaces provide a framework for supporting application authoring platforms to allow application developers to easily specify context information to the system and have it reliably forwarded to the correct input methods. Additionally, a context tagging tool may associate specific text input fields of an existing application with a n input scope without modifying the application itself. The context tagging tool may create a manifest that contains tags associating the specific text input fields with an input scope. Any advanced input methods use by the application may be updated with instructions for accessing the context information stored in the manifest.

Abstract: A common text framework for application programs having documents of primarily text and input devices for entering and editing text of such documents is disclosed. The framework provides for the interaction of applications and handlers for input devices, and includes a text store interface and a text input processor interface. The former interface permits an application having a document of primarily text to expose the document as an abstraction. The latter interface permits a handler for an input device to access the abstraction of the document and to insert additional text into the document.