Sandisk Technologies Inc.

Dell Products Inc.

Datarobot, Inc.

Cisco Technology, Inc.

Fusion-io, Inc.

Intelligent Intellectual Property Holdings 2 LLC

Swaminathan Sundararaman

Nisha D Talagala

Sriram Subramanian

David Flynn

Ashish Batwara

Vinay Sawal

James G Peterson

Bharath Ramsundar

Robert A Wipfel

Jingpei Yang

Nick Piggin

Ned Plasson

Gal Zuckerman

John Strasser

Krishnamurthy Subramanian

Saikrishna M Kotha

Lior Khermosh

David Nellans

Vivek Dharmadhikari

Haresh Shah

Dhananjoy Das

Srinath Sridharan

Hiren Desai

Amar Mudrankit

Chang Ahn

Sindhu Ghanta

Drew Roselli

Vinay Sridhar

Gregory N Gillis

Ronak K Desai

Robert Barry Wood

Anoop Ghanwani

Vladislav Bolkhovitin

Rohit Sharma

Mukesh Moopath Velayudhan

Lior Amar

Phaniraj Vattem

Eivind Sarto

Mohan Ayalasomayajula

Shaohua Li

Bhavini Gada

Apparatuses, systems, program products, and method are disclosed for detecting suitability of machine learning models for datasets. An apparatus includes a training evaluation module configured to calculate a first statistical data signature for a training data set of a machine learning system using one or more predefined statistical algorithms. An apparatus includes an inference evaluation module configured to calculate a second statistical data signature for an inference data set of a machine learning system using one or more predefined statistical algorithms. An apparatus includes a score module configured to calculate a suitability score describing the suitability of a training data set to an inference data set as a function of a first and a second statistical data signature. An apparatus includes an action module configured to perform an action related to a machine learning system in response to a suitability score satisfying an unsuitability threshold.

Detecting suitability of machine learning models for datasets

Apparatuses, systems, methods, and computer program products are disclosed for persistent memory management. Persistent memory management may include replicating a persistent data structure in volatile memory buffers of at least two non-volatile storage devices. Persistent memory management may include preserving a snapshot copy of data in association with completion of a barrier operation for the data. Persistent memory management may include determining which interface of a plurality of supported interfaces is to be used to flush data from a processor complex.

Persistent memory management

Apparatuses, systems, and methods are disclosed for snapshots of a non-volatile device. A method includes writing data in a sequential log structure for a non-volatile device. A method includes marking a point, in a sequential log structure, for a snapshot of data. A method includes preserving a logical-to-physical mapping for a snapshot based on a marked point and a temporal order for data in a sequential log structure.

Time sequence data management

Apparatuses, systems, program products, and method are disclosed for machine learning abstraction. An apparatus includes an objective module configured to receive an objective to be analyzed using machine learning. An apparatus includes a grouping module configured to select a logical grouping of one or more machine learning pipelines to analyze a received objective. An apparatus includes an adjustment module configured to dynamically adjust one or more machine learning settings for a logical grouping of one or more machine learning pipelines based on feedback generated in response to analyzing a received objective.

Machine learning abstraction

Described herein are systems and methods for providing data sets from a constantly changing database to a streaming machine learning component. In one embodiment, a data streaming sub-system receives multiple incoming streams of data sets, in which each stream is generated in real-time by one of multiple data sources. The streaming sub-system sends data sets, on-the-fly as they are received, to storage in the memory of a database, in which there is a linkage between the storage and the time of arrival or the time of storage, of the data sets. The database receives, from a machine learning component, a request to receive data sets according to a particular time or time period. In response to such request, the database identifies such data sets according to the particular time or time period and sends them to the machine learning component.

Systems and methods for storing and retrieving data sets based on temporal information

Apparatuses, systems, methods, and computer program products are disclosed for a persistent data structure. A method includes associating a logical identifier with a data structure. A method includes writing data of a data structure to a first region of a volatile memory module. A volatile memory module may be configured to ensure that data is preserved in response to a trigger. A method includes copying data of a data structure from a volatile memory module to a non-volatile storage medium such that the data of the data structure remains associated with a logical identifier.

Persistent data structures

Apparatuses, systems, methods, and computer program products are disclosed for persistent memory management. Persistent memory management may include providing a persistent data structure stored at least partially in volatile memory configured to ensure persistence of the data structure in a non-volatile memory medium. Persistent memory management may include replicating a persistent data structure in volatile memory buffers of at least two non-volatile storage devices. Persistent memory management may include preserving a snapshot copy of data in association with completion of a barrier operation for the data. Persistent memory management may include determining which interface of a plurality of supported interfaces is to be used to flush data from a processor complex.

Described herein are systems and methods for executing efficiently, in real-time, a plurality of machine learning processes. In one embodiment, a computing platform with multiple compute elements receives multiple data streams, each such stream associated with its own respective machine learning process. Each machine learning process is operative to use its data stream as input to train, in real-time, a respective mathematical model. Each of the processes has peaks and dips in processing demands. The system re-allocates, in real-time, compute elements from the processes with lower processing demands to processes with higher processing demands, thereby handling all of the multiple processes on-the-fly, preventing peak demands from causing the system to stall, and reducing overall the computational resources required by the system.

Systems and methods to execute efficiently a plurality of machine learning processes

An apparatus, system, and method are disclosed for data management. The method includes writing data in a sequential log structure. The method also includes receiving a time sequence request from a client. The method further includes servicing the time sequence request based on a temporal order of the data in the sequential log structure.

Embodiments are directed to a process and system for optimizing traffic paths for orphaned hosts in a VXLAN system, by configuring virtual link trunking (VLT) peers to advertise MAC addresses learned from all multi-homed hosts in the system using Anycast VXLAN tunnel endpoint-Internet Protocol address (VTEP-IP); configuring the virtual link trunking (VLT) peers to advertise MAC addresses learned from all single-homed hosts in the system using a secondary VTEP-IP; directing unicast traffic destined to the single-homed hosts to directly connected VLT peers using the Secondary VTEP-IP; and directing Broadcast, unknown unicast, and multi-cast (BUM) traffic destined to the single-homed hosts to directly connected VLT peers using the Inclusive Multicast Ethernet Tag route.

Optimizing traffic paths to orphaned hosts in VXLAN networks using virtual link trunking-based multi-homing

A storage system implements a sparse, thinly provisioned logical-to-physical translation layer. The storage system may perform operations to modify logical-to-physical mappings, including creating, removing, and/or modifying any-to-any and/or many-to-one mappings between logical identifiers and stored data (logical manipulation operations). The storage system records persistent metadata to render the logical manipulation (LM) operations persistent and crash-safe. The storage system may provide access to LM functionality through a generalized LM interface. Clients may leverage the LM interface to efficiently implement higher-level functionality and/or offload LM operations to the storage system.

Generalized storage virtualization interface

Snapshots for a non-volatile device

Apparatuses, systems, methods, and computer program products are disclosed for hybrid checkpointed memory. An extended memory module uses volatile memory of a host and a non-volatile memory medium as virtual memory for the host. A clone module clones data of a range of virtual memory in response to a checkpoint event for the range of virtual memory. A range of virtual memory may include data stored in a volatile memory and data stored in a non-volatile memory medium. A checkpoint module flushes dirty data of a range of virtual memory to a non-volatile memory medium in response to a checkpoint event. A hybrid checkpointed memory interface provides access to data of a range of virtual memory while dirty data is being flushed using data of a range of virtual memory, or using a clone of the data.

Hybrid checkpointed memory

A cache module leverages storage metadata to cache data of a backing store on a non-volatile storage device. The cache module maintains access metadata pertaining to access characteristics of logical identifiers in the logical address space, including access characteristics of un-cached logical identifiers (e.g., logical identifiers associated with data that is not stored on the non-volatile storage device). The access metadata may be separate and/or distinct from the storage metadata. The cache module determines whether to admit data into the cache and/or evict data from the cache using the access metadata. A storage module may provide eviction candidates to the cache module. The cache module may select candidates for eviction. The storage module may leverage the eviction candidates to improve the performance of storage recovery and/or grooming operations.

Systems and methods for cache and storage device coordination

A storage layer of a non-volatile storage device may be configured to provide key-value storage services. Key conflicts may be resolved by modifying the logical interface of data stored on the non-volatile storage device. Resolving a key conflict may comprise identifying an alternative key and implementing one or more range move operations configured to bind the stored data to the alternative key. The move operations may be implemented without relocating the data on the non-volatile storage device.

Systems and methods for storage collision management

Growing community of inventors

San Jose, CA, United States of America

Swaminathan Sundararaman