US 11,599,577 B2 · System and method for content-hashed object storage

Abstract

Features are detected from a sensor signal via a deep-learning network or other feature engineering methods in an edge processing node. Machine-learned metadata is created that describes the features, and a hash is created with the machine-learned metadata. The sensor signal is stored as a content object at the edge processing node, the object being keyed with the hash at, the edge processing node.

Drawings

Drawings are temporarily unavailable. Open the PDF at USPTO

Drawings

Claims20

Dependent claims

A method, comprising:

detecting a first feature from a sensor signal via a deep-learning network in an edge processing node;
creating first metadata indicating that the first feature has been detected in the sensor signal via a machine-learning processor that is pre-trained to detect the first feature, the first metadata describing the first feature using a label that is used consistently throughout a system in which the edge processing node operates;
creating a hash key with the first metadata, wherein the hash key comprises segments that represent different types of knowledge representations, the label representing one of the types of knowledge representation;
storing the sensor signal as a content object on an object-based storage at the edge processing node, the content object keyed in the object-based storage with the hash key, wherein the object-based storage manages the content object with a host based on an object identifier instead of block addresses;
extending the machine learning processor to learn a new feature in the sensor signal;
post-processing the content object to detect the new feature in the content object and creating new metadata indicating detection of the new feature;
and
creating an updated hash key with the new metadata, the updated hash key retaining a hash of the first metadata such that previous searches for the first feature on the updated hash key will obtain the same results as searching the hash key.

5 more elements

An edge processing node, comprising:

a processor;
a data storage medium configured for object-based storage, wherein the object-based storage manages objects with a host based on object identifiers instead of block addresses;
input/output circuitry coupled to the processor and the data storage medium and configured to receive a sensor signal;
and
memory that stores instructions that are operable by the processor to:
detect a first feature from the sensor signal via a deep-learning network or other feature engineering method;
create first metadata indicating that the first feature has been detected, via a machine-learning processor that is pre-trained to detect the first feature, the first metadata describing the first feature using a label that is used consistently throughout a system in which the edge processing node operates;
create a hash key with the first metadata, wherein the hash key comprises segments that represent different types of knowledge representations, the label representing one of the types of knowledge representation;
store the sensor signal as a content object in the data storage medium, the content object keyed in the object-based storage with the hash key;
extend the machine learning processor to learn a new feature in the sensor signal and create new metadata indicating detection of the new feature;
post-process the content object to detect the new feature in the content object;
and
create an updated hash key with the new metadata, the updated hash key retaining a hash of the first metadata such that previous searches for the first feature on the updated hash key will obtain the same results as searching the hash key.

10 more elements

A system comprising:

a plurality of sensors that generate a respective plurality of data streams;
a task management module that directs the plurality of data streams to a respective plurality of active object store nodes, each of the active object store nodes comprising an object-based storage wherein the object-based storage manages objects with a host based on object identifiers instead of block addresses, the active object storage nodes operable to:
detect a first feature from a data stream of the plurality of data streams via a machine-learning processor that utilizes a deep-learning network or other feature engineering methods pre-trained to detect the first feature;
create first metadata indicating that the first feature has been detected, the first metadata describing the first feature using a label that is used consistently throughout the system;
create a hash key with the first metadata, wherein the hash key comprises segments that represent different types of knowledge representations, the label representing one of the types of knowledge representation;
store the data stream as a content object in one of the active object store nodes, the content object keyed in the object-based storage with the hash key;
extend the machine learning processor to learn a new feature in the data stream;
post-process the content object to detect the new feature in the content object and create new metadata indicating detection of the new feature;
and
create an updated hash key with the new metadata, the updated hash key retaining a hash of the first metadata such that previous searches for the first feature on the updated hash key will obtain the same results as searching the hash key;
and
a key query handling module operable to:
receive and store the hash keys from the active object store nodes;
receive a query via the task management module;
process the query and identify matches in the stored hash keys;
and
return a list of the hash keys that satisfy the query, the list of the hash keys facilitating access to a subset of the content objects stored at a subset of the active object store nodes.

14 more elements

Dependencies are inferred from the claim text.

Description4,953 words

Summary104 words

#The present disclosure is directed to a system and method that facilitates content-hashed object storage. In one embodiment, features are detected from a sensor signal via a deep-learning network or other feature engineering methods in an edge processing node. Machine-learned metadata is created that describes the features, and a hash is created with the machine-learned metadata. The sensor signal is stored as a content object at the edge processing node, the object being keyed with the hash at the edge processing node.

#These and other features and aspects of various embodiments may be understood in view of the following detailed discussion and accompanying drawings.

Brief description of the drawings7 figures

#The discussion below makes reference to the following figures, wherein the same reference number may be used to identify the similar/same component in multiple figures.

#FIGS. 1-3 are diagrams of edge processing networks according to example embodiments;

#FIG. 4 is a block diagram of an edge processing architecture according to example embodiments;

Detailed description4,721 words

#The present disclosure generally relates to object-based data storage. In general-purpose computing devices (e.g., personal computers, mobile devices), storage devices such as hard disk drives (HDDs) and solid-state drives (SSDs) are usually configured as block storage devices. Generally, this means the storage device presents the storage media (e.g., disks, flash memory) as a series of data blocks identified by addresses, e.g., logical block addresses. In disk drives, these blocks would generally be directly mapped to the disk sectors on which the data was stored, although other arrangements (e.g., SSDs) may utilize a more flexible mapping of the logical to physical addresses.

#Regardless, a block storage device receives access commands from a host machine, such as reads and writes that are targeted to logical block addresses. The storage device services these requests without regard to the data that is being stored or the context in which the data is being used. The data stored in the blocks and associated structures are determined at a higher level of abstraction, e.g., via a disk controller card, host operating system, etc. For example, an operating system may implement a filesystem that governs how data hierarchies (e.g., directory trees) and data objects (e.g., files, streams) are defined and mapped to the underlying block addresses. The specific structures of the filesystem are stored at predefined block addresses as filesystem metadata.

#While operating system filesystems that utilize block access storage are still commonly-used, a different storage architecture known as object storage or object-based storage has been developed. In object storage, data are managed as objects, which include not only the underlying data of interest (e.g., pixels that define an image), but also includes metadata and a globally unique identifier. The underlying details of the storage (e.g., the physical addresses on which the data is stored) is abstracted away from the higher levels, e.g., applications. Object storage can be implemented at many different levels, e.g., the operating system, within the data storage device, within an interface therebetween. The present disclosure is directed to storage devices that implement object-based storage within the data storage device itself, such that the storage device and host communicate using object identifiers instead of (or in addition to) block addresses.

Facts

Application

16/597,911Filed 10 Oct 2019

Published

US 2021/0109971 A115 Apr 2021

Granted

7 Mar 2023

Term ends

18 Jun 2040

Estimated · how

20 years from earliest known nonprovisional domestic filing plus recorded patent term adjustment, including 252 days of adjustment. Maintenance-fee lapses, terminal disclaimers and statutory extensions have not been checked. The continuity chain may be incomplete; this is not an enforceability determination.

Classification9 codes

G06F 16/137
File access structures, e.g. distributed indices: hash-based
G06F 16/9014
Indexing; Data structures therefor; Storage structures: hash tables
G06F 16/152
File search processing: using file content signatures, e.g. hash values
G06F 16/2255
Indexing structures: hash tables

Timeline

10 Oct 2019Application filedUS 16/597,911
10 Oct 2019Assigned to Seagate Technology LLCassignment of assignor's interest
15 Apr 2021PublishedUS 2021/0109971 A1
7 Mar 2023GrantedUS 11,599,577 B2
12 Oct 2026TodayWithin its term
18 Jun 2040Term endsEstimated, includes 252 days of adjustment

System and method for content-hashed object storage

Life of this patent

System and method for content-hashed object storage

Life of this patent

Life of this patent

Abstract

Drawings

Drawings

Claims20

Description4,953 words

Drawings

Classification9 codes

Life of this patent

Description4,953 words