# US 2019/0379762 A1 · Data prefetching for large data systems

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0379762 A1
- Document ID: US20190379762A1
- Status: Published application
- Kind: A1
- Application number: 16/550,389
- Filed: 2019-08-26
- Published: 2019-12-12
- Earliest claimed: 2015-06-25
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Giovanni Cherubini; Jens Jelitto; Vinodh Venkatesan
- Web page: https://patentgrep.com/patent/US20190379762A1

## CPC

- H04L 67/63: Routing a service request depending on the request content or context
- G06F 12/08: in hierarchically structured memory systems, e.g. virtual memory systems
- G06F 12/0862: with prefetch
- G06F 12/0868: Data transfer between cache memory and other subsystems, e.g. storage devices or host systems
- G06F 16/172: Caching, prefetching or hoarding of files
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- H04L 67/1097: for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
- H04L 67/5681: Pre-fetching or pre-delivering data based on network characteristics
- G06F 12/084: with a shared cache
- G06F 12/0846: Cache with multiple tag or data arrays being simultaneously accessible
- G06F 12/0871: Allocation or management of cache space
- G06F 2212/1044: Space efficiency improvement
- G06F 2212/154: Networked environment
- G06F 2212/463: File
- G06F 2212/6024: History based prefetching
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

The present invention is notably directed to a computer-implemented method of accessing data by an application. The method comprises detecting an application that is accessing an initial file stored in a storage system; fetching an application model of the detected application, the application model comprising an association between access patterns of the detected application when accessing first files with first metadata information of the said first files; identifying one or more second files stored in the storage system, the identified one or more second files having second metadata information that are similar to the first metadata information; and prefetching the identified one or more second files.

## Claims

### Claim 1

A computer-implemented method of accelerating access to data in a storage system by an application, comprising:
detecting the application accessing an initial file stored in the storage system;
fetching, from an application model database, an application model of the application in response to detecting the access of the initial file, the application model comprising an access pattern of the application when accessing first files having first metadata information similar to metadata information of the initial file;
identifying one or more second files stored in the storage system and having second metadata information that are similar to the first metadata information;
filtering the one or more second files by computing, for each second file, a probability of each second file to be accessed in a next time period, according to the access pattern of the application model, and comparing the probability to a threshold that incorporates a value for percentage of free cache; and
prefetching from cold storage to cache memory those of the one or more second files for which the probability to be accessed satisfies the threshold;
wherein prefetching the identified one or more second files further comprises associating a prediction confidence to each second file to be prefetched and prefetching second files associated with a prediction confidence that exceeds a predetermined confidence threshold.

### Claim 2 (depends on 1)

The method of claim 1, further comprising before detecting the application:
building, in the application model database, the application model of the application based on access patterns of the application when accessing first files with first metadata information of the first files.

### Claim 3 (depends on 2)

The method of claim 2, wherein a machine learning algorithm builds the application model of the application by computing correlations between access patterns of the application accessing first files and first metadata information of the first files that are accessed.

### Claim 4 (depends on 3)

The method of claim 3, wherein the machine learning algorithm is an agglomerative information bottleneck algorithm.

### Claim 5 (depends on 1)

The method of claim 1, wherein identifying the one or more second files further comprises determining, for each second file to be prefetched, a degree of similarity between its second metadata information and metadata information of the initial file, and prefetching the identified one or more second files further comprises prefetching the identified one or more second files that exceeds a predetermined degree of similarity.

### Claim 6 (depends on 1)

The method of claim 1, wherein metadata associated with a file include at least one among data owner, data owner's group, data path, data encoding, date of creation of the data, date of last modification of the data, date of last access of the data.

### Claim 7 (depends on 1)

The method of claim 1, wherein metadata further comprises domain-dependent metadata.

### Claim 8

A computer-implemented system for accelerating access to data in a storage system by an application, the system comprising:
a detector module that detects the application accessing an initial file stored on the storage system;
an application model database that stores an application model of the application, the application model comprising an access pattern of the application when accessing first files that have first metadata information similar to metadata information of the initial file;
an access pattern predictor that identifies one or more second files stored in the storage system, the one or more second files having second metadata information that are similar to the first metadata information, and that filters the one or more second files by computing, for each second file, a probability of each second file to be accessed in a next time period, according to the access pattern of the application model, and comparing the probability to a threshold that incorporates a value for percentage of free cache;
a prefetch module that prefetches from cold storage to cache memory those of the one or more second files for which the probability to be accessed exceeds the threshold; and
a memory that stores the one or more second files that are prefetched;
wherein the access pattern predictor associates a prediction confidence to each second file to be prefetched and the prefetch module prefetches each second file that is associated with a prediction confidence that exceeds a predetermined confidence threshold.

### Claim 9 (depends on 8)

The computer-implemented system of claim 8, wherein the access pattern predictor determines, for each second file to be prefetched, a degree of similarity between its second metadata information and metadata information of the initial file, and the prefetch module prefetches second files that exceed a predetermined degree of similarity.

### Claim 10 (depends on 8)

The computer-implemented system of claim 8, further comprising:
a machine learning algorithm that computes correlations between access patterns of the application accessing first files and first metadata of the first files that are accessed and builds an application model from the computed correlations.

### Claim 11 (depends on 10)

The computer-implemented system of claim 10, wherein the machine learning computes the correlation with an agglomerative information bottleneck algorithm.

### Claim 12 (depends on 10)

The computer-implemented system of claim 10, wherein the machine learning continuously updates the application model by comparing the identified access patterns with the actual access patterns.

### Claim 13 (depends on 8)

The computer-implemented system of claim 8, wherein metadata information include at least one among data owner, data owner's group, data path, data encoding, date of creation of the data, date of last modification of the data, date of last access of the data, domain-dependent metadata.

### Claim 14

A computer program product for accelerating access to data in a storage system by an application, the computer program product comprising a computer readable non-transitory storage medium embodying program instructions, the program instructions being executable by a computerized system to cause:
detecting the application accessing an initial file stored on the storage system;
fetching, from an application model database, an application model of the application in response to detecting the access of the initial file, the application model comprising an access pattern of the application when accessing first files having first metadata information similar to metadata information of the initial file;
identifying one or more second files stored in the storage system, the identified one or more second files having second metadata information that are similar to the first metadata information;
filtering the one or more second files by computing, for each second file, a probability of each second file to be accessed in a next time period, according to the access pattern of the application model, and comparing the probability to a threshold that incorporates a value for percentage of free cache; and
prefetching from cold storage to cache memory those of the one or more second files for which the probability to be accessed exceeds the threshold;
wherein prefetching the identified one or more second files further comprises associating a prediction confidence to each second file to be prefetched and prefetching second files associated with a prediction confidence that exceeds a predetermined confidence threshold.
