Granted Patent B2

US 12,287,721 · App. 17/585,695 · Granted Apr 29, 2025

Inventors: Yan Li (Beijing, CN); Run Qian Bj Chen (Beijing, CN); Chen Guang Zhao (Beijing, CN); Qin Qin Zhou (Beijing, CN); Guang Han Sui (Beijing, CN); Jing Li (Beijing, CN); You Bing Li (Beijing, CN); Yu Xiang Chen (Shanghai, CN)

Assignee: International Business Machines Corporation

G06F11/3442G06F9/505G06F9/5077

Abstract

Claims (83)

1. A computer-implemented method for managing and optimizing container image storage, the computer-implemented method comprising:

inputting, by a processor, image and data information of a historical workload record into a machine learning model configured to predict future workload requirements based on workload trends of the historical workload record;

learning, by the processor, image and data requirement trends of the historical workload record using the machine learning model;

outputting, by the processor, predicted image and data requirements for future workloads based on the image and data requirement trends learned from the machine learning model;

engaging, by the processor, a checking cycle wherein a daemon process checks whether an image file for an upcoming future workload as predicted by the predicted image and data requirements needs to be downloaded to one or more nodes prior to running the upcoming future workload; and

triggering, by the processor, a pulling task upon a current local time plus a max recorded download time of the image file plus a pre-defined buffer time is less than or equal to a predicted start time of the upcoming future workload.

2. The computer-implemented method of claim 1 , wherein the pulling task comprises:

selecting, by the processor, a plurality of nodes with properties satisfying image requirements and workload requirements of the upcoming future workload;

ordering, by the processor, the plurality of nodes with the properties satisfying the image requirements and workload requirements by highest availability of one or more computing resources; and

iteratively checking, by the processor, the plurality of nodes sequentially, starting with a first node in the ordering of the plurality of nodes, for whether a selected node has the image file for the upcoming future workload stored by a local storage device of the selected node, wherein upon confirming the image file for the upcoming future workload is locally stored by the local storage device of the selected node, tasking the selected node to run the upcoming future workload on the selected node at the predicted start time of the upcoming future workload.

3. The computer-implemented method of claim 2 , further comprising:

upon confirming the image file for the upcoming future workload is not locally stored by the local storage device of the second node, comparing, by the processor, available space of the local storage device with image size of the image file for deploying the upcoming future workload;

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size exceeds the available space of the local storage device:

triggering, by the processor, a disk utilization process; and

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size is less that the available space on the local storage device:

pulling, by the processor, the image file from a registry; and

storing, by the processor, the image file to the local storage device.

4. The computer-implemented method of claim 3 , wherein finding that the image size is less that the available space on the local storage device further comprises:

determining, by the processor, network input/output (I/O) of the selected node satisfies network I/O requirements of the upcoming future workload.

5. The computer-implemented method of claim 3 , wherein the disk utilization process comprises:

setting, by the processor, a target amount of free space on the local storage device that meets or exceeds the image size;

mapping, by the processor, a relationship between existing image files stored on the local storage device and related workloads predicted to be scheduled to be run at a future point in time;

ordering, by the processor, the existing image files by predicted start times of related workloads, wherein earliest start time is ordered first, and latest start time is last;

while the available space on the local storage device is less than or equal to the target amount of free space, deleting, by the processor, a last image in the ordering of the existing image files until the available space on the local storage device exceeds the target amount of free space on the local storage device; and

pulling, by the processor, the image file from the registry to the local storage device.

6. The computer-implemented method of claim 1 , wherein the image and data requirements are selected from the group consisting of image required, node kind required, number of nodes required, a target deadline, network I/O requirement and a combination thereof.

7. The computer-implemented method of claim 1 , wherein the image and data information of the historical workload record are selected from the group consisting of image name, image size, image version, historical workload start time, download time, workload name, workload type, workload running time, kinds of nodes, number of nodes, and a combination thereof.

8. A computing program product for managing and optimizing container image storage comprising:

one or more computer-readable storage media having computer-readable program instructions stored on the one or more computer-readable storage media, said program instructions executes a computer-implemented method comprising:

inputting, by a processor, image and data information of a historical workload record into a machine learning model configured to predict future workload requirements based on workload trends of the historical workload record;

learning, by the processor, image and data requirement trends of the historical workload record using the machine learning model;

outputting, by the processor, predicted image and data requirements for future workloads based on the image and data requirement trends learned from the machine learning model;

engaging, by the processor, a checking cycle wherein a daemon process checks whether an image file for an upcoming future workload as predicted by the predicted image and data requirements needs to be downloaded to one or more nodes prior to running the upcoming future workload; and

triggering, by the processor, a pulling task upon a current local time plus a max recorded download time of the image file plus a pre-defined buffer time is less than or equal to a predicted start time of the upcoming future workload.

9. The computing program product of claim 8 , wherein the pulling task comprises:

selecting, by a processor, a plurality of nodes with properties satisfying image requirements and workload requirements of the upcoming future workload;

ordering, by the processor, the plurality of nodes with the properties satisfying image requirements and workload requirements by highest availability of one or more computing resources; and

iteratively checking, by the processor, the plurality of nodes sequentially, starting with a first node in the ordering of the plurality of nodes, for whether a selected node has the image file for the upcoming future workload stored by a local storage device of the selected node, wherein upon confirming the image file for the upcoming future workload is locally stored by the local storage device of the selected node, tasking the selected node to run the upcoming future workload on the selected node at the predicted start time of the upcoming future workload.

10. The computing program product of claim 9 , further comprising:

upon confirming the image file for the upcoming future workload is not locally stored by the local storage device of the second node, comparing, by the processor, available space of the local storage device with image size of the image file for the upcoming future workload;

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size exceeds the available space of the local storage device:

triggering, by the processor, a disk utilization process; and

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size is less that the available space on the local storage device:

pulling, by the processor, the image file from a registry; and

storing, by the processor, the image file to the local storage device.

11. The computing program product of claim 10 , wherein finding that the image size is less that the available space on the local storage device further comprises:

determining, by the processor, network input/output (I/O) of the selected node satisfies network I/O requirements of the upcoming future workload.

12. The computing program product of claim 10 , wherein the disk utilization process comprises:

setting, by the processor, a target amount of free space on the local storage device that meets or exceeds the image size;

mapping, by the processor, a relationship between existing image files stored on the local storage device and related workloads predicted to be scheduled to be run at a future point in time;

ordering, by the processor, the existing image files by predicted start times of related workloads, wherein earliest start time is ordered first, and latest start time is last;

while the available space on the local storage device is less than or equal to the target amount of free space, deleting, by the processor, a last image in the ordering of the existing image files until the available space on the local storage device exceeds the target amount of free space on the local storage device; and

pulling, by the processor, the image file from the registry to the local storage device.

13. The computing program product of claim 8 , wherein the image and data requirements are selected from the group consisting of image required, node kind required, number of nodes required, a target deadline, network I/O requirement and a combination thereof.

14. The computing program product of claim 8 , wherein the image and data information of the historical workload record are selected from the group consisting of image name, image size, image version, historical workload start time, download time, workload name, workload type, workload running time, kinds of nodes, number of nodes, and a combination thereof.

15. A computer system for managing and optimizing container image storage comprising:

a processor; and

a computer-readable storage media coupled to the processor, wherein the computer-readable storage media contains program instructions executing, via the processor, a computer-implemented method comprising:

inputting, by the processor, image and data information of a historical workload record into a machine learning model configured to predict future workload requirements based on workload trends of the historical workload record;

learning, by the processor, image and data requirement trends of the historical workload record using the machine learning model;

outputting, by the processor, predicted image and data requirements for future workloads based on the image and data requirement trends learned from the machine learning model;

engaging, by the processor, a checking cycle wherein a daemon process checks whether an image file for an upcoming future workload as predicted by the predicted image and data requirements needs to be downloaded to one or more nodes prior to running the upcoming future workload; and

triggering, by the processor, a pulling task upon a current local time plus a max recorded download time of the image file plus a pre-defined buffer time is less than or equal to a predicted start time of the upcoming future workload.

16. The computer system of claim 15 , wherein the pulling task comprises:

selecting, by a processor, a plurality of nodes with properties satisfying image requirements and workload requirements of the upcoming future workload;

ordering, by the processor, the plurality of nodes with the properties satisfying image requirements and workload requirements by highest availability of one or more computing resources; and

iteratively checking, by the processor, the plurality of nodes sequentially, starting with a first node in the ordering of the plurality of nodes, for whether a selected node has the image file for the upcoming future workload stored by a local storage device of the selected node, wherein upon confirming the image file for the upcoming future workload is locally stored by the local storage device of the selected node, tasking the selected node to run the upcoming future workload on the selected node at the predicted start time of the upcoming future workload.

17. The computer system of claim 16 , further comprising:

upon confirming the image file for the upcoming future workload is not locally stored by the local storage device of the second node, comparing, by the processor, available space of the local storage device with image size of the image file for the upcoming future workload;

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size exceeds the available space of the local storage device:

triggering, by the processor, a disk utilization process; and

upon comparing the available space of the local storage device with the image size of the image file, finding that the image size is less that the available space on the local storage device:

pulling, by the processor, the image file from a registry; and

storing, by the processor, the image file to the local storage device.

18. The computer system of claim 17 , wherein finding that the image size is less that the available space on the local storage device further comprises:

determining, by the processor, network input/output (I/O) of the selected node satisfies network I/O requirements of the upcoming future workload.

19. The computer system of claim 17 , wherein the disk utilization process comprises:

setting, by the processor, a target amount of free space on the local storage device that meets or exceeds the image size;

mapping, by the processor, a relationship between existing image files stored on the local storage device and related workloads predicted to be scheduled to be run at a future point in time;

ordering, by the processor, the existing image files by predicted start times of related workloads, wherein earliest start time is ordered first, and latest start time is last;

while the available space on the local storage device is less than or equal to the target amount of free space, deleting, by the processor, a last image in the ordering of the existing image files until the available space on the local storage device exceeds the target amount of free space on the local storage device; and

pulling, by the processor, the image file from the registry to the local storage device.

20. The computer system of claim 15 , wherein the image and data requirements are selected from the group consisting of image required, node kind required, number of nodes required, a target deadline, network I/O requirement and a combination thereof.

Assignments (1)

Continuity (1)

References Cited (29)