Machine-learning-based load balancing for cloud-based disaster recovery apparatuses, processes and systems

Patent metadata

Field	Value
Publication number	US-12223362-B2
Application number	US-202117399008-A
Country	US
Kind code	B2
Filing date	Aug 10, 2021
Priority date	Aug 10, 2021
Publication date	Feb 11, 2025
Grant date	Feb 11, 2025

Abstract

The Machine-Learning-Based Load Balancing for Cloud-Based Disaster Recovery Apparatuses, Processes and Systems (“MLLB”) transforms workload agent installation request, AWCD training request, NWCD training request, asset workload classification request, node workload classification request, asset virtualization request inputs via MLLB components into workload agent installation response, AWCD training response, NWCD training response, asset workload classification response, node workload classification response, asset virtualization response outputs. An asset virtualization request datastructure is obtained. A set of asset workload classification labels for the asset determined using an asset workload classification datastructure is retrieved. A set of node workload classification labels for each node in a set of available compute nodes determined using a node workload classification datastructure is retrieved. A set of compatible candidate compute nodes is determined using a set of capacity threshold rules. A virtual machine corresponding to the asset is instantiated on a selected candidate compute node.

First claim

What is claimed is:

1. A load balancing asset virtualizing apparatus, comprising:
at least one memory;
a component collection stored in the at least one memory;
at least one processor disposed in communication with the at least one memory, the at least one processor executing processor-executable instructions from the component collection, the component collection structured with processor-executable instructions, comprising:
obtain, via the at least one processor, an asset virtualization request datastructure, the asset virtualization request datastructure structured to include a first data field for identifying an asset and a second data field for specifying an expected workload timeframe corresponding to the expected workload timeframe associated with the asset virtualization request datastructure;
retrieve, via the at least one processor, a set of asset workload classification labels for the asset, the set of asset workload classification labels determined using an asset workload classification datastructure, the asset workload classification datastructure trained using a machine learning method and asset telemetry training data;
retrieve, via the at least one processor, a set of node workload classification labels for each node in a set of available compute nodes, each set of node workload classification labels determined using a node workload classification datastructure, the node workload classification datastructure trained using a machine learning method and node telemetry training data from a repository, the node telemetry training data including labeled training records over a certain period of time;
determine, via the at least one processor, a set of candidate compute nodes from the set of available compute nodes, in which the set of node workload classification labels for a candidate compute node is determined to be compatible with the set of asset workload classification labels for the asset, in which compatibility is determined using a set of capacity threshold rules;
select, via the at least one processor, a candidate compute node from the set of candidate compute nodes; and
instantiate, via the at least one processor, a virtual machine corresponding to the asset on the selected candidate compute node.

2. The apparatus of claim 1 , in which the asset is one of: a desktop, a workstation, a laptop, a mobile device, a server.

3. The apparatus of claim 1 , in which the asset is structured to execute backup software that is structured to utilize a kernel-resident agent to periodically collect workload info data regarding the asset.

4. The apparatus of claim 1 , in which the asset virtualization request datastructure is structured to include a data field for identifying a snapshot, in which the set of asset workload classification labels for the asset is associated with the snapshot.

5. The apparatus of claim 1 , in which the set of asset workload classification labels for the asset is associated with a time window matching the expected workload timeframe.

6. The apparatus of claim 1 , in which the asset virtualization request datastructure is structured to include a data field for specifying a virtualization definition, the virtualization definition specifying a set of resources requested for the virtual machine.

7. The apparatus of claim 6 , in which the component collection is further structured with processor-executable instructions, comprising:
determine, via the at least one processor, that the virtualization definition is under-provisioned or over-provisioned for a resource; and
modify, via the at least one processor, the virtualization definition to use a resource amount for the resource that corresponds to observed workload for the asset.

8. The apparatus of claim 6 , in which a resource is one of: CPU, RAM, disk, network, energy, time of day.

9. The apparatus of claim 1 , in which the set of asset workload classification labels for the asset comprises at least one of: a label indicating an overall workload footprint of the asset, a plurality of labels indicating workload footprints of the asset for different resources, one or more labels indicating resources heavily utilized by the asset.

10. The apparatus of claim 1 , in which the machine learning method is one of: logistic regression, k-nearest neighbors, random forest, a neural-network-based learning method.

11. The apparatus of claim 1 , in which the asset workload classification datastructure and the node workload classification datastructure are the same datastructure.

12. The apparatus of claim 1 , in which the candidate compute node is selected randomly.

13. The apparatus of claim 1 , in which the instructions to select a candidate compute node are structured as:
determine, via the at least one processor, virtualized assets corresponding to guest virtual machines already running on a respective candidate compute node from the set of candidate compute nodes;
determine, via the at least one processor, a set of asset workload classification labels for each of the virtualized assets;
determine, via the at least one processor, remaining capacity of the respective candidate compute node based on: a capacity metric associated with the respective candidate compute node, the set of node workload classification labels for the respective candidate compute node, and the set of asset workload classification labels for each of the virtualized assets;
determine, via the at least one processor, that the remaining capacity of the respective candidate compute node meets capacity requirements of the asset; and
select, via the at least one processor, the respective candidate compute node.

14. The apparatus of claim 13 , in which the capacity requirements of the asset are determined based on the set of asset workload classification labels for the asset.

15. The apparatus of claim 13 , in which the capacity requirements of the asset are determined based on a virtualization definition for the asset.

16. A load balancing asset virtualizing processor-readable, non-transient medium, the medium storing a component collection, the component collection structured with processor-executable instructions comprising:
obtain, via at least one processor, an asset virtualization request datastructure, the asset virtualization request datastructure structured to include a first data field for identifying an asset and a second data field for specifying an expected workload timeframe corresponding to the expected workload timeframe associated with the asset virtualization request datastructure;
retrieve, via the at least one processor, a set of asset workload classification labels for the asset, the set of asset workload classification labels determined using an asset workload classification datastructure, the asset workload classification datastructure trained using a machine learning method and asset telemetry training data;
retrieve, via the at least one processor, a set of node workload classification labels for each node in a set of available compute nodes, set of node workload classification labels determined using a node workload classification datastructure, the node workload classification datastructure trained using a machine learning method and node telemetry training data from a repository, the node telemetry training data including labeled training records over a certain period of time;
determine, via the at least one processor, a set of candidate compute nodes from the set of available compute nodes, in which the set of node workload classification labels for a candidate compute node is determined to be compatible with the set of a

Assignees

• Datto Inc

Inventors

• Sherwood Jr William J

Classifications

• G06F9/5077

  Logical partitioning of resources; Management or configuration of virtualized resources (specific details on emulation or internal functioning of virtual machines G06F9/455) · CPC title

• G06F9/5016

  the resource being the memory · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F2209/508

  Monitor · CPC title

• G06F2209/503

  Resource availability · CPC title