Dynamic resource distribution using periodicity-aware predictive modeling

• US 10,484,301 B1
• Filed: 09/30/2016
• Issued: 11/19/2019
• Est. Priority Date: 09/30/2016
• Status: Active Grant

• Current Assignee: Nutanix Incorporated
• Original Assignee: Nutanix Incorporated
• Inventors: Shukla, Himanshu, Singh, Rahul, Nagpal, Abhinay, Ramesh, Aditya
• Primary Examiner(s): Pollack, Melvin H
• Application Number:
  US15/283,004
• Time in Patent Office: 1,145 Days
• Field of Search: 709226
• CPC Class Codes: G06F 2009/4557 Distribution of virtual machine instances; Migration and load balancing
• CPC Class Codes: G06F 3/0605 by facilitating the interaction with a user or administrator
• CPC Class Codes: G06F 3/0611 in relation to response time
• CPC Class Codes: G06F 3/0631 by allocating resources to storage systems
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
• CPC Class Codes: G06F 9/45533 Hypervisors; Virtual machine monitors
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: H04L 41/0896 Bandwidth or capacity management, i.e. automatically increasing or decreasing capacities flow or congestion control using dynamic resource allocation, e.g. in-call renegotiation, H04L47/76
• CPC Class Codes: H04L 41/0897 by horizontal or vertical scaling of resources, or by migrating entities, e.g. virtual resources or entities
• CPC Class Codes: H04L 41/147 for predicting network behaviour
• CPC Class Codes: H04L 41/16 using machine learning or artificial intelligence
• CPC Class Codes: H04L 41/22 comprising specially adapted graphical user interfaces [GUI]
• CPC Class Codes: H04L 43/0876 Network utilisation, e.g. volume of load or congestion level
• CPC Class Codes: H04L 43/0894 Packet rate
• CPC Class Codes: H04L 47/822 Collecting or measuring resource availability data
• CPC Class Codes: H04L 47/83 based on usage prediction
• CPC Class Codes: 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]

Abstract

Resource allocation techniques for distributed data storage. A set of distributed storage system historical resource usage measurements are collected and stored using distributed storage system measurement techniques. The resource usage metrics are associated with and/or derived from processing entities in the distributed storage computing system. An analysis module determines a training window time period corresponding to a portion of the collected distributed storage system historical resource usage measurements. The training window time period is determined so as to provide an earlier time boundary and a later time boundary that defines a periodically recurring portion of the distributed storage system historical resource usage measurements. A latest cycle of those periodically recurring measurements are then used to train a predictive model, which in turn is used to produce distributed storage system predicted resource usage characteristics. Resource allocation decisions are made based at least in part on predictions from the trained predictive model.

Claims

1. A method, comprising:

maintaining, at a storage pool, measurements corresponding to prior usage of a distributed storage resource of a cluster by a plurality of nodes in the cluster, the storage pool comprising an aggregation of one or more storage devices directly attached to respective nodes in the cluster;

receiving, at a first node of the plurality of nodes in the cluster, the measurements from the storage pool;

determining, at the first node, a time period for a training window corresponding to a portion of the measurements, wherein the training window bounds a periodically recurring portion of the measurements;

training, at the first node, a predictive model based at least in part on the periodically recurring portion of the measurements to produce a trained predictive model;

determining a resource allocation operation based at least in part on a prediction from the trained predictive model.

2. The method of claim 1, further comprising executing the resource allocation operation.

3. The method of claim 1, further comprising comparing a peak resource usage measurement and a mean resource usage measurement to a storage pool predicted resource usage characteristics to determine the resource allocation operation.

4. The method of claim 1, further comprising presenting the resource allocation operation in a user interface.

5. The method of claim 4, further comprising receiving a selection of the resource allocation operation.

6. The method of claim 1, wherein the resource allocation operation is determined based at least in part on a resource allocation rule.

7. The method of claim 1, wherein the measurements pertain to at least one of, a CPU capacity usage, a storage capacity usage, a bandwidth capacity usage, or an IO contention, or a storage IO latency.

8. The method of claim 1, wherein determining the resource allocation operation is performed on a virtual machine above a hypervisor.

9. The method of claim 1, wherein the resource allocation operation pertains to at least one of, migrating a virtualized controller, migrating a virtual machine, migrating a container, adding a node, or adding storage capacity.

10. The method of claim 1, wherein each of the plurality of nodes in the cluster include a controller virtual machine above a hypervisor that presents the aggregation of one or more storage devices to one or more user virtual machines or containers.

11. A non-transitory computer readable medium having stored thereon a sequence of instructions which, when executed by a processor performs a set of acts comprising:

maintaining, at a storage pool, measurements corresponding to prior usage of a distributed storage resource of a cluster by a plurality of nodes in the cluster, the storage pool comprising an aggregation of one or more storage devices directly attached to respective nodes in the cluster;

receiving, at a first node of the plurality of nodes in the cluster, the measurements from the storage pool;

determining, at the first node, a time period for a training window corresponding to a portion of the measurements, wherein the training window bounds a periodically recurring portion of the measurements;

training, at the first node, a predictive model based at least in part on the periodically recurring portion of the measurements to produce a trained predictive model;

and

determining a resource allocation operation based at least in part on a prediction from the trained predictive model.

12. The computer readable medium of claim 11, wherein the set of acts further comprise executing the resource allocation operation.

13. The computer readable medium of claim 11, wherein the set of acts further comprise comparing a peak resource usage measurement and a mean resource usage measurement to a storage pool predicted resource usage characteristics to determine the resource allocation operation.

14. The computer readable medium of claim 11, wherein the set of acts further comprise presenting the resource allocation operation in a user interface.

15. The computer readable medium of claim 14, wherein the set of acts further comprise receiving a selection of the resource allocation operation.

16. The computer readable medium of claim 11, wherein the resource allocation operation is determined based at least in part on a resource allocation rule.

17. The computer readable medium of claim 11, wherein the measurements pertain to at least one of, a CPU capacity usage, a storage capacity usage, a bandwidth capacity usage, or an TO contention, or a storage TO latency.

18. The computer readable medium of claim 11, wherein determining the resource allocation operation is performed on a virtual machine above a hypervisor.

19. A system, comprising:

a storage medium having stored thereon a sequence of instructions; and

a processor that executes the sequence of instructions to perform a set of acts comprising;

maintaining, at a storage pool, measurements corresponding to prior usage of a distributed storage resource of a cluster by a plurality of nodes in the cluster, the storage pool comprising an aggregation of one or more storage devices directly attached to respective nodes in the cluster;

receiving, at a first node of the plurality of nodes in the cluster, the measurements from the storage pool;

determining, at the first node, a time period for a training window corresponding to a portion of the measurements, wherein the training window bounds a periodically recurring portion of the measurements;

training, at the first node, a predictive model based at least in part on the periodically recurring portion of the measurements to produce a trained predictive model;

and

determining a resource allocation operation based at least in part on a prediction from the trained predictive model.

20. The system of claim 19, wherein the measurements pertain to at least one of, a CPU capacity usage, a storage capacity usage, a bandwidth capacity usage, or an IO contention, or a storage IO latency.