# US 2010/0169253 A1 · Artificial neural network for balancing workload by migrating computing tasks across hosts

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

- Number: US 2010/0169253 A1
- Document ID: US20100169253A1
- Status: Published application
- Kind: A1
- Application number: 12/344,463
- Filed: 2008-12-27
- Published: 2010-07-01
- Earliest claimed: 2008-12-27
- Assignees: VMware, Inc.
- Inventors: Tzu-Chin TAN
- Web page: https://patentgrep.com/patent/US20100169253A1

## CPC

- G06F 9/5088: {involving task migration}
- G06F 2209/5019: Workload prediction
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Methods and apparatuses for balancing computing workload via migrating computing tasks are disclosed. An artificial neural network (ANN) is trained based on the workload distribution over time for a host. The ANN predicts the workload for the host, and an indication may be sent to migrate at least one computing task away from the host. The indication is sent when the method is operating in a proactive mode and when the predicted workload is outside of a desired operating range. Some embodiments monitor the workload; and automatically switch the method to the proactive mode, when a difference between the monitored workload and the predicted workload is small. Other embodiments monitor the workload; and automatically switch the method to a reactive mode, when the monitored workload is outside of a failsafe operating range for the particular host.

## Claims

### Claim 1

A method of balancing a computing workload, where the method is computer implemented and is adapted for use with at least two hosts that are configured to migrate computing tasks among themselves, the method comprising:
training a first artificial neural network (ANN), where training data for the ANN is based on a distribution over time of a computing workload for a particular one of the hosts;
predicting a predicted workload for the particular host when the method is operating in a prediction mode, where the predicting is performed by the first ANN; and
sending an indication to at least one of the hosts to migrate at least one of the computing tasks away from the particular host, when the method is operating in a proactive mode and when the predicted workload is outside of a proactive operating range for the particular host.

### Claim 2 (depends on 1)

The method of claim 1, where the indication includes, when the predicted workload is less than a low proactive threshold, a reduce power flag indicating that all computing tasks currently active on the particular host are to be migrated away from the particular host and then a reduced power mode is to be set for the particular host.

### Claim 3 (depends on 1)

The method of claim 1, where the indication includes, when the predicted workload is greater than a high proactive threshold, an offload flag indicating that at least one computing task currently active on the particular host is to be migrated away from the particular host.

### Claim 4 (depends on 1)

The method of claim 1, further comprising:
monitoring the computing workload for the particular host; and
automatically switching the method to the proactive mode, when a difference between the monitored workload and the predicted workload is less than an autostart accuracy threshold.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
monitoring the computing workload for the particular host; and
sending an indication to at least one of the hosts to migrate computing tasks away from the particular host, when the method is operating in a the reactive mode and when the monitored workload is outside of the reactive operating range for the particular host.

### Claim 6 (depends on 1)

The method of claim 1, further comprising:
monitoring the computing workload for the particular host; and
automatically switching the method to a reactive mode, when the monitored workload is outside of a failsafe operating range for the particular host.

### Claim 7 (depends on 6)

The method of claim 6, further comprising:
automatically switching the first ANN to the training mode, when the monitored workload is outside of the failsafe operating range for the particular host.

### Claim 8 (depends on 1)

The method of claim 1, further comprising:
displaying to a user any combination of information selected from: i) a set of configuration settings for the method; ii) a monitored workload of the particular host over a period of time; or iii) the predicted workload for the particular host over the same period of time; and
optionally receiving from the user one or more new values for the configuration settings.

### Claim 9 (depends on 8)

The method of claim 8, where the configuration settings include one or more selected from: a command for the ANN to enter a training mode; a command for the ANN to enter a predict only mode; a command to enter the proactive mode; a command to enter a reactive mode; a command to enter an autostart mode; a command to stay in a current mode; a command to disable automatic migration of a particular computing task; a command to require that at least two computing tasks be migrated together whenever any one of them is migrated; a priority for migration processes versus other computing tasks; a command to set an autostart accuracy threshold; a command to set a failsafe operating range; a command to set a proactive accuracy range; a command to set a reactive accuracy range; a command that applies to a category of days; or a command that applies to all days.

### Claim 10 (depends on 1)

The method of claim 1, where the computing workload is selected from one or more of: i) a processor utilization indication; ii) a memory utilization indication; or iii) an input output (IO) utilization indication.

### Claim 11 (depends on 1)

The method of claim 1, where the distribution over time of the computing workload is selected from one or more of: i) a distribution based on a time of day; or ii) a distribution based both on a time of day and on a day category.

### Claim 12 (depends on 1)

The method of claim 1, further comprising:
training a second artificial neural network (ANN), where training data for the ANN is based on a distribution over time of a computing workload for a particular one of the hosts, and where the training data only includes computing workload information for days within a second day category;
where the training data for the first ANN only includes computing workload information within a first day category; and
where the predicting is performed by the first ANN when a current day is within the first day category and is performed by the second ANN when the current day is within the second day category.

### Claim 13 (depends on 1)

The method of claim 1, where the at least one of the computing tasks to be migrated is executed by an operating system and where the migration also includes:
the operating system;
a memory image of the particular computing task;
a control state for any IO devices in use by the particular computing task; and
a processor state for the particular computing task.

### Claim 14 (depends on 13)

The method of claim 13, where the computing task, the operating system, the memory image, the disk image, the control state, and the processor state form a virtual machine.

### Claim 15

A medium readable by a computer, where the medium includes a sequence of instructions that when processed by the computer control the performance of a method for balancing a computing workload, where the method is adapted for use with at least two hosts that are configured to migrate computing tasks among themselves, and where the method comprises:
training a first artificial neural network (ANN), where training data for the ANN is based on a distribution over time of a computing workload for a particular one of the hosts;
predicting a predicted workload for the particular host when the method is operating in a prediction mode, where the predicting is performed by the first ANN; and
sending an indication to at least one of the hosts to migrate at least one of the computing tasks away from the particular host, when the method is operating in a proactive mode and when the predicted workload is outside of a proactive operating range for the particular host.

### Claim 16 (depends on 15)

The medium of claim 15, where the indication includes, when the predicted workload is less than a low proactive threshold, a reduce power flag indicating that all computing tasks currently active on the particular host are to be migrated away from the particular host and then a reduced power mode is to be set for the particular host.

### Claim 17 (depends on 15)

The medium of claim 15, where the indication includes, when the predicted workload is greater than a high proactive threshold, an offload flag indicating that at least one computing task currently active on the particular host is to be migrated away from the particular host.

### Claim 18 (depends on 15)

The medium of claim 15, where the method further comprises:
monitoring the computing workload for the particular host; and
automatically switching the method to the proactive mode, when a difference between the monitored workload and the predicted workload is less than an autostart accuracy threshold.

### Claim 19 (depends on 15)

The medium of claim 15, where the method further comprises:
monitoring the computing workload for the particular host; and
sending an indication to at least one of the hosts to migrate computing tasks away from the particular host, when the method is operating in a the reactive mode and when the monitored workload is outside of the reactive operating range for the particular host.

### Claim 20 (depends on 15)

The medium of claim 15, where the method further comprises:
monitoring the computing workload for the particular host; and
automatically switching the method to a reactive mode, when the monitored workload is outside of a failsafe operating range for the particular host.

### Claim 21 (depends on 20)

The medium of claim 20, where the method further comprises:
automatically switching the first ANN to the training mode, when the monitored workload is outside of the failsafe operating range for the particular host.

### Claim 22 (depends on 15)

The medium of claim 15, where the method further comprises:
displaying to a user any combination of information selected from: i) a set of configuration settings for the method; ii) a monitored workload of the particular host over a period of time; or iii) the predicted workload for the particular host over the same period of time; and
optionally receiving from the user one or more new values for the configuration settings.

### Claim 23 (depends on 22)

The medium of claim 22, where the configuration settings include one or more selected from: a command for the ANN to enter a training mode; a command for the ANN to enter a predict only mode; a command to enter the proactive mode; a command to enter a reactive mode; a command to enter an autostart mode; a command to stay in a current mode; a command to disable automatic migration of a particular computing task; a command to require that at least two computing tasks be migrated together whenever any one of them is migrated; a priority for migration processes versus other computing tasks; a command to set an autostart accuracy threshold; a command to set a failsafe operating range; a command to set a proactive accuracy range; a command to set a reactive accuracy range; a command that applies to a category of days; or a command that applies to all days.

### Claim 24 (depends on 15)

The medium of claim 15, where the computing workload is selected from one or more of: i) a processor utilization indication; ii) a memory utilization indication; or iii) an input output (IO) utilization indication.

### Claim 25 (depends on 15)

The medium of claim 15, where the distribution over time of the computing workload is selected from one or more of: i) a distribution based on a time of day; or ii) a distribution based both on a time of day and on a day category.

### Claim 26 (depends on 15)

The medium of claim 15, where the method further comprises:
training a second artificial neural network (ANN), where training data for the ANN is based on a distribution over time of a computing workload for a particular one of the hosts, and where the training data only includes computing workload information for days within a second day category;
where the training data for the first ANN only includes computing workload information within a first day category; and
where the predicting is performed by the first ANN when a current day is within the first day category and is performed by the second ANN when the current day is within the second day category.

### Claim 27 (depends on 15)

The medium of claim 15 where the at least one of the computing tasks to be migrated is executed by an operating system and where the migration also includes:
the operating system;
a memory image of the particular computing task;
a control state for any IO devices in use by the particular computing task; and
a processor state for the particular computing task.

### Claim 28 (depends on 27)

The medium of claim 27, where the computing task, the operating system, the memory image, the disk image, the control state, and the processor state form a virtual machine.

### Claim 29

A system for balancing a computing workload, where the system is adapted for use with at least two hosts that are configured to migrate computing tasks among themselves, the system comprising:
an artificial neural network (ANN), configured to process training data that is based on a distribution over time of a computing workload for a particular one of the hosts, and configured to predict a predicted workload for the particular host when the system is operating in a prediction mode, where the particular host has a proactive operating range; and
a workload controller configured to send an indication to the hosts to migrate at least one of the computing tasks away from the particular host, when the system is operating in a proactive mode and when the predicted workload is outside of the proactive operating range.

### Claim 30

A system adapted for use with a host means operable to migrate computing tasks among particular hosts within the host means, the system comprising:
artificial neural network (ANN) means for training, when operating in a training mode, where training data for the ANN is based on a distribution over time of a computing workload for a particular one of the hosts within the host means, and where the ANN means is further a means for predicting, when operating in a prediction mode, a predicted workload for the particular host means; and
control means for sending, when operating in a proactive mode, an indication to the host means to migrate at least one virtual machine (VM) away from the particular host when the predicted workload is outside of a proactive operating range for the particular host means.

### Claim 31 (depends on 30)

The system of claim 30, further comprising:
means for monitoring the computing workload for the particular host means; and
means for automatically switching the control means to the proactive mode, when a difference between the monitored workload and the predicted workload is less than an autostart accuracy threshold.

### Claim 32 (depends on 30)

The system of claim 30, further comprising:
means for monitoring the computing workload for the particular host;
where the control means is further a means for automatically switching to a reactive mode, when the monitored computing workload is outside of a proactive operating range for the particular host, where when in the reactive mode the control means signals the host means to migrate at least one VM away from the particular host based on the monitored workload.

### Claim 33 (depends on 32)

The system of claim 32, where the control means is further a means for switching the ANN to the training mode, when the monitored computing workload is outside of a proactive operating range for the particular host means.

### Claim 34 (depends on 30)

The system of claim 30, where the VM also includes:
a memory image for the particular computing task;
a disk image for the particular computing task; and
a control state for any IO devices in use by the particular computing task.

### Claim 35 (depends on 34)

The system of claim 34, where at least a substantial part of the memory image within the VM is migrated to a destination host means prior to ending execution of the VM on the particular host means.
