# US 2006/0188011 A1 · Automated diagnosis and forecasting of service level objective states

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

- Number: US 2006/0188011 A1
- Document ID: US20060188011A1
- Status: Published application
- Kind: A1
- Application number: 10/987,611
- Filed: 2004-11-12
- Published: 2006-08-24
- Earliest claimed: 2004-11-12
- Assignees: Hewlett-Packard Development Company, L.P.
- Inventors: Moises Goldszmidt; Ira Cohen; Terence Kelly; Julie Symons
- Web page: https://patentgrep.com/patent/US20060188011A1

## CPC

- G06Q 10/04: Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202)
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

Systems, methods, and software used in performing automated diagnosis and identification of or forecasting service level object states. Some embodiments include building classifier models based on collected metric data to detect and forecast service level objective (SLO) violations. Some such systems, methods, and software further include automated detecting and forecasting of SLO violations along with providing alarms, messages, or commands to administrators or system components. Some such messages include diagnostic information with regard to a cause of a SLO violation. Some embodiments further include storing data representative of system performance and detected and forecast system SLO states. This data can then be used to generate reports of system performance including representations of system SLO states.

## Claims

### Claim 1

A method of automated diagnostics and forecasting of service level objective (SLO) states, the method comprising:
collecting datasets from one or more metrics over time, each dataset including a state value representative of compliance with a service level objective;
synchronizing the collected datasets; and
building a classifier model based on the datasets and their state values, wherein the model includes a number of SLO state representations equal to a number of unique SLO state values represented in the collected datasets.

### Claim 2 (depends on 1)

The method of claim 1, wherein building classifier models based on the datasets and their state values comprises:
dividing the datasets into regions representative of each unique state value, wherein each dataset is divided into the region representative of its state value;
identifying and ranking, for each region, key metrics with values correlated with a state of the region; and
computing, for each region, sufficient statistics for at least the key metrics, the sufficient statistics defining the classifier model for each region.

### Claim 3 (depends on 2)

The method of claim 2, wherein identifying key metrics further includes one or more previously identified key metrics, and values for those metrics, to include in at least one of classifier models.

### Claim 4 (depends on 1)

The method of claim 1, wherein building classifier models based on the datasets and their state values comprises:
dividing the datasets into regions representative of each unique state value, wherein each dataset is divided into the region representative of its state value;
computing sufficient statistics for each metric of the datasets for each of the two regions;
computing a relevance to an SLO state for each metric in each region, the relevance being a probability that the metric is indicative of the SLO state;
ordering the metrics in each region by relevance; and
selecting one or more of the metrics in each region to form a SLO state representation for each region.

### Claim 5 (depends on 4)

The method of claim 4, further comprising:
testing the classifier model to determine an accuracy of the classifier model using the collected datasets, wherein the accuracy of the model is a probability that the model correctly identifies the SLO states associated with the collected datasets.

### Claim 6 (depends on 5)

The method of claim 5, wherein testing the classifier model to determine an accuracy of the classifier model comprises:
extracting metrics measured at a particular point in time;
determining which SLO state representation each of the metrics most closely fits;
comparing the determined SLO state representation with the actual SLO state of the metrics; and
computing an accuracy probability for the model based on the comparing.

### Claim 7 (depends on 5)

The method of claim 5, further comprising:
comparing the accuracy probability of the classifier model with an accuracy threshold; and
rebuilding the classifier model if it fails to meet the accuracy threshold, wherein rebuilding the model includes manipulating the metrics included in the model.

### Claim 8 (depends on 1)

The method of claim 1, wherein the number of unique state values is two, a first state value representative of SLO compliance and a second state value representative of SLO violation.

### Claim 9

A method of system diagnosis comprising:
inducing a probabilistic model of a relationship between a first set of n metrics collected over time to a set of m Service Level Objective (SLO) system states correlated in time to when the first set of n metrics were measured; and
applying the model to a second set of n metrics collected at t point in time to determine one of the m system SLO states to which the second set of n metrics is mostly correlated.

### Claim 10 (depends on 9)

The method of claim 9, further comprising:
sending a message to a network system component in response to a determined system SLO state.

### Claim 11 (depends on 10)

The method of claim 10, wherein the system SLO state is a SLO violation and the message causes the network system component to perform an action in response to the SLO violation.

### Claim 12 (depends on 9)

The method of claim 9, further comprising:
rebuilding the model upon an occurrence of an event.

### Claim 13 (depends on 12)

The method of claim 12, wherein the event is a passage of a predetermined period of time.

### Claim 14 (depends on 13)

The method of claim 13, wherein the event includes a detection that the model is no longer indicative of system performance.

### Claim 15 (depends on 9)

The method of claim 9, further comprising:
storing data from this inducing of the probabilistic model and applying the model;
generating a report based upon the stored data, wherein the report provides a historical view of system performance; and
outputting the report.

### Claim 16 (depends on 15)

The method of claim 15, wherein the historical view of system performance includes a representation of system performance in relation to a SLO.

### Claim 17 (depends on 9)

The method of claim 9, wherein inducing the model comprises:
time-shifting the system states to the first set of n metrics measured Δ period of time prior to t point in time when the system SLO states were measured.

### Claim 18 (depends on 17)

The method of claim 17, wherein Δ period of time is five minutes.

### Claim 19 (depends on 17)

The method of claim 17, wherein a result of applying the model to the second set of n metrics is a forecasted system SLO state for t+Δ point in time.

### Claim 20 (depends on 17)

The method of claim 17, further comprising:
sending a message to a network system component in response to a forecasted system state.

### Claim 21 (depends on 20)

The method of claim 20, wherein the system SLO state is a SLO violation and the message causes the network system component to perform an action in response to the SLO violation.

### Claim 22

A computer readable medium, with instructions thereon to cause a computer to:
induce a model of a relationship between a first set of n metrics collected over time to a set of m Service Level Objective (SLO) system states correlated in time to when the first set of n metrics were measured; and
apply the model to a second set of n metrics collected at t point in time to determine one of the m system SLO states to which the second set of n metrics is mostly correlated.

### Claim 23 (depends on 22)

The computer readable medium of claim 22, wherein the instructions further cause a computer to:
rebuild the classifier model at a regular interval.

### Claim 24 (depends on 23)

The computer readable medium of claim 23, wherein the regular interval is daily.

### Claim 26 (depends on 22)

The computer readable medium of claim 22, wherein the system comprises a group of one or more devices operable on a network.

### Claim 26 (depends on 22)

The computer readable medium of claim 22, wherein the instructions to cause the computer to induce the model comprise instructions to cause the computer to:
time-shift the system SLO states to the first set of n metrics measured Δ period of time prior to t point in time when the system SLO states were measured; and
apply the model to the second set of n metrics to generate a system SLO state forecast for t+Δ point in time.

### Claim 27

A method comprising:
receiving a current metric set of measured computing system performance;
comparing at least a subset of the metrics in the current metric set to a classifier model, the classifier model including two or more Service Level Objective (SLO) state representations;
determining which SLO state representation the metric subset most closely fits; and
generating an alarm if the SLO state representation to which the metric subset most closely fits is a SLO state representation for a negative system state, wherein the alarm includes diagnostic information about a probable cause for the negative system state.

### Claim 28 (depends on 27)

The method of claim 27, wherein the classifier model for the negative system state is representative of a SLO violation state.

### Claim 29 (depends on 27)

The method of claim 27, wherein the classifier model is generated from historical metric sets correlated to subsequent system states measured Δ period of time after the historical metric sets.

### Claim 30 (depends on 29)

The method of claim 29, wherein the generated alarm is a forecast of a future system state Δ period of time after measurement of the current metric set.

### Claim 31

A system comprising:
a processor;
a memory;
a network interface;
software in the memory and operable on the processor to:
collect metric data over the network interface, the metric data including an indicator of a service level objective state correlated with items of the metric data;
generate a service level objective state representation for each of the service level objective states correlated to the items of metric data;
testing the SLO state representations for accuracy and regenerating SLO state representations substantially failing to meet an accuracy threshold;
receiving subsequent metric data over the network interface; and
identifying a service level objective state using the SLO state representations, the service level objective state identified by determining which SLO state representation the subsequent metric data most closely fits.

### Claim 32 (depends on 31)

The system of claim 31, wherein the SLO state representations are part of a probabilistic based classifier model.

### Claim 33 (depends on 31)

The system of claim 31, wherein identifying a service level objective state using the SLO state representations includes a forecast of a system state.

### Claim 34 (depends on 31)

The system of claims 31 wherein identifying a service level objective state includes diagnosing a cause of a service level objective violation.
