International Business Machines Corporation

Karla Kay Arndt

Aspen L Payton

James M Caffrey

Nicholas C Matsakis

David Harold Surman

Joseph W Gentile

Nicholas R Jones

Keyur Patel

John Ryan McGarvey

Robert M Abrams

Juliet C Candee

Mark Joseph Cavage

William Robert Richardson

John D Sullivan

Kristin Marie Hazlewood

Dieter Wellerdiek

Robert L Snyder

Friedrich Matthias Gubitz

John Ryan Mcgarvey

Richard Allyn Heller

A method, system, and computer program product are provided for performance anomaly detection. Velocity data is periodically received from a workload manager for one or more address spaces. An expected velocity value is created for each of the one or more address spaces. A factor of the expected velocity value is compared to a current velocity value from the velocity data. Based on the current velocity value being lower than the factor, a remedial action is generated indicating an anomaly.

Programmatic performance anomaly detection

Techniques are described for monitoring a performance metric. A multiple modeling approach is used to improve predictive analysis by avoiding the issuance of warnings during spikes which occur as a part of normal system processing. This approach increases the accuracy of predictive analytics on a monitored computing system, does not require creating rules defining periodic processing cycles, reduces the amount of data required to perform predictive modeling, and reduces the amount of CPU required to perform predictive modeling.

Multiple modeling paradigm for predictive analytics

Present disclosure relates to methods, processing systems and computer program products of analyzing contention data and following resource blockers to find root causes of computer problems. The method may include: detecting one or more resource waiters in a computer system, iteratively determining whether the resource blockers are a resource waiter, until a final resource blocker not waiting for another resource is found, determining, whether final resource blocker is caused by a resource blocker in a different computer system, iteratively executing, the method on the different computer system to find the final resource blocker not waiting for another resource is found, determining, whether the final resource blocker has more than one symptom that may or may not be a contention problem, selecting a symptom that has the highest priority as the root cause of the computer problems, and generating, using the processor, a report of root causes of the computer problems.

Analyzing contention data and following resource blockers to find root causes of computer problems

Aspects relate to a computer system and a computer implemented method for resolving abnormal contention on the computer system. The method includes detecting, using a processor and at a hypervisor level of the computer system, abnormal contention of a serially reusable resource caused by a first virtual machine. The abnormal contention includes the first virtual machine experiencing resource starvation of computer system resources used for processing the first virtual machine, causing the first virtual machine to block the serially reusable resource from a second virtual machine that is waiting to use the serially reusable resource. The method also includes adjusting resource allocation at the hypervisor level of the computer system resources for the first virtual machine, processing the first virtual machine based on the resource allocation, and releasing the serially reusable resource by the first virtual machine in response to the first virtual machine processing.

Method and apparatus for resolving contention at the hypervisor level

Aspects relate to a computer implemented method for resolving abnormal contention on a computer system. The method includes detecting, using a processor, abnormal contention of a serially reusable resource caused by a first process, wherein the abnormal contention includes the first process blocking the serially reusable resource from a second process that is waiting to use the serially reusable resource. The method includes collecting, in a computer storage medium, resource data in a serialized resource history database and analyzing the resource data associated with the serially reusable resource and adjusting, using the processor, resource allocation for the first process of the serially reusable resource based on the resource data. The method also includes processing, using the processor and the serially reusable resource, the first process based on the resource allocation and releasing, using the processor, the serially reusable resource by the first process in response to the first process completing.

Method and apparatus for resolving contention in a computer system

Systems, methods, and computer program products to perform an operation comprising collecting, metric data for a first job upon determining that the first job: uses a first resource of a computing system at a level that exceeds a first threshold, wherein the metric data describes a usage level of the first resource by the first job, and has been executing for a duration of time that exceeds a time threshold.

Dynamically choosing data to collect in a system

Method to perform an operation comprising collecting, metric data for a first job upon determining that the first job: uses a first resource of a computing system at a level that exceeds a first threshold, wherein the metric data describes a usage level of the first resource by the first job, and has been executing for a duration of time that exceeds a time threshold.

Techniques are described for determining what node of a classification and regression tree (CART) should be used by a predictive analysis application. A first approach is to use a standard deviation of the data at a given the level of the CART to determine whether data in the next, lower node is more consistent than the data in the current node. A second approach is to measure a correlation between data points in a given node and the time at which each point was sampled (or other correlation metric) to identify a preferred node.

Determining a preferred node in a classification and regression tree for use in a predictive analysis

A computing system resource recovery method can include identifying a resource manager associated with a computing transaction, classifying the computing transaction to determine a predetermined metric, measuring an actual metric of the computing transaction, comparing the predetermined metric to the actual metric to detect abnormal behavior in the transaction and modeling the abnormal behavior to determine how the resource manager is affected by the abnormal behavior.

User-coordinated resource recovery

In an embodiment, deltas are calculated between respective current metric values for respective entities and previous metric values for the respective entities. A subset of the deltas is determined. A sum of the subset is calculated, and the sum is divided by a number of the subset to create an average delta for the subset. If one of the respective entities has one of the deltas that is greater than or equal to the average delta for the subset and the one of the respective entities was not previously used to create the previous prediction, then a current prediction is created.

Determining when to create a prediction based on deltas of metric values

A computing system includes a processor configured to identify a resource manager associated with a computing transaction, classify the computing transaction to determine a predetermined metric, measure an actual metric of the computing transaction, compare the predetermined metric to the actual metric to detect abnormal behavior in the transaction and model the abnormal behavior to determine how the resource manager is affected by the abnormal behavior.

Growing community of inventors

Rochester, MN, United States of America

Karla Kay Arndt