Abstract: Systems, methods, and other embodiments associated with multi-distance clustering are described. In one embodiment, a method includes reading a multi-distance similarity matrix S that records pair-wise multi-distance similarities between respective pairs of data points in a data set. Each pair-wise similarity is based on distances between a pair of data points calculated using K different distance functions, where K is greater than one. The method includes clustering the data points in the data set into n clusters based on the similarity matrix S. The number of clusters n is not determined prior to the clustering.

Abstract: The disclosed embodiments provide a system that prevents oscillatory load behavior for a multi-node distributed system. During operation, the system uses a load-balancing policy to distribute requests to nodes of the distributed system. The system determines operational characteristics for the nodes as they process a set of requests, and then uses these operational characteristics to compute machine queuing models that describe the machine state of each node. The system then uses this machine state for the nodes to determine whether the load-balancing policy and the distributed system are susceptible to oscillatory load behavior.

Abstract: A method for generating instructions for a monitoring facility. The method includes obtaining a signal from a data server, where the data server is monitored by the monitoring facility using a task, identifying a signal characteristic of the signal, and determining a first category for the signal based on the signal characteristic. The method further includes identifying a first signal processing tool for the signal based on the task and the first category and instructing the monitoring facility to perform the task by applying the first signal processing tool to the signal.

Abstract: A method for processing signals from a data server system including generating, by a monitoring module on a monitoring facility, a plurality of frequency data items, where the monitoring facility comprises a central processing unit, obtaining, by the monitoring module, a first signal value from the data server system, where the signal value is a measurement of an element of the data server system, and calculating a plurality of fit errors for the plurality of frequency data items using the first signal value. The method further includes selecting a frequency data item of the plurality of frequency data items with a lowest fit error to obtain a selected frequency data item, removing a cyclic component of the first signal value using the selected frequency data item to obtain a first processed signal value, and displaying the first processed signal value on a graph.

Abstract: Systems, methods, and other embodiments associated with anomaly detection using tripoint arbitration are described. In one embodiment, a method includes identifying a set of clusters that correspond to a nominal sample of data points in a sample space. A point z is determined to be an anomaly with respect to the nominal sample when, for each cluster, a tripoint arbitration similarity between data points in the cluster calculated with z as arbiter is greater than a threshold.

Abstract: Systems, methods, and other embodiments associated with similarity analysis using tri-point arbitration are described. In one embodiment, a method includes selecting a data point pair and an arbiter point from a data set. A tri-point arbitration coefficient (?TAC) is calculated for data point pairs based, at least in part, on a distance between the first and second data points and the arbiter point. A similarity metric is determined for the data set based, at least in part, on an aggregation of tri-point arbitration coefficients for data point pairs in the set of data points using the selected arbiter point.

Abstract: Systems, methods, and other embodiments associated with transient detection for predictive health management are described. In one embodiment, a method includes receiving a health signal from a data outlet. The health signal corresponds to a derivative variable derived from a combination of data processing system metrics not exposed beyond the data outlet. A transient is detected in the health signal. In response to detecting the transient, the method includes performing a corrective action. The example method may also include detecting the transient in a health signal from a data processing system by collecting historical values of the health signal; selecting a first statistical model that best fits the historical values; receiving a present value of the health signal; and applying the first statistical model to the present value of the health signal to determine whether the present value of the health signal is a transient.

Abstract: Systems, methods, and other embodiments associated with adaptively determining a preventive maintenance schedule based on historical system operation are described. The prognostic parameter values are continuously partitioned into a number of operating states based on observed maintenance costs associated with the prognostic parameter values. The operating states range from absolutely healthy, one or more degrees of degradation, to fully degraded. A system cost function is used as the discriminant function. The system cost function is an expected maintenance cost when a given preventive maintenance (PM) schedule is adopted. The system cost function calculates the expected cost based on the observed cost of operation in each of the operating states and a probability of the computing system being in each of the operating states as determined by the PM schedule. The PM schedule that minimizes the cost function is determined to be the optimal PM schedule.

Abstract: In general the invention relates to a method for processing signals from a data server. The method includes obtaining, by a monitoring facility, a measured signal from the data server, wherein the monitoring facility comprises a central processing unit, generating a first probing characteristic from the measured signal, and determining that the first probing characteristic is measureless. The method further includes decomposing, by the central processing unit in response to the determination, the first probing characteristic into a plurality of component parts, constructing a reconstructed signal using a first one of the plurality of component parts, generating a second probing characteristic using the reconstructed signal, wherein the second probing characteristic is not measureless, and processing the reconstructed signal by the monitoring facility.

Abstract: Systems, methods, and other embodiments associated with tripoint arbitration for data classification are described. In one embodiment, a method includes receiving a query data point for classification, an arbiter point, a first set of classified data points belonging to a first class and a second set of classified data points belonging to a second class. A class-aggregated similarity metric for the query data point and each set is determined based, at least in part, on tripoint arbitration coefficients for data point pairs that include the query data point and a classified data point in the set, as calculated using the arbiter point. The query data point is classified based on a comparison of the similarity metric determined for the first set and the similarity metric determined for the second set.

Abstract: Systems, methods, and other embodiments associated with decision making with analytically combined split conditions are provided. In one embodiment, a method for classifying data is provided. An input data sample is received for classification as belonging to one of two possible classes. The input data sample includes a set of attribute values. The method includes evaluating the set of attribute values with a tree function that defines a decision boundary of a classification tree. The tree function classifies an input data sample as belonging to one of the two possible classes based, at least in part, on the attribute values of the input data sample. In another embodiment parameters of the tree function are derived by applying a gradient descent parameter update rule to the training data samples.

Abstract: Some embodiments of the present invention provide a system that characterizes a response of a component in a computer system to vibrations generated by the computer system. First, the system measures the response of the component to vibrations in a frequency range while the component is located outside of the computer system. The system also measures vibrations generated by the computer system in the frequency range during operation of the computer system, wherein the vibrations are measured at a location in the computer system which is configured to receive the component. The system then characterizes the response of the component to vibrations generated by the computer system based on the measured response of the component to vibrations in the frequency range and the measured vibrations in the frequency range at the location.

Abstract: Systems, methods, and other embodiments associated with clustering using tri-point arbitration are described. In one embodiment, a method includes selecting a data point pair and a set of arbiter points. A tri-point arbitration similarity is calculated for data point pairs based, at least in part, on a distance between the first and second data points and the arbiter points. In one embodiment, similar data points are clustered.

Abstract: A method for generating instructions for a monitoring facility. The method includes obtaining a signal from a data server, where the data server is monitored by the monitoring facility using a task, identifying a signal characteristic of the signal, and determining a first category for the signal based on the signal characteristic. The method further includes identifying a first signal processing tool for the signal based on the task and the first category and instructing the monitoring facility to perform the task by applying the first signal processing tool to the signal.

Abstract: A method for processing signals from a data server system including generating, by a monitoring module on a monitoring facility, a plurality of frequency data items, where the monitoring facility comprises a central processing unit, obtaining, by the monitoring module, a first signal value from the data server system, where the signal value is a measurement of an element of the data server system, and calculating a plurality of fit errors for the plurality of frequency data items using the first signal value. The method further includes selecting a frequency data item of the plurality of frequency data items with a lowest fit error to obtain a selected frequency data item, removing a cyclic component of the first signal value using the selected frequency data item to obtain a first processed signal value, and displaying the first processed signal value on a graph.

Abstract: A method for generating a service action for a computer system is described. During the method, a longevity index value for a packaging technology in the computer system is calculated using a longevity model and thermal and vibration telemetry data that is collected in the computer system. This longevity model may be based on accelerated failure testing of the packaging technology, field failures of the packaging technology in a group of computer systems that includes the computer system, and/or thermal and vibration telemetry data for the group of computer systems. Furthermore, using the longevity index value, the service action for the computer system is determined. Based on the longevity index value, remedial as action, e.g., repairs to the computer system, may be scheduled and performed.

Abstract: Systems, methods, and other embodiments associated with similarity analysis using tri-point arbitration are described. In one embodiment, a method includes selecting a data point pair and an arbiter point from a data set. A tri-point arbitration coefficient (?TAC) is calculated for data point pairs based, at least in part, on a distance between the first and second data points and the arbiter point. A similarity metric is determined for the data set based, at least in part, on an aggregation of tri-point arbitration coefficients for data point pairs in the set of data points using the selected arbiter point.

Abstract: In general the invention relates to a method for processing signals from a data server. The method includes obtaining, by a monitoring facility, a measured signal from the data server, wherein the monitoring facility comprises a central processing unit, generating a first probing characteristic from the measured signal, and determining that the first probing characteristic is measureless. The method further includes decomposing, by the central processing unit in response to the determination, the first probing characteristic into a plurality of component parts, constructing a reconstructed signal using a first one of the plurality of component parts, generating a second probing characteristic using the reconstructed signal, wherein the second probing characteristic is not measureless, and processing the reconstructed signal by the monitoring facility.

Abstract: A system for characterizing vibration of a rack structure having at least one hard disk drive disposed therein and a vibration exciter operatively associated with the rack structure includes one or more computers. The one or more computers are configured to command the vibration exciter to provide vibration input to the rack structure and to command a plurality of reads from the at least one hard disk drive such that, for each of the plurality of reads, data stored on the at least one hard disk drive is retrieved from the at least one hard disk drive. The one or more computers are also configured to obtain read rate information related to the plurality of reads and to identify at least one resonant frequency of the rack structure based on the read rate information.

Abstract: Systems, methods, and other embodiments associated with adaptively determining a preventive maintenance schedule based on historical system operation are described. The prognostic parameter values are continuously partitioned into a number of operating states based on observed maintenance costs associated with the prognostic parameter values. The operating states range from absolutely healthy, one or more degrees of degradation, to fully degraded. A system cost function is used as the discriminant function. The system cost function is an expected maintenance cost when a given preventive maintenance (PM) schedule is adopted. The system cost function calculates the expected cost based on the observed cost of operation in each of the operating states and a probability of the computing system being in each of the operating states as determined by the PM schedule. The PM schedule that minimizes the cost function is determined to be the optimal PM schedule.