摘要:
A method, system and program storage device for measurement acquisition using predictive models which: (a) can improve scalability; (b) can accommodate devices that operate in a disconnected mode; and (c) enable integration of data from sources with different time granularities. Various features can be embodied in software and an object-oriented implementation is described. Different implementations are described, such as standalone predictive models implemented only on a manager (for example systems management/load balancing) or managed system (for example router management); or a parallel implementation with predictive models running on both the manager and managed (agent) systems (for example financial trading or system health monitoring). In a parallel model implementation, the agent constructs a predictive model that is conveyed to the manager system. The models are used in parallel, possibly with no communication for an extended time. The manager uses its model to provide tentative values of measurement variables to management applications. The agent uses its model to check its accuracy. If the model is found to be insufficiently accurate, an updated model is transmitted to the manager. The invention allows other measurement acquisition protocols to operate concurrently on the same measurement variables.
摘要:
A method, system and program storage device for measurement acquisition using predictive models which: (a) can improve scalability; (b) can accommodate devices that operate in a disconnected mode; and (c) enable integration of data from sources with different time granularities. Various features can be embodied in software and an object-oriented implementation is described. Different implementations are described, such as standalone predictive models implemented only on a manager (for example systems management/load balancing) or managed system (for example router management); or a parallel implementation with predictive models running on both the manager and managed (agent) systems (for example financial trading or system health monitoring). In a parallel model implementation, the agent constructs a predictive model that is conveyed to the manager system. The models are used in parallel, possibly with no communication for an extended time. The manager uses its model to provide tentative values of measurement variables to management applications. The agent uses its model to check its accuracy. If the model is found to be insufficiently accurate, an updated model is transmitted to the manager. The invention allows other measurement acquisition protocols to operate concurrently on the same measurement variables.
摘要:
Automated or autonomic techniques for managing deployment of one or more resources in a computing environment based on varying workload levels. The automated techniques may comprise predicting a future workload level based on data associated with the computing environment. Then, an estimation is performed to determine whether a current resource deployment is insufficient, sufficient, or overly sufficient to satisfy the future workload level. Then, one or more actions are caused to be taken when the current resource deployment is estimated to be insufficient or overly sufficient to satisfy the future workload level. Actions may comprise resource provisioning, resource tuning and/or admission control.
摘要:
Techniques for use in accordance with application performance decomposition are provided which take advantage of the communications protocol used to carry a transaction between application components in a distributed computing network. Specifically, the invention extends the communications protocol by embedding data, such as timestamp and duration measurement data, in the protocol itself, rather than extending or altering the application or transaction data carried by the protocol as in existing approaches. Thus, the invention provides natural correlation of interactions of distributed application components on such transactions without modification to the application or transaction data. Because the correlation is performed in-line with the application component interactions, minimal data management overhead is required, and correlated performance decomposition is made possible in real-time for the transaction. Furthermore, subsequent processing stages of the distributed application can interpret the communications protocol to glean processing durations of previous stages in order to make decisions regarding treatment of the transaction.
摘要:
In one embodiment, functional system elements are added to an autonomic manager to enable automatic online sample interval selection. In another embodiment, a method for determining the sample interval by continually characterizing the system workload behavior includes monitoring the system data and analyzing the degree to which the workload is stationary. This makes the online optimization method less sensitive to system noise and capable of being adapted to handle different workloads. The effectiveness of the autonomic optimizer is thereby improved, making it easier to manage a wide range of systems.
摘要:
Techniques for performing adaptive and robust prediction. Prediction techniques are adaptive in that they use a minimal amount of historical data to make predictions, the amount of data being selectable. The techniques are able to learn quickly about changes in the workload traffic pattern and make predictions, based on such learning, that are useful for proactive response to workload changes. To counter the increased variability in the prediction as a result of using minimal history, robustness is improved by checking model stability at every time interval and revising the model structure as needed to meet designated stability criteria. Furthermore, the short term prediction techniques can be used in conjunction with a long term forecaster.
摘要:
Computer-based methods and systems for performing automated distribution of a software package to one or more target machines in one or more regions of a distributed network of target machines, comprises the following steps. First, a base software package is prepared for each of the one or more regions based on at least one of: (i) policy data indicating which of the one or more regions are candidates for receiving the software package, (ii) dependency information indicating requisites for a service provided by the software package, and (iii) configuration information for each of the candidate regions. The base software package is then distributed to each of the candidate regions of the distributed network. The base software package received at each of the candidate regions is then customized based on at least one of: (i) regional distribution policies, (ii) dependency information specific to one or more roles performed by the target machines in that region, and (iii) individual target machine configuration information. Lastly, the software package customized in each of the candidate regions is distributed to at least one of the target machines in the candidate regions of the distributed network.
摘要:
Predictive models are widely used for tasks in many domains. The present invention addresses the problem of prediction of non-stationary processes by dynamically managing multiple models. The system comprises a model assessor, a model adapter, a plurality of sub-models, a plurality of model combiner functions, training data that is used to estimate model parameters, and test data that is used to test for change points. Two processes are described, one for handling data updates and another that addresses prediction requests.
摘要:
A new form of pattern is provided, referred to as a mutual dependence pattern or m-pattern. The m-pattern captures mutual dependence among a set of items. Intuitively, the m-pattern represents a set of items that often occur together. In our experience, such m-patterns often provide great values for certain tasks, such as event correlation in event management. Further, an efficient algorithm is provided for discovering all m-patterns in data for a given minimum mutual dependence threshold. Specifically, a linear algorithm is provided for testing whether a pattern is an m-pattern. Further, a pruning algorithm is provided that prunes the search space effectively. Still further, a level-wise algorithm for mining m-patterns is provided.
摘要:
Methods and apparatus are described that use models of measurement variables to provide self-adjusting policies that: (a) reduce the administrative overhead of specifying thresholds and (b) provide a means for pro-active management by automatically constructing warning thresholds based on the probability of an alarm occurring within a time horizon. The invention includes components for model construction, threshold construction, policy evaluation, and action taking. Several kinds of meta policies are used in accordance with the invention to provide adaptation and pro-active management.