Abstract: In a datacenter setting, a summary of differences and similarities between two or more states of the same or similar systems are predicted. Initially, a Long Short-Term Memory (LSTM) neural network is trained with to predict a summary describing the state change between at least two states of the datacenter. Given a set of training data (at least two datacenter states that are annotated with a state change description), the LSTM neural network learns which similarities and differences between the datacenter states correspond to the annotations. Accordingly, given a set of test data comprising at least two states of a datacenter represented by context graphs that indicate a plurality of relationships among a plurality of nodes corresponding to components of a datacenter, the LSTM neural network is able to determine a state change description that summarizes the differences and similarities between the at least two states of the datacenter.

Abstract: Determining a similarity between a pair of graphs or patterns can be a computationally expensive and time-consuming process. To reduce the similarity calculation costs, patterns can be simplified based on equivalent classes of components. A similarity score can be calculated between nodes of a pattern. The nodes which represent a same component type and have similar attributes will likely have a high similarity score and can be combined into a single node representing the entire class of the components. The decision to combine nodes also considers a node's topological features such as relationships and connections to other nodes. By combining equivalent nodes, the search space for mapping and determining similarity between two graphs can be reduced. Reducing the search space, exponentially reduces the number of iterations required for determining an optimal similarity score and improves the performance and scalability of the overall root cause analysis framework.

Abstract: To aid in the root cause analysis of current system errors or anomalies, a graph-based root cause analysis software determines whether a graph representing an anomalous region of a system, referred to as a pattern, is similar to a previously stored pattern in a pattern library. The analysis software extracts a sub-graph or pattern representing components currently experiencing an anomaly from an overall system graph. The analysis software calculates a similarity score based on the comparison of the extracted pattern to patterns in the pattern library. The patterns in the pattern library represent previously encountered anomalies and include attributes, event data, expert/system administrator notes, etc., that can aid in diagnosing the current system anomaly.

Abstract: In a datacenter setting, annotations or descriptions of relevant parts or subgraphs corresponding to components in the datacenter are predicted. Given a set of training data (library of subgraphs seen in the past labeled with a textual description explaining why were they considered relevant enough to be placed in the historical database), the recurrent neural network (RNN) learns how to combine the different textual annotations coming from each relevant region into a single annotation that describes the whole system. Accordingly, given a set of input or test data (datacenter state modeled a context graph that is not annotated), the system determines which regions of the input graph are more relevant, and for each of these regions, the RNN predicts an annotation even in a previously unseen or different datacenter infrastructure.

Abstract: Incomplete state information for nodes of a datacenter is completed utilizing historical state information. A context graph is received having a plurality of nodes that correspond to components of the datacenter, each node including properties that correspond to the represented component. It is determined that at least one of the properties for a node is incomplete. A context hash is derived from the context graph and compared to a plurality of subgraph hashes. Based upon the comparison, one or more subgraphs are identified that are similar to the context hash. Where a single similar subgraph is identified, the property information for the single subgraph is imputed to the context graph to complete the incomplete information. Where more than one similar subgraph is identified, the property information for the identified similar subgraphs is aggregated and the aggregated property information is imputed to the context graph to complete the incomplete information.

Abstract: Incomplete state information for nodes of a datacenter is completed utilizing historical state information. A context graph is received having a plurality of nodes that correspond to components of the datacenter, each node including properties that correspond to the represented component. It is determined that at least one of the properties for a node is incomplete. A context hash is derived from the context graph and compared to a plurality of subgraph hashes. Based upon the comparison, one or more subgraphs are identified that are similar to the context hash. Where a single similar subgraph is identified, the property information for the single subgraph is imputed to the context graph to complete the incomplete information. Where more than one similar subgraph is identified, the property information for the identified similar subgraphs is aggregated and the aggregated property information is imputed to the context graph to complete the incomplete information.

Abstract: In a datacenter setting, root causes of anomalies corresponding to components in the datacenter are predicted. Initially, a convolutional neural network (CNN) is utilized to consider the evolution sequence of the datacenter infrastructure. Given a set of training data (sequences of datacenter states that are labeled with root causes of the anomalies present in the sequences), the CNN learns which sequences of datacenter states correspond to the labels of root causes. Accordingly, given a set of input or test data (sequences of datacenter states that are not labeled with root causes of the anomalies present in the sequences), the CNN is able to predict a root cause for the anomaly even in a previously unseen or different datacenter infrastructure.

Abstract: In a datacenter setting, annotations or descriptions of relevant parts or subgraphs corresponding to components in the datacenter are predicted. Given a set of training data (library of subgraphs seen in the past labeled with a textual description explaining why were they considered relevant enough to be placed in the historical database), the recurrent neural network (RNN) learns how to combine the different textual annotations coming from each relevant region into a single annotation that describes the whole system. Accordingly, given a set of input or test data (datacenter state modeled a context graph that is not annotated), the system determines which regions of the input graph are more relevant, and for each of these regions, the RNN predicts an annotation even in a previously unseen or different datacenter infrastructure.

Abstract: In a datacenter setting, future states of nodes of a datacenter (each node representing a component of the datacenter in a context graph) are predicted. Initially, historical metrics collected from the datacenter nodes, as well as historical metrics of neighboring nodes. The metrics are aggregated into historical metric summary vector representations of the nodes which are utilized to train a future state predictor to predict future datacenter states. Once trained, metrics may be input into the future state predictor and the future state predictor may be utilized to predict a future state of one or more of the nodes of the datacenter.

Abstract: In a datacenter setting, a summary of differences and similarities between two or more states of the same or similar systems are predicted. Initially, a Long Short-Term Memory (LSTM) neural network is trained with to predict a summary describing the state change between at least two states of the datacenter. Given a set of training data (at least two datacenter states that are annotated with a state change description), the LSTM neural network learns which similarities and differences between the datacenter states correspond to the annotations. Accordingly, given a set of test data comprising at least two states of a datacenter represented by context graphs that indicate a plurality of relationships among a plurality of nodes corresponding to components of a datacenter, the LSTM neural network is able to determine a state change description that summarizes the differences and similarities between the at least two states of the datacenter.