DETAILED ACTION
This office action is in response to amendments filed on 12/09/2025.
Claims 1-6, 8-10, 12, 14, and 16 have been amended. Claims 1-16 are pending.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Claim Objections:
In light of applicant’s amendments to the claims (pg. 2-6), the objections to the claims have been withdrawn.
Rejections under 35 U.S.C. § 112:
In light of applicant’s amendments to the claims (pg. 2-6), the rejections under 35 U.S.C. § 112 have been withdrawn.
Rejections under 35 U.S.C. § 101:
Applicant's arguments regarding the rejections under 35 U.S.C. § 101 (pg. 8-11) have been fully considered but they are not persuasive.
Applicant argues in regard to Step 2A of the Alice/Mayo test that the claimed invention is not directed to an abstract idea because it is directed to an improvement in the technical field of root cause analysis, as it reduces the computational burden of GNN-based causal inference by dividing low level nodes into groups and only evaluating causal relations between nodes within the groups, rather than among all nodes at the same time. Examiner acknowledges that this improvement is present in the specification (see e.g. paragraphs 0043-0050), but respectfully notes that the determination of patent eligibility due to an improvement to technology requires that “the claim must include the components or steps of the invention that provide the improvement described in the specification” (MPEP 2106.05(a)). While amended independent claims 1, 10, and 14 recite a causal graph including groups of lower level nodes, they do not require connections between lower level nodes to be limited to connections within a group. Therefore, the asserted improvement of reducing the number of evaluated causal relations, and thus reducing the computational burden, is not adequately claimed.
Applicant additionally argues in regard to Step 2B of the Alice/Mayo test that the claimed combination of features, specifically the grouping of nodes for the purpose of reducing the causal search space in a distributed computing environment, is not well-understood, routine, and conventional in the art. Examiner respectfully notes that when determining whether a claim recites significantly more than a judicial exception, the quality of being well-understood, routine, and conventional is a consideration applied to the claim’s additional elements, not the judicial exception itself (MPEP 2106.05(d)). As can be seen in the rejection under 35 U.S.C. § 101 below, the grouping of nodes in a causal graph can be considered a mental process, and the additional elements amount to mere instructions to implement the abstract idea on a computer, generally linking the use of the abstract idea to a particular field of use, and insignificant extra-solution activity which is well-understood, routine, and conventional in the art.
Applicant finally argues that the claimed invention is inherently computer based and does not have a parallel outside a computer environment, and thus cannot be directed to an abstract idea or mental process. Examiner respectfully notes that, as can be seen in the rejection under 35 U.S.C. § 101 below, the claims are directed to processes which can practically be performed in the human mind or with the aid of pen and paper. Per MPEP 2106.04(a)(2)(III), “[c]laims can recite a mental process even if they are claimed as being performed on a computer.”
The rejections of claims 1-5 and 10-15 under 35 U.S.C. § 101 are maintained, and have been updated to include the amended limitations and to clarify the reasoning given for the limitations that were not amended.
Prior Art Rejections:
Applicant's arguments regarding the prior art rejections (pg. 12-19) have been fully considered but they are not persuasive.
In regard to independent claim 10, applicant argues (pg. 12) that Meng fails to teach the limitation directed to “propagating the system failure over a learned causal graph for all of a plurality of nodes, the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups”. Examiner notes that the rejection of claims 10-11 under 35 U.S.C. § 102 has been withdrawn and replaced by a rejection under 35 U.S.C. § 103. As can be seen below in the rejection under 35 U.S.C. § 103, this amended limitation is taught by the combination of Meng and Chen.
In regard to independent claim 10, applicant argues (pg. 12) that Meng teaches away from the causal graph being divided into parts or groups. However, the causal graph referenced by Meng (figure 8) is ineffective because the two parts of the graph are completely disconnected, while the claimed causal graph consists of groups of nodes which are still connected by at least higher level connections. Since the structure that Meng may teach away from is different from the claimed graph structure, Meng does not teach away from the claimed subject matter.
In regard to independent claim 10, applicant argues (pg. 12-13) that Meng does not mention the use or implementation of a GNN. However, examiner respectfully notes that a GNN is not present in this claim, and in claims that do involve a GNN, the GNN is taught by Xu.
In regard to independent claim 1, applicant argues (pg. 13-19) that the amended limitation directed to “using a time series generated by each of a plurality of nodes to train a graph neural network to generate a causal graph for all of the plurality of nodes, the causal graph representing a group wherein a group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups” is not taught or suggested by Xu, Chen, or any of the other cited references. Examiner notes that, as can be seen below in the rejection under 35 U.S.C. § 103, this limitation is taught by the combination of Xu and Chen.
In regard to independent claim 1, applicant appears to argue (pg. 14) that Xu does not teach the above limitation because Xu’s causal graph is fully connected and does not include groups of nodes. Examiner respectfully disagrees that the graph shown in Xu’s figure 1 is fully connected between all the nodes; there are many nodes between which no connection is shown. Further, examiner respectfully notes that nothing in the amended claim restricts the causal graph from being fully connected between all nodes anyway. Finally, examiner notes that, as can be seen below in the rejection under 35 U.S.C. § 103, the groups of nodes are taught by Chen.
In regarding to independent claim 1, applicant argues (pg. 14-15) that Chen does not teach the above limitation because Chen’s causal graphs correspond to physical structures while the claimed causal graphs correspond to microservices. However, examiner respectfully notes that the physical embodiment disclosed by Chen mirrors the microservice architecture which is claimed and further described in specification paragraph 0018 of the instant application: “In a microservice system, for example, the lower level can be the level of physical pods running microservices, the higher level can be the level of network servers containing such pods, and the system level represents the performance of the whole group of servers.” Chen’s causal graph, as shown in figure 2, includes a set of servers running services (i.e. a higher level of network servers) represented by large dashed circles, with each service comprising a group of local nodes, each associated with performance metrics (i.e. a lower level running microservices), represented by small solid circles.
The prior art rejections have been updated to include the amended limitations and to clarify the reasoning given for the limitations that were not amended.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-5 and 10-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claim 1:
Step 1: The claim is directed to a method, which falls within the statutory category of a process.
Step 2A Prong 1: The claim is directed to an abstract idea. Specifically, the claim recites:
using a time series generated by each of a plurality of nodes [to train a graph neural network] to generate a causal graph for all of the plurality of nodes, the causal graph representing a group wherein a group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups; (Abstract idea – mental process. Generating a causal graph representing groups of nodes based on time series data can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the data on a display, mentally noting time-lagged correlations between variables indicating causal relationships, and drawing a graph on paper where nodes represent pods running microservices, connections between nodes represent causal relationships, and the nodes are arranged in groups. See MPEP 2106.04(a)(2)(III).)
identifying interdependent causal networks that depict hierarchical causal links from low-level nodes to high-level nodes to a system key performance indicator (KPI); (Abstract idea – mental process. Identifying interdependent networks with hierarchical causal links can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing KPI and metric data on a display and noting time-lagged correlations between entities of different levels, indicating hierarchical causal relationships. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2: The additional elements recited in the claim do not integrate the abstract idea into a practical application, individually or in combination. Specifically, the claim recites the additional elements:
train a graph neural network (Training a generic graph neural network is standard in the field of machine learning, and thus this limitation amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
simulating causal relations between entities by aggregating embeddings from neighbors in each layer; and (Adding insignificant extra-solution activity to the judicial exception – see MPEP2106.05(g).)
generating output embeddings for entity metrics prediction and between-level aggregation. (Generating output embeddings via a machine learning model is standard in the field of machine learning, and thus this limitation amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Specifically, the claim recites the additional elements:
train a graph neural network (Training a generic graph neural network is standard in the field of machine learning, and thus this limitation amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
simulating causal relations between entities by aggregating embeddings from neighbors in each layer; and (Adding insignificant extra-solution activity to the judicial exception – see MPEP2106.05(g). Further, the limitation is directed to aggregating embeddings from neighbors in a graph neural network, which is well-understood, routine, and conventional per Liu: “Graph neural networks have shown significant success in the field of graph representation learning. Graph convolutions perform neighborhood aggregation and represent one of the most important graph operations” (Liu et al., “Towards Deeper Graph Neural Networks”, Pg. 338, Abstract). See MPEP 2106.05(d)(II).)
generating output embeddings for entity metrics prediction and between-level aggregation. (Generating output embeddings via a machine learning model is standard in the field of machine learning, and thus this limitation amounts to adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
Claims 2-5:
Claim 2 recites The method as recited in claim 1, wherein the entity metrics data includes CPU utilization, memory usage, system KPI data, and combinations thereof. This limitation merely qualifies the data output by the model, and thus amounts to generally linking the use of a judicial exception to a particular technological environment or field of use - see MPEP 2106.05(h). Therefore, the claim does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 3 recites The method as recited in claim 2, wherein the KPI is a latency time, a connection time, or a combination thereof. This limitation merely qualifies the data output by the model, and thus amounts to generally linking the use of a judicial exception to a particular technological environment or field of use - see MPEP 2106.05(h). Therefore, the claim does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 4 recites The method as recited in claim 3, further comprising collecting the time series from each of the plurality of nodes by monitoring system components of each of the plurality of nodes. Monitoring system components to collect time series data amounts to insignificant extra-solution activity (necessary data gathering) (see MPEP2106.05(g)), and is well-understood, routine, and conventional in the field per Jensen: “The increase in deployment of sensors for monitoring large industrial systems and the ability to analyze the collected data efficiently provide the means for automation and remote management to be utilized at an unprecedented scale” (Jensen et al., “Time Series Management Systems: A Survey”, pg. 2581, section 1). Therefore, the claim does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 5 recites The method as recited in claim 4, wherein a causal structure learning process for the interdependent causal networks is divided into intra-level learning and inter-level learning. This limitation merely qualifies the abstract idea recited in claim 1 by dividing causal graph generation (mental process) into two stages of identifying intra-level and inter-level causal relationships, and does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 10:
Step 1: The claim is directed to a method, which falls within the statutory category of a process.
Step 2A Prong 1: The claim is directed to an abstract idea. Specifically, the claim recites:
detecting a system failure; (Abstract idea – mental process. Detecting a system failure can practically be performed in the human mind or with the aid of pen and paper, for example, by observing a defect in the system’s performance. See MPEP 2106.04(a)(2)(III).)
conducting topological cause learning by extracting causal relations from entity metrics data and system key performance indicator (KPI) data; (Abstract idea – mental process. Extracting causal relations between metrics and KPI data can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the data on a display and mentally noting time-lagged correlations between variables indicating causal relationships. See MPEP 2106.04(a)(2)(III).)
propagating the system failure over a learned causal graph for all of a plurality of nodes, the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups; (Abstract idea – mental process. Propagating a failure over a causal graph including groups of nodes representing microservice pods can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the causal graph on a display and mentally tracing causal relationships from the point of failure. See MPEP 2106.04(a)(2)(III).)
generating a topological cause score representing how much a component can be the root cause; (Abstract idea – mental process. Generating a topological cause score can practically be performed in the human mind or with the aid of pen and paper, for example, by counting the number of times a component is visited during propagation, and assigning a score to the component based on the count. See MPEP 2106.04(a)(2)(III).)
generating an individual cause score based on entity metrics using extreme value theory; (Abstract idea – mental process. Generating an individual cause score using extreme value theory can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally computing an extreme value threshold and calculating a cause score based on an entity metric’s relation to the threshold. See MPEP 2106.04(a)(2)(III).)
detecting anomalous entities based on performance of individual components; (Abstract idea – mental process. Detecting anomalous entities based on performance of individual components can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing metric data associated with the components on a display and mentally identifying outliers/anomalies. See MPEP 2106.04(a)(2)(III).)
aggregating the topological cause score and individual cause score to obtain a root cause ranking to discover the most probable root causes; and (Abstract idea – mental process. Aggregating topological score and individual score to obtain a root cause ranking can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally adding the two scores together for each entity and ranking the entities by combined cause score. See MPEP 2106.04(a)(2)(III).)
identifying a top K system entities associated with the most probable root causes. (Abstract idea – mental process. Identifying the top K root cause entities can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally identifying the K highest combined cause scores and their associated entities. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2: The claim does not recite additional elements which integrate the abstract idea into a practical application, individually or in combination.
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claims 11-13:
Claim 11 recites The method as recited in claim 10, wherein individual causes of the entity metrics are detected based on an extreme value theory. This limitation is directed to an abstract idea (mental process) because detecting individual causes of metrics based on EVT can practically be performed in the human mind or with the aid of pen and paper, for example, by identifying the component associated with a high individual cause score calculated based on EVT. See MPEP 2106.04(a)(2)(III). Therefore, the claim merges with the abstract idea recited in claim 10, and does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 12 recites The method as recited in claim 11, wherein abnormal values of the entity metrics are normalized using a Sigmoid function, and a mean value of the normalized values are used as the individual causal score of the associated system entity. This limitation is directed to an abstract idea (mathematical concept) because normalization using a Sigmoid function and computation of a mean value are both mathematical calculations. See MPEP 2106.04(a)(2)(I). Therefore, the claim merges with the abstract idea recited in claim 11, and does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 13 recites The method as recited in claim 12, wherein identifying most probable root causes does not require any domain/prior knowledge as input for root cause localization. This limitation merely qualifies the abstract idea by stating that the method does not require domain knowledge, and does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim 14:
Step 1: The claim is directed to a system, which falls within the statutory category of a machine/manufacture.
Step 2A Prong 1: The courts have recognized that claims can recite a mental process even if they are claimed as being performed on a computer. The claim is directed to an abstract idea. Specifically, the claim recites:
monitor system key performance indicator (KPI), detect a system failure, (Abstract idea – mental process. Monitoring a KPI and detecting a system failure can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing values of the KPI on a display and observing a defect in the system’s performance. See MPEP 2106.04(a)(2)(III).)
conduct topological cause learning by extracting causal relations from entity metrics data and system key performance indicator (KPI) data, (Abstract idea – mental process. Extracting causal relations between metrics and KPI data can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the data on a display and mentally noting time-lagged correlations between variables indicating causal relationships. See MPEP 2106.04(a)(2)(III).)
propagate the system failure over a learned causal graph for all of a plurality of nodes, the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system. wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups, and (Abstract idea – mental process. Propagating a failure over a causal graph including groups of nodes representing microservice pods can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the causal graph on a display and mentally tracing causal relationships from the point of failure. See MPEP 2106.04(a)(2)(III).)
generate a topological cause score representing how much a component can be the root cause; (Abstract idea – mental process. Generating a topological cause score can practically be performed in the human mind or with the aid of pen and paper, for example, by counting the number of times a component is visited during propagation, and assigning a score to the component based on the count. See MPEP 2106.04(a)(2)(III).)
generate an individual cause score based on the entity metrics using extreme value theory, and (Abstract idea – mental process. Generating an individual cause score using extreme value theory can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally computing an extreme value threshold and calculating a cause score based on an entity metric’s relation to the threshold. See MPEP 2106.04(a)(2)(III).)
detect anomalous entities based on performance of individual components; and (Abstract idea – mental process. Detecting anomalous entities based on performance of individual components can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing metric data associated with the components on a display and mentally identifying outliers/anomalies. See MPEP 2106.04(a)(2)(III).)
an integration tool configured to aggregate the topological causal score and the individual causal score to obtain a root cause ranking to discover the most probable root causes, (Abstract idea – mental process. Aggregating topological score and individual score to obtain a root cause ranking can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally adding the two scores together for each entity and ranking the entities by combined cause score. See MPEP 2106.04(a)(2)(III).)
identify a top K system entities associated with the most probable root causes, wherein identifying most probable root causes does not require any domain/prior knowledge as input for root cause localization, and (Abstract idea – mental process. Identifying the top K root cause entities can practically be performed in the human mind or with the aid of pen and paper, for example, by mentally identifying the K highest combined cause scores and their associated entities. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2: The additional elements recited in the claim do not integrate the abstract idea into a practical application, individually or in combination. Specifically, the claim recites the additional elements:
A system for identifying most probable root causes, comprising: one or more processors; a display screen coupled to the one or more processors through a bus; memory coupled to the one or more processors through the bus, wherein the memory includes a topological causal discover tool (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
an individual causal discovery tool configured to receive entity metrics, (Adding insignificant extra-solution activity (necessary data gathering) to the judicial exception – see MPEP2106.05(g).)
display the most probable root causes to a user on the display screen. (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
Step 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Specifically, the claim recites the additional elements:
A system for identifying most probable root causes, comprising: one or more processors; a display screen coupled to the one or more processors through a bus; memory coupled to the one or more processors through the bus, wherein the memory includes a topological causal discover tool (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
an individual causal discovery tool configured to receive entity metrics, (Adding insignificant extra-solution activity (necessary data gathering) to the judicial exception – see MPEP2106.05(g). Further, the limitation is directed to receiving or transmitting data over a network, which the courts have found to be well-understood, routine, and conventional in the computer arts. See MPEP 2106.05(d)(II).)
display the most probable root causes to a user on the display screen. (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea – see MPEP 2106.05(f).)
Claim 15:
Claim 15 recites The system as recited in claim 14, wherein a random walk with restart on interdependent causal networks is used to estimate the topological causal score of each node. This limitation is directed to an abstract idea (mental process) because performing a random walk with restart to estimate a causal score can practically be performed in the human mind or with the aid of pen and paper, for example, by viewing the causal graph on a display, mentally tracing causal links based on transfer probabilities, counting the number of times each node is visited, and using that count as the causal score for that node. See MPEP 2106.04(a)(2)(III). Therefore, the claim merges with the abstract idea recited in claim 14, and does not recite additional elements that are sufficient to amount to significantly more than the abstract idea.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-7 are rejected under 35 U.S.C. 103 as being unpatentable over
Xu et al. (hereinafter Xu), “Multivariate Time Series Forecasting with Transfer Entropy Graph” in view of
Chen et al. (hereinafter Chen), “CauseInfer: Automatic and Distributed Performance Diagnosis with Hierarchical Causality Graph in Large Distributed Systems”.
Regarding Claim 1,
Xu teaches A method for training a hierarchical graph neural network, comprising:
using a time series generated by each of a plurality of nodes to train a graph neural network to generate a causal graph for all of the plurality of nodes; (Pg. 2, section 1: “In this work, a novel framework, termed Graph Neural Network with Transfer Entropy (TEGNN) is proposed and applied for MTS [multivariate time series] forecasting tasks, which considers the causal relationships among variables. For the introduction of causality, The pairwise TE between variables is calculated, thus obtain the TE matrix, which is regarded as the adjacency matrix of the graph structure and each variable is one node of this graph.” Transfer entropy is a measure of causality, and the generated transfer entropy graph is thus a causal graph.)
simulating causal relations between entities by aggregating embeddings from neighbors in each layer; and (Pg. 5, section 3.4: “k-GNNs only perform information fusion between a certain node and its neighbors, ignoring the information of other non-neighbor nodes. In this way, for the prediction of a time series, only other series with significant causality are considered.” Performing information fusion between neighbors is aggregating embeddings.)
generating output embeddings for entity metrics prediction and between-level aggregation. (Pg. 5, section 3.4: “
h
i
(
l
)
is the hidden state of node
v
i
in the
l
t
h
layer…, the output dimension of the last graph neural network layer is 1, which is used as the prediction result.” Hidden state
h
i
(
l
)
is the output embedding for between-level aggregation, and the prediction result is the output embedding for metrics prediction.)
Xu does not appear to explicitly disclose
the causal graph representing a group wherein a group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups;
identifying interdependent causal networks that depict hierarchical causal links from low-level nodes to high-level nodes to a system key performance indicator (KPI);
However, Chen teaches the causal graph representing a group wherein a group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups; (Pg. 2, figure 2: “The basic structure and the work flow of CauseInfer… the top is the abstracted service and metric causality graph. In the causality graph, the big dashed circle denotes service, the red node denotes the root cause, the black node denotes performance metric, the green node denotes SLO metric, the arc denotes the causality or dependency relationship and the arrow denotes the direction of anomaly propagation.” The causality graph, as depicted in the top of figure 2, includes groups of connected nodes surrounded by dashed circles representing microservices (i.e. the lower level), as well as connections between the set of groups (i.e. the higher level).)
identifying interdependent causal networks that depict hierarchical causal links from low-level nodes to high-level nodes to a system key performance indicator (KPI); (Pg. 1, section 1: “CauseInfer automatically constructs a two layered hierarchical causality graph: a coarse-grained graph with the purpose of locating the causes at service level and a fine-grained graph with the purpose of finding the real culprits of performance problems. Once an SLO (Service Level Objective) violation in the front end servers occurs, the inference procedure is triggered. We first locate the performance anomaly at specific service(s) (e.g. tomcat) by detecting the violations of SLO metric then find out the root cause(s) by detecting the violations of other performance metrics in a local node.” The local nodes located by the fine-grained graphs are low level nodes representing root causes, the services located by the coarse-grained graph are high level nodes representing interdependent networks, the Service Level Objective (SLO) is a key performance indicator (KPI), and the connections of the hierarchical causality graph are hierarchical causal links.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Xu and Chen. Xu teaches a method for time-series forecasting using a graph neural network to model causal relationships. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. One of ordinary skill would have motivation to combine Xu and Chen because Xu’s TEGNN model captures accurate causal graphs: “With in-depth theoretical analysis and experimental verification, we confirm that TEGNN successfully captures the causal relationship among variables and uses graph neural network to select key variables for accurate forecasting” (Xu, pg. 9, section 5), and Chen’s CauseInfer system uses causal graphs for root cause localization to “achieve a high precision and recall for performance diagnosis and scale up readily in large distributed systems” (Chen, pg. 9, section VI).
Regarding Claim 2, Xu and Chen teach The method as recited in claim 1, as shown above.
Chen also teaches wherein the entity metrics data includes CPU utilization, memory usage, system KPI data, and combinations thereof. (Pg. 6, section IV.B: “Figure 7 demonstrates part of the causality graph built for Httpd service,” where each node represents a metric, including “CPU_Totl” (i.e. CPU utilization), “MEM_Used” (i.e. memory usage), and “TCP_LATENCY”, which is a unified service level objective (i.e. KPI): “we propose a new unified SLO metric, tcp request latency (abbreviated as TCP LATENCY)” (pg. 3, section III.A).)
Regarding Claim 3, Xu and Chen teach The method as recited in claim 2, as shown above.
Chen also teaches wherein the KPI is a latency time, a connection time, or a combination thereof. (Pg. 3, section III.A: “we propose a new unified SLO metric, tcp request latency (abbreviated as TCP LATENCY).”)
Regarding Claim 4, Xu and Chen teach The method as recited in claim 3, as shown above.
Chen also teaches further comprising collecting the time series from each of the plurality of nodes by monitoring system components of each of the plurality of nodes. (Pg. 3, section III.A: “The data collection module collects high dimensional runtime information from multiple data sources across different software stacks covering application, process and operating system.”)
Regarding Claim 5, Xu and Chen teach The method as recited in claim 4, as shown above.
Chen also teaches wherein a causal structure learning process for the interdependent causal networks is divided into intra-level learning and inter-level learning. (Pg. 1, section 1: “CauseInfer automatically constructs a two layered hierarchical causality graph: a coarse-grained graph with the purpose of locating the causes at service level and a fine-grained graph with the purpose of finding the real culprits of performance problems.” Constructing the fine-grained graph within a service is intra-level learning, and constructing the coarse-grained graph between services is inter-level learning.)
Regarding Claim 6, Xu and Chen teach The method as recited in claim 5, as shown above.
Chen teaches wherein information of low-level nodes to the high-level nodes is aggregated for constructing a cross-level causal relations, so an initial embedding of high level nodes,
z
¨
(
0
)
, is a concatenation of their time-lagged data
x
¨
t
-
1
,
…
,
x
¨
t
-
p
and aggregated low-level embeddings, which can be formulated as
z
¨
(
0
)
=
C
a
t
x
¨
t
-
1
,
…
,
x
¨
t
-
p
,
W
¨
∙
z
L
; (Pg. 5, section III.C: “For a service without any dependent service such as database service, the causality graph is built using only the local performance metrics mentioned in data collection section. But for the one with some dependent services such as web service, the causality graph is built using not only the local performance metrics but also the TCP_LATENCY metrics of its dependent services.” A service with dependent services is a high-level node, and its graph is built using input (i.e. initial embedding) comprised of local performance metrics (i.e. its own time lagged data) as well as metrics from the nodes of dependent services (i.e. aggregated low-level embeddings).)
Xu teaches where
W
¨
is a weight matrix that controls contributions of low-level embeddings to high-level embeddings. (Pg. 5, section 3.4: “we propose TEGNN model and use the following propagation model for calculating the forward-pass update of a node denoted by
v
i
:
h
i
(
l
+
1
)
=
σ
h
i
(
l
)
W
1
(
l
)
+
∑
j
∈
N
(
i
)
h
j
(
l
)
W
2
(
l
)
, where
W
1
(
l
)
and
W
2
(
l
)
are parameter matrices,
h
i
(
l
)
is the hidden state of node
v
i
in the
l
t
h
layer and
N
(
i
)
denotes the neighbors of node
i
.” The hidden state of layer
l
+
1
(i.e. high-level embeddings) is computed based on the hidden state of layer
l
(i.e. low-level embeddings) and the parameter matrices (i.e. weight matrices).)
Regarding Claim 7, Xu and Chen teach The method as recited in claim 6, as shown above.
Chen also teaches wherein learned interdependent causal graphs meet an acyclicity requirement. (Pg. 3, section III.C: “And in this causal relationship, it’s not allowed two variables cause each other. So finally, all the causal relationships can be encoded by a DAG (Directed Acyclic Graph)”)
Claims 8 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Chen and further in view of
Meng et al. (hereinafter Meng), “Localizing Failure Root Causes in a Microservice through Causality Inference” and
Pan et al. (hereinafter Pan), “Automatic Multimedia Cross-modal Correlation Discovery”.
Regarding Claim 8, Xu and Chen teach The method as recited in claim 7, as shown above.
Xu and Chen do not appear to explicitly disclose wherein a random walk with restart on interdependent causal networks is used to estimate a topological causal score of each of the plurality of nodes.
However, Meng teaches wherein a random walk [with restart] on interdependent causal networks is used to estimate a topological causal score of each of the plurality of nodes. (Pg. 6-7, section IV.C: “In this step, we propose cause oriented random walk to find the possible root cause via the causal relationship between the metrics and KPIs… After Nrw steps, the walker stops and each node is visited
c
i
times…
c
-
i
is the normalized visit time
c
i
.” Normalized visit time is a topological cause score representing each node’s causal relationship with the anomalous KPI.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Xu, Chen, and Meng. Xu teaches a method for time-series forecasting using a graph neural network to model causal relationships. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Meng teaches a method for root cause localization in microservices architecture via cause oriented random walk over a causal graph. One of ordinary skill would have motivation to combine Xu, Chen, and Meng because, according to Meng, “Random walk has been widely used for root cause localization” (Meng, pg. 4, section III.A), however “Current random walk algorithm, which is based on the assumption that, an abnormal metric more correlated with an anomalous KPI is more likely to be the root cause, is not always true in our scenario” (Meng, pg. 2, section I). Meng’s MicroCause framework and temporal cause oriented random walk (TCORW) “achieves the best performance for the task of failure root cause localization” when compared with baseline random walk algorithms (Meng, pg. 9, section V.D). “Quick failure root cause localization can improve quality of service and reduce loss in efficiency and revenue” (Meng, pg. 10, section VII).
Pan teaches random walk with restart (Pg. 3, section 3: “The “random walk with restarts” operates as follows: to compute the affinity of node ‘B’ for node ‘A’, consider a random walker that starts from node ‘A’. The random walker chooses randomly among the available edges every time, except that, before he makes a choice, with probability c, he goes back to node ‘A’ (restart).”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Xu, Chen, Meng, and Pan. Xu teaches a method for time-series forecasting using a graph neural network to model causal relationships. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Meng teaches a method for root cause localization in microservices architecture via cause oriented random walk over a causal graph. Pan teaches a graph based approach to discovering cross-modal correlations via random walk with restart. One of ordinary skill would have motivation to combine Xu, Chen, Meng, and Pan because, as can be seen in Pan’s figure 5(a), increasing restart probability c in a random walk increases accuracy (Pan, pg. 5, figure 5).
Regarding Claim 9, Xu, Chen, Meng, and Pan teach The method as recited in claim 8, as shown above.
Meng also teaches wherein transition probabilities of a particle on the interdependent causal networks is calculated as:
H
=
H
G
G
H
G
A
H
A
G
H
A
A
where HGG and HAA depict the random walk within a same-level network, and HGA and HAG describe the random walk across different level networks. (Pg. 6, section IV.C: “We calculate the matrix Q in random walk as follows…” Equation (3) shows calculation of the probability
Q
i
j
of a “Forward step (walk from result indicator to cause indicator)”, equation (4) shows calculation of the probability
Q
j
i
of a “Backward step (walk from cause indicator to result indicator)”, and equation (5) shows calculation of the probability
Q
i
i
of a “Self step (stay in the present node)”. Forward and backward steps represent walks across different level networks, and self steps represent walks within the same-level network.)
Claims 10-11 are rejected under 35 U.S.C. 103 as being unpatentable over Meng in view of Chen.
Regarding Claim 10,
Meng teaches A method for identifying most probable root causes, comprising:
detecting a system failure; (Pg. 3, section II.B: “When a KPI of a microservice becomes anomalous, the microservice will be deemed to be failed.”)
conducting topological cause learning by extracting causal relations from entity metrics data and system key performance indicator (KPI) data; (Pg. 5, section IV.A: “In improved PC algorithm, given a failure case, such as failure case X, the monitoring indicators of microservice A dataset
I
t
i
, t = 0, …, T, i = 1,…, N with N time series, including metrics and KPIs, will be used as input… the result of the improved PC is a causal graph…” Generating a causal graph based on metrics and KPIs amounts to extracting causal relations from the metric and KPI data.)
propagating the system failure over a learned causal graph for all of a plurality of nodes, (Pg. 6, section IV.C: “In this step, we propose cause oriented random walk to find the possible root cause via the causal relationship between the metrics and KPIs.” Random walk over the failure causal graph, oriented by causal correlation, amounts to propagating the system failure over the causal graph.)
generating a topological cause score representing how much a component can be the root cause; (Pg. 6-7, section IV.C: “After Nrw steps, the walker stops and each node is visited
c
i
times…
c
-
i
is the normalized visit time
c
i
.” Normalized visit time is a topological cause score representing the node’s causal relationship with the anomalous KPI.)
generating an individual cause score based on entity metrics using extreme value theory; (Pg. 6, section IV.B: “SPOT detects the sudden change in time series via the extreme value theory… In addition to detecting the anomalies of indicators, we also evaluate the anomaly degree of metrics based on the SPOT result in this module. Here we define the anomaly degree of the metric i as
η
m
a
x
i
.” Anomaly degree
η
m
a
x
i
is an individual cause score based on entity metric i, which is calculated based on the threshold generated by SPOT [Streaming Peaks-Over-Threshold] using the extreme value theory.)
detecting anomalous entities based on performance of individual components; (Pg. 6, section IV.B: “In MicroCause, we assume that the root cause metrics should become anomalous in some time before failure time. We adopt the SPOT [18], to detect anomaly of the metrics.”)
aggregating the topological cause score and individual cause score to obtain a root cause ranking to discover the most probable root causes; and (Pg. 6, section IV.C: “Beside the casual relationship with the anomalous KPI, we also take the anomaly degree
η
m
a
x
i
of the metrics into the consideration to localize the root cause… Here we define the potential root cause score of metric i as
γ
i
. It is calculated as:
γ
i
=
λ
c
-
i
+
(
1
-
λ
)
η
-
m
a
x
i
.”
c
-
i
is the normalized visit time (i.e. the topological cause score),
η
-
m
a
x
i
is the normalized anomaly degree (i.e. the individual cause score), and the two are aggregated to determine root cause score
γ
i
, which is then used to rank root causes: “we design an algorithm, which combines the potential root cause score of the metric, the priority of the metric and the anomaly time of the metric to rank all the potential root cause of the failure ticket, as shown in Algorithm 1.”)
identifying a top K system entities associated with the most probable root causes. (Pg. 5, section III.B: “…the temporal cause oriented random walk (denoted as TCORW) will give the top N potential root cause of the failure based on the failure causal graph and the anomaly information of the metrics.”)
Meng does not appear to explicitly disclose the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups;
However, Chen teaches the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups; (Pg. 2, figure 2: “The basic structure and the work flow of CauseInfer… the top is the abstracted service and metric causality graph. In the causality graph, the big dashed circle denotes service, the red node denotes the root cause, the black node denotes performance metric, the green node denotes SLO metric, the arc denotes the causality or dependency relationship and the arrow denotes the direction of anomaly propagation.” The causality graph, as depicted in the top of figure 2, includes groups of connected nodes surrounded by dashed circles representing microservices (i.e. the lower level), as well as connections between the set of groups (i.e. the higher level).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng and Chen. Meng teaches a method for root cause localization in microservices architecture via cause oriented random walk over a causal graph. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. One of ordinary skill would have motivation to combine Meng and Chen because Chen’s hierarchical causality analysis “largely reduces the data exchange between hosts during performance diagnosis” (Chen, pg. 2, section I) while “achiev[ing] a high precision and recall for performance diagnosis and scal[ing] up readily in large distributed systems” (Chen, pg. 9, section VI).”
Regarding Claim 11, Meng and Chen teach The method as recited in claim 10, as shown above.
Meng also teaches wherein individual causes of the entity metrics are detected based on an extreme value theory. (Pg. 6, section IV.B: “In MicroCause, we assume that the root cause metrics should become anomalous in some time before failure time. We adopt the SPOT [18], to detect anomaly of the metrics. Because SPOT detects the sudden change in time series via the extreme value theory.”)
Claims 12-13 are rejected under 35 U.S.C. 103 as being unpatentable over Meng in view of Chen and further in view of
Wu et al. (hereinafter Wu), “Developing an Unsupervised Real-Time Anomaly Detection Scheme for Time Series With Multi-Seasonality” and
Trentini et al. (hereinafter Trentini), “Model-centered Ensemble for Anomaly Detection in Time Series”.
Regarding Claim 12, Meng and Chen teach The method as recited in claim 11, as shown above.
Meng and Chen do not appear to explicitly disclose wherein abnormal values of the entity metrics are normalized using a Sigmoid function,
However, Wu teaches wherein abnormal values of the entity metrics are normalized using a Sigmoid function, (Pg. 4148, section 1: “We also propose a method to map the LTI [Local Trend Inconsistency] value of a frame to its Anomaly Score (AS) by a logistic-shaped function.” Local Trend Inconsistency values represent anomalies in the metrics, and they are normalized by a logistic-shaped (i.e. sigmoid) function.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng, Chen, and Wu. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Wu teaches an anomaly detection scheme in which anomalous time series data is normalized via sigmoid function to produce an anomaly score for each frame. One of ordinary skill would have motivation to combine Meng, Chen, and Wu because mapping to a normalized anomaly score represents the probability of anomaly, independent of the specific application/metric (Wu, pg. 4153, section 4.3). Further, Wu’s anomaly detection scheme “outperformed several representative anomaly detection schemes commonly used in practice” (Wu, pg. 4158, section 6).
Meng, Chen, and Wu do not appear to explicitly disclose a mean value of the normalized values are used as the individual causal score of the associated system entity.
However, Trentini teaches a mean value of the normalized values are used as the individual causal score of the associated system entity. (Pg. 702, section 4: “To address the first issue, a damping function is applied to the anomaly scores, in order to prevent it from being dominated by a few components… The second issue is addressed in this paper by using a weighted average on the damped scores…” Pg. 705, section 6.3: “We also evaluate the combination of these techniques using the following ensemble strategies: Simple average Ensemble (SA), Damped Average Ensemble (DA), and Simple Weighted Average Ensemble (SWA).” In Damped Average Ensemble, anomaly scores, which have been processed by a damping function (i.e. normalized), are aggregated by a simple average (i.e. mean).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng, Chen, Wu, and Trentini. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Wu teaches an anomaly detection scheme in which anomalous time series data is normalized via sigmoid function to produce an anomaly score for each frame. Trentini teaches a time series anomaly detection method in which various anomaly scores are averaged to produce a final anomaly score. One of ordinary skill would have motivation to combine Meng, Chen, Wu, and Trentini in order to best aggregate anomaly scores identified at each time series frame into a single score: “Table 1 shows the precision, recall, F0.1 and F1 for all the baseline models and the ensemble models evaluated, trained and tested with Power Demand. As we can see, the Simple Average (SA) and Damped Average (DA) ensembles achieved the best results” (Trentini, pg. 706, section 6.3.1).
Regarding Claim 13, Meng, Chen, Wu, and Trentini teach The method as recited in claim 12, as shown above.
Chen also teaches wherein identifying most probable root causes does not require any domain/prior knowledge as input for root cause localization. (Pg. 5, section III.C: “we set up two methods: a conservative one and an aggressive one. By the word of aggressive, we mean it doesn’t use any prior knowledge to build the causality graph.”)
Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over Meng in view of Chen and
Lee, U.S. Patent Application Publication US 20160171414 A1.
Regarding Claim 14,
Meng teaches A system for identifying most probable root causes, comprising:
a topological causal discover tool configured to monitor system key performance indicator (KPI), detect a system failure, (Pg. 3, section II.B: “When a KPI of a microservice becomes anomalous, the microservice will be deemed to be failed.”)
conduct topological cause learning by extracting causal relations from entity metrics data and system key performance indicator (KPI) data, (Pg. 5, section IV.A: “In improved PC algorithm, given a failure case, such as failure case X, the monitoring indicators of microservice A dataset
I
t
i
, t = 0, …, T, i = 1,…, N with N time series, including metrics and KPIs, will be used as input… the result of the improved PC is a causal graph…” Generating a causal graph based on metrics and KPIs amounts to extracting causal relations from the metric and KPI data.)
propagate the system failure over a learned causal graph for all of a plurality of nodes, (Pg. 6, section IV.C: “In this step, we propose cause oriented random walk to find the possible root cause via the causal relationship between the metrics and KPIs.” Random walk over the failure causal graph, oriented by causal correlation, amounts to propagating the system failure over the causal graph.)
generate a topological cause score representing how much a component can be the root cause; (Pg. 6-7, section IV.C: “After Nrw steps, the walker stops and each node is visited
c
i
times…
c
-
i
is the normalized visit time
c
i
.” Normalized visit time is a topological cause score representing the node’s causal relationship with the anomalous KPI.)
an individual causal discovery tool configured to receive entity metrics, (Pg. 5, section III.B: “In the meantime, the metrics in the input dataset will be checked whether there are anomalies in anomaly detection module.” The anomaly detection module is an individual causal discovery tool.)
generate an individual cause score based on the entity metrics using extreme value theory, and (Pg. 6, section IV.B: “SPOT detects the sudden change in time series via the extreme value theory… In addition to detecting the anomalies of indicators, we also evaluate the anomaly degree of metrics based on the SPOT result in this module. Here we define the anomaly degree of the metric i as
η
m
a
x
i
.” Anomaly degree
η
m
a
x
i
is an individual cause score based on entity metric i, which is calculated based on the threshold generated by SPOT [Streaming Peaks-Over-Threshold] using the extreme value theory.)
detect anomalous entities based on performance of individual components; and (Pg. 6, section IV.B: “In MicroCause, we assume that the root cause metrics should become anomalous in some time before failure time. We adopt the SPOT [18], to detect anomaly of the metrics.”)
an integration tool configured to aggregate the topological causal score and the individual causal score to obtain a root cause ranking to discover the most probable root causes, (Pg. 6-7, section IV.C: “Beside the casual relationship with the anomalous KPI, we also take the anomaly degree
η
m
a
x
i
of the metrics into the consideration to localize the root cause… Here we define the potential root cause score of metric i as
γ
i
. It is calculated as:
γ
i
=
λ
c
-
i
+
(
1
-
λ
)
η
-
m
a
x
i
.”
c
-
i
is the normalized visit time (i.e. the topological cause score),
η
-
m
a
x
i
is the normalized anomaly degree (i.e. the individual cause score), and the two are aggregated to determine root cause score
γ
i
, which is then used to rank root causes: “we design an algorithm, which combines the potential root cause score of the metric, the priority of the metric and the anomaly time of the metric to rank all the potential root cause of the failure ticket, as shown in Algorithm 1.”)
identify a top K system entities associated with the most probable root causes, (Pg. 5, section III.B: “…the temporal cause oriented random walk (denoted as TCORW) will give the top N potential root cause of the failure based on the failure causal graph and the anomaly information of the metrics.”)
Meng does not appear to explicitly disclose
one or more processors;
a display screen coupled to the one or more processors through a bus;
memory coupled to the one or more processors through the bus, wherein the memory includes
display the most probable root causes to a user on the display screen.
However, Lee teaches A system for identifying most probable root causes, comprising:
one or more processors; ([0072]: “FIG. 3 is a block diagram 300 of an exemplary computer 302 used in the EDCS 100 according to an implementation.” [0078]: “The computer 302 includes a processor 305. Although illustrated as a single processor 305 in FIG. 3, two or more processors may be used according to particular needs, desires, or particular implementations of the computer 302 and/or the EDCS 100.”)
a display screen coupled to the one or more processors through a bus; ([0088]: “To provide for interaction with a user, implementations of the subject matter described in this specification can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube), LCD (liquid crystal display), LED (Light Emitting Diode), or plasma monitor, for displaying information to the user…”)
memory coupled to the one or more processors through the bus, ([0079]: “The computer 302 also includes a memory 306 that holds data for the computer 302 and/or other components of the EDCS 100.”)
display the most probable root causes to a user on the display screen. ([0070]: “At 234b, method 200 can initiate display of information related to the most significant root cause. Method 200 can also initiate display of information related to a set of most significant root causes.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng and Lee. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Lee teaches a computer system for identifying and displaying root causes of KPI violations. One of ordinary skill would have motivation to combine Meng and Lee because Lee’s system facilitates user interaction with the root cause analysis: “the system can identify the most significant problem(s) and root cause(s) and can provide a recommendation to address inefficiency issues. In this manner, a user can simply follow the recommendation to make an improvement” (Lee, 0013).
Meng and Lee do not appear to explicitly disclose
the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups,
wherein identifying most probable root causes does not require any domain/prior knowledge as input for root cause localization, and
However, Chen teaches the learned causal graph representing a group wherein the group includes a portion of the plurality of nodes at a lower level of a microservice system, wherein the lower level corresponds to pods running microservices and a higher level corresponds to a set of the groups, (Pg. 2, figure 2: “The basic structure and the work flow of CauseInfer… the top is the abstracted service and metric causality graph. In the causality graph, the big dashed circle denotes service, the red node denotes the root cause, the black node denotes performance metric, the green node denotes SLO metric, the arc denotes the causality or dependency relationship and the arrow denotes the direction of anomaly propagation.” The causality graph, as depicted in the top of figure 2, includes groups of connected nodes surrounded by dashed circles representing microservices (i.e. the lower level), as well as connections between the set of groups (i.e. the higher level).)
wherein identifying most probable root causes does not require any domain/prior knowledge as input for root cause localization, and (Pg. 5, section III.C: “we set up two methods: a conservative one and an aggressive one. By the word of aggressive, we mean it doesn’t use any prior knowledge to build the causality graph.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng, Lee, and Chen. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Lee teaches a computer system for identifying and displaying root causes of KPI violations. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. One of ordinary skill would have motivation to combine Meng, Lee, and Chen because Chen’s hierarchical causality analysis “largely reduces the data exchange between hosts during performance diagnosis” (Chen, pg. 2, section I) while “achiev[ing] a high precision and recall for performance diagnosis and scal[ing] up readily in large distributed systems” (Chen, pg. 9, section VI).”
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Meng in view of Lee and Chen and further in view of Pan.
Regarding Claim 15, Meng, Lee, and Chen teach The system as recited in claim 14, as shown above.
Meng also teaches wherein a random walk [with restart] on interdependent causal networks is used to estimate the topological causal score of each node. (Pg. 6-7, section IV.C: “In this step, we propose cause oriented random walk to find the possible root cause via the causal relationship between the metrics and KPIs… After Nrw steps, the walker stops and each node is visited
c
i
times…
c
-
i
is the normalized visit time
c
i
.” Normalized visit time is a topological cause score representing each node’s causal relationship with the anomalous KPI.)
Meng, Lee, and Chen do not appear to explicitly disclose random walk with restart
However, Pan teaches random walk with restart (Pg. 3, section 3: “The “random walk with restarts” operates as follows: to compute the affinity of node ‘B’ for node ‘A’, consider a random walker that starts from node ‘A’. The random walker chooses randomly among the available edges every time, except that, before he makes a choice, with probability c, he goes back to node ‘A’ (restart).”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng, Lee, Chen, and Pan. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Lee teaches a computer system for identifying and displaying root causes of KPI violations. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Pan teaches a graph based approach to discovering cross-modal correlations via random walk with restart. One of ordinary skill would have motivation to combine Meng, Lee, Chen, and Pan because, as can be seen in Pan’s figure 5(a), increasing restart probability c in a random walk increases accuracy (Pan, pg. 5, figure 5).
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Meng in view of Lee, Chen, and Pan, and further in view of Xu.
Regarding Claim 16, Meng, Lee, Chen, and Pan teach The system as recited in claim 15, as shown above.
Chen also teaches wherein information of low-level nodes to high-level nodes is aggregated for constructing a cross-level causal relations, so an initial embedding of high level nodes,
z
¨
(
0
)
, is a concatenation of their time-lagged data
x
¨
t
-
1
,
…
,
x
¨
t
-
p
and aggregated low-level embeddings, which can be formulated as
z
¨
(
0
)
=
C
a
t
x
¨
t
-
1
,
…
,
x
¨
t
-
p
,
W
¨
∙
z
L
; (Pg. 5, section III.C: “For a service without any dependent service such as database service, the causality graph is built using only the local performance metrics mentioned in data collection section. But for the one with some dependent services such as web service, the causality graph is built using not only the local performance metrics but also the TCP_LATENCY metrics of its dependent services.” A service with dependent services is a high-level node, and its graph is built using input (i.e. initial embedding) comprised of local performance metrics (i.e. its own time lagged data) as well as metrics from the nodes of dependent services (i.e. aggregated low-level embeddings).)
Meng, Lee, Chen, and Pan do not appear to explicitly disclose where
W
¨
is a weight matrix that controls contributions of low-level embeddings to high-level embeddings.
However, Xu teaches where
W
¨
is a weight matrix that controls contributions of low-level embeddings to high-level embeddings. (Pg. 5, section 3.4: “we propose TEGNN model and use the following propagation model for calculating the forward-pass update of a node denoted by
v
i
:
h
i
(
l
+
1
)
=
σ
h
i
(
l
)
W
1
(
l
)
+
∑
j
∈
N
(
i
)
h
j
(
l
)
W
2
(
l
)
, where
W
1
(
l
)
and
W
2
(
l
)
are parameter matrices,
h
i
(
l
)
is the hidden state of node
v
i
in the
l
t
h
layer and
N
(
i
)
denotes the neighbors of node
i
.” The hidden state of layer
l
+
1
(i.e. high-level embeddings) is computed based on the hidden state of layer
l
(i.e. low-level embeddings) and the parameter matrices (i.e. weight matrices).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Meng, Lee, Chen, Pan, and Xu. Meng teaches a method for root cause localization in microservices architecture by learning causal graphs and analyzing KPI/metric anomalies. Lee teaches a computer system for identifying and displaying root causes of KPI violations. Chen teaches a method for root cause localization in a distributed system using a hierarchical causality graph based on time-series data. Pan teaches a graph based approach to discovering cross-modal correlations via random walk with restart. Xu teaches a method for time-series forecasting using a graph neural network to model causal relationships. One of ordinary skill would have motivation to combine Meng, Lee, Chen, Pan, and Xu because Xu’s TEGNN model captures accurate causal graphs: “With in-depth theoretical analysis and experimental verification, we confirm that TEGNN successfully captures the causal relationship among variables and uses graph neural network to select key variables for accurate forecasting” (Xu, pg. 9, section 5).
Conclusion
Claim 1-16 are rejected.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Viker Lamardo can be reached at (571) 270-5871. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/B.M.R./Examiner, Art Unit 2147 /VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147