Automatic datacenter state summarization

• US 10,346,450 B2
• Filed: 12/28/2016
• Issued: 07/09/2019
• Est. Priority Date: 12/21/2016
• Status: Active Grant

• Current Assignee: CA, Inc. (d/b/a CA Technologies) (Broadcom, Inc.)
• Original Assignee: CA, Inc. (d/b/a CA Technologies) (Broadcom, Inc.)
• Inventors: Noguero, David Solans, Llagostera, Jaume Ferrarons, Segura, Alberto Huelamo, Mulero, Victor Muntes, Charles, David Sanchez, Simo, Marc Sole
• Primary Examiner(s): Ly, Anh
• Application Number:
  US15/392,577
• Publication Number:
  US 20180173687A1
• Time in Patent Office: 923 Days
• CPC Class Codes: G06F 16/334 Query execution G06F16/335 takes precedence
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence
• CPC Class Codes: H04L 41/147 for predicting network behaviour
• CPC Class Codes: H04L 43/0817 by checking functioning

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.

Claims

1. A method comprising:

receiving a context graph indicating a plurality of relationships among a plurality of nodes corresponding to components of a datacenter, each node comprising properties corresponding to a particular component of the datacenter;

identifying one or more subgraphs from the context graph, wherein each subgraph is a portion of the context graph that is determined to be a relevant region associated with the datacenter;

for each subgraph identified as a relevant region in the context graph, determining whether each subgraph identified as a relevant region corresponds to a relevance condition and lacks an annotation describing the relevance condition;

determining, for each subgraph identified as a relevant region, a plurality of context hashes based on selected properties, the context hashes generated by applying a hash function to each subgraph;

comparing the plurality of context hashes to a plurality of subgraph hashes derived from a library of subgraphs to determine a set of subgraph hashes that are similar to the plurality of context hashes derived for each relevant region;

identifying annotations for each subgraph identified as a relevant region by referencing corresponding subgraph hashes in the set of subgraph hashes;

combining the annotations in accordance with an importance into an annotation for each relevant region; and

annotating the context graph with the combined annotations, the combined annotations describing the datacenter as a whole with a textual description.

2. The method of claim 1, wherein the set of subgraph hashes include annotations provided by a user that describe relevance conditions corresponding to each subgraph hash of the set.

3. The media of claim 1, further comprising receiving properties from a historical database to build the context graph.

4. The method of claim 1, further comprising receiving properties from a most recent timestamp to build the context graph.

5. The method of claim 1, wherein the importance is based on a ranking function.

6. The method of claim 5, wherein the ranking function considers a centrality of the set of subgraph hashes to the relevance condition.

7. The method of claim 5, wherein the ranking function considers a frequency of a pattern in the set of subgraph hashes.

8. The method of claim 1, wherein combining the annotations comprises concatenating the annotations.

9. The method of claim 1, wherein combining the annotations comprises utilizing natural language processing techniques to generate the textual description.

10. The method of claim 1, wherein the combined annotations describes the property or the particular component that is causing the anomalous condition.

11. A method comprising:

receiving historical metrics from a plurality of historical nodes for relevance conditions seen by historical datacenters;

aggregating the historical metrics into historical vector representations for the plurality of historical nodes in the historical datacenter, the historical vector representations including information derived from the historical metrics of neighbors of the historical nodes and having annotations describing relevance conditions seen by the historical datacenters;

training a classifier based on the historical vector representations and annotations; and

predicting an annotation for a datacenter represented by a context graph indicating a plurality of relationships among a plurality of nodes corresponding to components of the datacenter, the context graph having a subgraph identified as a relevant region corresponding to a relevance condition not having an annotation describing the relevance condition with a textual description, wherein the classifier identifies similar subgraphs to one or more context subgraphs derived from the context graph and annotations corresponding to the similar subgraphs.

12. The method of claim 11, further comprising receiving the context graph.

13. The method of claim 12, further comprising, for the relevant region of the context graph corresponding to the relevance condition and lacking the annotation describing the relevance condition, determining a context vector representation based on selected properties.

14. The method of claim 13, further comprising comparing the context vector representation to historical vector representations to determine a set of historical vector representations that, when aggregated, are similar to the context vector representation.

15. The method of claim 14, further comprising identifying annotations corresponding to the set of historical vector representations.

16. The method of claim 15, further comprising combining the annotations in accordance with an importance that is based on a ranking function, the ranking function considering a centrality of the set of historical vector representations to the relevance condition or a frequency of a pattern in the set of historical vector representations.

17. The method of claim 16, further comprising annotating the node with the combined annotations, the combined annotations describing the relevance condition with a textual description.

18. The method of claim 17, wherein the combined annotations describe the property or the particular component that is causing the relevance condition.

19. The method of claim 11, wherein the historical datacenter corresponding to the historical metrics and the datacenter corresponding to the context graph are two different datacenters.

20. A computerized system comprising:

a processor; and

a non-transitory computer storage medium storing computer-useable instructions that, when used by the processor, cause the system to;

receive a context graph indicating a plurality of relationships among a plurality of nodes corresponding to components of a datacenter, each node comprising properties corresponding to a particular component of the datacenter;

identifying one or more relevant regions from the context graph, wherein each relevant region is a portion of the context graph and is associated with the datacenter;

for each relevant region of the context graph, determining whether each relevant region identified corresponds to a relevance condition and lacks an annotation describing the relevance condition;

determine a plurality of context hashes for each relevant region based on selected properties of a node of the plurality of nodes associated with the relevant region;

compare the plurality of context hashes to a plurality of subgraph hashes derived from a library of subgraphs to determine a set of subgraph hashes that are similar to the plurality of context hashes corresponding to each relevant region;

identify annotations corresponding to the set of subgraph hashes for each relevant region; and

annotate the node with one or more combined annotations, the combined annotations describing the relevance conditions of the datacenter at a particular state with a textual description.