MODULAR MODEL WORKFLOW IN A DISTRIBUTED COMPUTATION SYSTEM

• US 20170063886A1
• Filed: 10/30/2015
• Published: 03/02/2017
• Est. Priority Date: 08/31/2015
• Status: Active Grant

• Current Assignee: Splunk Inc.
• Original Assignee: Splunk Inc.
• Inventors: Kavacheri, Sathyanarayanan, Muddu, Sudhakar, Tryfonas, Christos
• Granted Patent: US 10,110,617 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/254 Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/444 Spatial browsing, e.g. 2D maps, 3D or virtual spaces
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence
• CPC Class Codes: G06F 3/0482 Interaction with lists of selectable items, e.g. menus
• CPC Class Codes: G06F 3/0484 for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
• CPC Class Codes: G06F 3/04842 Selection of displayed objects or displayed text elements G06F3/0482 takes precedence
• CPC Class Codes: G06F 3/04847 Interaction techniques to control parameter settings, e.g. interaction with sliders or dials
• CPC Class Codes: G06F 40/134 Hyperlinking
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06V 10/225 based on a marking or identifier characterising the area
• CPC Class Codes: H04L 2463/121 Timestamp cryptographic mechanisms or cryptographic arrangements involving time stamps H04L9/3297
• CPC Class Codes: H04L 41/0893 Assignment of logical groups to network elements
• CPC Class Codes: H04L 41/145 involving simulating, designing, planning or modelling of a network
• CPC Class Codes: H04L 41/22 comprising specially adapted graphical user interfaces [GUI]
• CPC Class Codes: H04L 43/00 : Arrangements for monitoring or testing data switching networks
• CPC Class Codes: H04L 43/045 : for graphical visualisation of monitoring data
• CPC Class Codes: H04L 43/062 : related to network traffic
• CPC Class Codes: H04L 43/08 : Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
• CPC Class Codes: H04L 63/06 : for supporting key management in a packet data network cryptographic mechanisms or cryptographic arrangements for key management H04L9/08
• CPC Class Codes: H04L 63/1408 : by monitoring network traffic monitoring network traffic per se H04L43/00
• CPC Class Codes: H04L 63/1416 : Event detection, e.g. attack signature detection
• CPC Class Codes: H04L 63/1425 : Traffic logging, e.g. anomaly detection
• CPC Class Codes: H04L 63/1433 : Vulnerability analysis
• CPC Class Codes: H04L 63/1441 : Countermeasures against malicious traffic countermeasures against attacks on cryptographic mechanisms H04L9/002
• CPC Class Codes: H04L 63/20 : for managing network security; network security policies in general filtering policies H04L63/0227
• CPC Class Codes: H05K 999/99 : dummy group

Abstract

A security platform employs a variety techniques and mechanisms to detect security related anomalies and threats in a computer network environment. The security platform is “big data” driven and employs machine learning to perform security analytics. The security platform performs user/entity behavioral analytics (UEBA) to detect the security related anomalies and threats, regardless of whether such anomalies/threats were previously known. The security platform can include both real-time and batch paths/modes for detecting anomalies and threats. By visually presenting analytical results scored with risk ratings and supporting evidence, the security platform enables network security administrators to respond to a detected anomaly or threat, and to take action promptly.

Claims

1. A method comprising:

extracting, from a model registry, a model type definition that includes a reference to model execution code and a processing mode specifier of a processing mode for a model deliberation workflow or a model training workflow, the processing mode specifier specifying at least one of a real-time processing mode or a batch processing mode;

implementing a model execution engine in a distributed computation system to execute processes that use machine learning to detect computer security related anomalies or threats in a computer network;

assigning, by the model execution engine, the model training workflow or the model deliberation workflow to the distributed computation system based on the processing mode specifier; and

scheduling, according to the model training workflow or the model deliberation workflow, a model processing thread, that corresponds to a portion of the model execution code and that is either a model training thread or a model deliberation thread, to a computing node for parallel processing in the distributed computation system.

2. The method of claim 1, further comprising storing the model type definition in the model registry implemented in a cache cluster or a distributed file system.

3. The method of claim 1, further comprising storing the model type definition in the model registry implemented in Redis or Hadoop Filesystem.

4. The method of claim 1, wherein the model training workflow or the model deliberation workflow is assigned to the distributed computation system when the processing mode specifier specifies the real-time processing mode and the distributed computation system has real-time task-parallel processing capability.

5. The method of claim 1, wherein the model training workflow or the model deliberation workflow is assigned to the distributed computation system when the processing mode specifier specifies the batch processing mode and the distributed computation system has batch data-parallel processing capability.

6. The method of claim 1, wherein the processing mode specifier specifies whether to process inputs in real-time or in batch mode when executing the model processing thread.

7. The method of claim 1, wherein the processing mode specifier is for the model training workflow and the model type definition specifies another processing mode specifier for the model deliberation workflow.

8. The method of claim 1, wherein the distributed computation system includes a distributed resource manager.

9. The method of claim 1, wherein the distributed computation system includes a distributed messaging system.

10. The method of claim 1, wherein the distributed computation system includes a task-parallel, real-time, distributed computation engine capable of running a data processing thread that reliably processes an unbounded data stream.

11. The method of claim 1, wherein the distributed computation system includes Apache Storm.

12. The method of claim 1, wherein the distributed computation system includes a data parallel, cluster-based, distributed computation engine.

13. The method of claim 1, wherein the distributed computation system includes Apache Spark.

14. The method of claim 1, wherein said assigning includes assigning the model deliberation workflow to the distributed computation system;

and further comprising instantiating a model deliberation thread by configuring model deliberation processing logic defined by the model execution code with a model state from a model store.

15. The method of claim 1, wherein said assigning includes assigning the model training workflow to the distributed computation system;

and further comprising instantiating the model training thread according to model training processing logic defined by the model execution code.

16. The method of claim 1, wherein the model type definition specifies an event view subscription configured to filter for a specific type of data events;

and further comprising inputting an event feature set corresponding to the specific type of data events to the model processing thread.

17. The method of claim 1, wherein the model type definition specifies a model type topology;

and further comprising determining, based on the model type topology, how many model processing threads of the model type definition to instantiate during either the model training workflow or the model deliberation workflow.

18. The method of claim 1, wherein the model type definition specifies a model type topology;

and further comprising;

identifying entities falling within the model type topology; and

instantiating model processing threads corresponding respectively to the entities.

19. The method of claim 1, wherein the model type definition specifies a model type topology;

and further comprising;

identifying entities falling within the model type topology by querying a machine data recording device in the computer network; and

instantiating model processing threads that respectively correspond to the entities.

20. The method of claim 1, wherein the model type definition specifies a model type topology;

and further comprising partitioning, according to the model type topology, event feature sets to feed into model processing threads running on different processing nodes of the distributed computation system.

21. The method of claim 1, further comprising executing the model training thread in the distributed computation system;

wherein said executing the model training thread includes;

processing event feature sets to compute a model state of the model type definition; and

storing the model state in a model store.

22. The method of claim 1, further comprising training a plurality of models of the model type definition based on event feature sets, each of the plurality of models corresponding to a different entity.

23. The method of claim 1, further comprising training an entity-specific model or a purpose-specific model of the model type definition;

and storing multiple versions of the entity-specific model or the purpose-specific model at different stages of said training.

24. The method of claim 1, wherein the model type definition includes a model type topology and an event view subscription;

and further comprising;

identifying event feature sets to feed into a plurality of model processing threads for the model training workflow or the model deliberation workflow according to the events view subscription; and

partitioning the event feature sets and the model processing threads into groups, wherein each group corresponds to a worker node in the distributed computation system.

25. The method of claim 1, wherein the model type definition includes a model type topology;

and further comprising;

performing a consistent hash on event feature sets selected based on the model type definition; and

partitioning, based on the consistent hash, the event feature sets such that a worker node in the distributed computation system running the model processing thread receives only a subset of the event feature sets.

26. The method of claim 1, wherein the model type definition includes a model type topology;

and further comprising;

determining a number of entities corresponding to the model type topology; and

partitioning event feature sets selected based on the model type definition such that each worker node in the distributed computation executing at least a model processing thread of the model training workflow or the model deliberation workflow receives only a subset of the event feature sets, wherein said partitioning includes determining a number of partitions based on the number of entities and a number of available workers in the distributed computation system.

27. The method of claim 1, wherein the distributed computation system includes a plurality of computing machines, each of the computing machines implementing at least a worker node capable of running at least a model processing thread;

and wherein said assigning includes assigning either a plurality of entity-specific model training threads of the model training workflow or a plurality of entity-specific model deliberation threads of the model deliberation workflow to each worker node.

28. The method of claim 1, wherein the model type definition includes a model type topology;

wherein the distributed computation system includes a plurality of computing machines, each of the computing machines implementing at least a worker node capable of running at least a model processing thread; and

wherein each worker node receives from a single partition of event feature sets according to the model type topology, the event feature sets identified based on an event view subscription of the model type definition.

29. A system comprising:

a distributed computation system;

a model registry configured to store a model type definition including a reference to model execution code and a processing mode specifier for a processing mode of a model deliberation workflow or a model training workflow, the processing mode specifier specifying at least one of a real-time processing mode or a batch processing mode; and

a model execution engine implemented on the distributed computation system to execute processes that use machine learning to detect computer security related anomalies or threats in a computer network;

wherein the model execution engine is configured to;

assign the model training workflow or the model deliberation workflow to the distributed computation system based on the processing mode specifier; and

schedule, according to the model training workflow or the model deliberation workflow, a model processing thread, that corresponds to a portion of the model execution code and that is either a model training thread or a model deliberation thread, to a computing node for parallel processing in the distributed computation system.

30. A computer readable medium storing instructions there on which, when executed by a processor, cause the processor to:

define a model type definition including a reference to model execution code and a processing mode specifier of a processing mode for a model deliberation workflow or a model training workflow, the processing mode specifier specifying at least one of a real-time processing mode or a batch processing mode;

implement a model execution engine on a distributed computation system to execute processes that use machine learning to detect computer security related anomalies or threats in a computer network;

assign, by the model execution engine, the model training workflow or the model deliberation workflow to the distributed computation system based on the processing mode specifier; and

schedule, according to the model training workflow or the model deliberation workflow, a model processing thread, that corresponds to a portion of the model execution code and that is either a model training thread or a model deliberation thread, to a computing node for parallel processing in the distributed computation system.