Diese Präsentation wurde erfolgreich gemeldet.
Die SlideShare-Präsentation wird heruntergeladen. ×

Kafka at the core of an AIOps pipeline | Sunanda Kommula, Selector.ai and Alankar Sharma, Comcast

Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige
Anzeige

Hier ansehen

1 von 21 Anzeige

Kafka at the core of an AIOps pipeline | Sunanda Kommula, Selector.ai and Alankar Sharma, Comcast

Herunterladen, um offline zu lesen

Large networks consist of a diverse range of equipment, across private, public, hybrid clouds and partner networks. A hierarchical network has layers of infrastructure, catering to access, core, or distribution roles, managed by different organizations specialized to architect the right network hardware, software, and features for that network layer. The nature of data generated by each component can vary in type and form, including logs, events, metrics, or alarms.

The diversity of data generated by a large network is beyond human scale. Apache Kafka® is a critical hub in large networks, empowering AIOps to enhance decision making, improve analysis and insights by contextualizing large volumes of operational data. Kafka solved the big problem of collecting, processing, storing and normalizing data at scale, allowing us to focus on building the AIOps pipeline.

Our platform connects the dots across relevant operations data and provides operations teams with simple and powerful access to insights, from within increasingly popular collaboration environments like Slack and Microsoft teams. The pipeline must also integrate with automation solutions.

This session will cover how large volumes of streaming messages can be received by parallel Kafka consumers, and turned into action by network operations teams, dramatically reducing downtime and improving performance.

Large networks consist of a diverse range of equipment, across private, public, hybrid clouds and partner networks. A hierarchical network has layers of infrastructure, catering to access, core, or distribution roles, managed by different organizations specialized to architect the right network hardware, software, and features for that network layer. The nature of data generated by each component can vary in type and form, including logs, events, metrics, or alarms.

The diversity of data generated by a large network is beyond human scale. Apache Kafka® is a critical hub in large networks, empowering AIOps to enhance decision making, improve analysis and insights by contextualizing large volumes of operational data. Kafka solved the big problem of collecting, processing, storing and normalizing data at scale, allowing us to focus on building the AIOps pipeline.

Our platform connects the dots across relevant operations data and provides operations teams with simple and powerful access to insights, from within increasingly popular collaboration environments like Slack and Microsoft teams. The pipeline must also integrate with automation solutions.

This session will cover how large volumes of streaming messages can be received by parallel Kafka consumers, and turned into action by network operations teams, dramatically reducing downtime and improving performance.

Anzeige
Anzeige

Weitere Verwandte Inhalte

Diashows für Sie (20)

Ähnlich wie Kafka at the core of an AIOps pipeline | Sunanda Kommula, Selector.ai and Alankar Sharma, Comcast (20)

Anzeige

Weitere von HostedbyConfluent (20)

Aktuellste (20)

Anzeige

Kafka at the core of an AIOps pipeline | Sunanda Kommula, Selector.ai and Alankar Sharma, Comcast

  1. 1. Kafka at the core of an AIOps pipeline Presented by: Sunanda Kommula (Distinguished Engineer, Selector.ai) Alankar Sharma (Sr Principal Architect, Comcast) Hybrid Cloud AIOps
  2. 2. WWW.SELECTOR.AI Agenda • Comcast hybrid cloud • Key role of Kafka • Selector AI • Challenges and Solutions • System architecture • Observability of Data Ingestion • CI/CD with Kafka • AIOps
  3. 3. WWW.SELECTOR.AI Use Case • Hybrid cloud: infrastructure for Internet, voice, video, storage • Goal - Application connectivity • Wide variety of applications • Large cloud environment, operationally complex • Application perspective vs cloud status • End to end connectivity Hybrid Cloud Compute AI/ML Load Balancer
  4. 4. WWW.SELECTOR.AI The Hybrid Cloud • Cloud environment is complex, diverse and evolving • Application connectivity is key • Detect grey failures • Isolate wide-spread issues • Why AIOps? • Examine millions of data-points • Connect the dots • Faster root-cause, resolution and remediation Internet AWS Region 1 AWS Region 2 Data Center 1 Data Center 2 Data Center 3 Backbone Future Azure
  5. 5. WWW.SELECTOR.AI Kafka empowers AIOps • Critical hub of large networks • Connects varied data sources • Reliably delivers high velocity, volume & variety of data • Enables communication between microservices • Bridges organizations Broker Broker Broker AWS Cloud Watch Logs Kafka Clusters Synthetics Producer Producer Producer Producer Broker Broker Broker Network Monitoring Telemetry Engine Engine Broker Broker Broker ChatOps Portal ML Storage Kafka Cluster Device Metrics Query Visualize
  6. 6. WWW.SELECTOR.AI Selector AI: Turn-key AIOps for Instant Actionable Insights GET RESULTS WITHIN HOURS OF DATA ONBOARDING Declarative ETL Hybrid, Edge & Public Cloud Data Zero Config Analytics Automate Closed Loop Anomaly Remediation Zero touch user onboarding with NLP Slack as a Collaborative Notebook NETWORK APM SYNTHETICS DATA HYPERVISOR UNIQUE I-RANK ENGINE 6
  7. 7. WWW.SELECTOR.AI Challenges – The 6 V's of Big Data • Single topic • ~350Mbps • Noisy data • Subscribe • Filter Volume Variety • Metrics • Logs • SNMP • Avro • JSON Velocity • ~230Kpps • Deserialize • Time-sensitive • Ordered • Batch Veracity • Trust • Access model • Changing data • Changing model Variability • Statistical • Events • Correlate Value
  8. 8. WWW.SELECTOR.AI Solutions – Velocity & Volume • Message filter at ingest (I/O filter) - Head of line drop Broker message rate (~230Kpps) Post I/O filter (5Kpps) Byte pattern I/O filter 97% noise reduction
  9. 9. WWW.SELECTOR.AI Solutions – Velocity & Volume Most busy consumers (1.8Kpps each) Internal gochannel sizes per decoder • Leveraged Golang – performance, concurrency, channels • Live dashboards for cluster KPI monitoring and performance tuning
  10. 10. WWW.SELECTOR.AI Solutions – Velocity & Volume • Scale-out models Engine Pod 1 Engine Pod 2 Engine Pod 3 Kafka Cluster Broker Broker Broker Telemetry Metrics Engine Configuration Sharded Data Ingestion Shared consumer group - Data sharding Engine Pod 1 Engine Pod 2 Engine Pod 3 Kafka Cluster Broker Broker Broker Telemetry Metrics Engine Configuration Independent consumers – I/O filters Full Data Ingestion
  11. 11. WWW.SELECTOR.AI Solutions – Variety & Variability Dynamic schema Blocked patterns Allowed patterns Schema driven filters subscriptions: - label: sevone-spdb # kafka topic pathgroup: - group: port # port group paths: - 'indicatorName=ifHCInOctets' - 'indicatorName=ifHCInUcastPkts' - 'indicatorName=ifHCInMulticastPkts' - group: disk # disk group paths: - 'indicatorName=s1_sizebytes' - 'indicatorName=s1_usedbytes' - group: system # system group paths: - 'indicatorName=sysUpTime' - 'indicatorName=hrProcessorLoad' - group: bgp # bgp group paths: - 'indicatorName=bgpInTotalMessages' - 'indicatorName=bgpOutTotalMessages' - 'indicatorName=bgpPeerInUpdates' - 'indicatorName=bgpPeerOutUpdates' - 'indicatorName=bgpPeerState' Dynamic subscriptions
  12. 12. WWW.SELECTOR.AI Solutions – Value & Veracity • Declarative ETL • Extract, enrich, normalize • GraphQL based record selection • Deployment specific, zero code changes • Model events – syslogs and SNMP from devices • Correlate metrics, events and synthetics • Access considerations • Distributed control plane • Disaggregated data pipeline • SaaS, DMZ, on-prem PODs
  13. 13. WWW.SELECTOR.AI • CONFIDENTIAL AND PROPRIETARY System Architecture 13 Disaggregated data pipeline DMZ Engine Engine Engine Broker Broker Broker Producer Producer Producer Customer AWS Cloud Watch Network Monitoring Elastic Search Selector SaaS Kafka Cluster Broker Broker Broker Kafka Cluster Ingest Engine ML Storage ChatOps Query Visualize Portal
  14. 14. WWW.SELECTOR.AI • CONFIDENTIAL AND PROPRIETARY Engine Micro Architecture 14 Ingest I/O Filter Decode Match Label Format Export Extract Selector Pipeline
  15. 15. WWW.SELECTOR.AI Observability • Highly visible Kafka ingest pipeline • Aggregated & granular KPIs • broker-clusters, topics, consumers, partitions, decoders, match rates • Metric ingest KPIs • cumulative, port, BGP, system, memory • Reflects subscription configuration models Match rates per partition Metric ingest rate (~0.5Kpps) Port metrics ingest rate BGP metrics ingest rate
  16. 16. WWW.SELECTOR.AI CI/CD Pipeline • Why – Replicate every deployment in a test environment • What – On demand launch of Kafka cluster, fake data, fake signals • How – Jenkins pipelines • Deployment specific testbeds • Kafka cluster & producers • Integration, performance & scale tests • Fake test data for Kafka producer • Generate network metrics, syslogs, SNMP traps • Policy driven metrics • In & out of bounds traffic rates, cpu, memory, protocol metrics • Errors, alarms, synthetics and signals • Deterministic violations
  17. 17. WWW.SELECTOR.AI AIOps – Application Perspective Application Hop Counts Application Latency Application Packetloss
  18. 18. WWW.SELECTOR.AI AIOps – Network Fabric Status Fabric Status Device Logs BGP Peering Status Device Port Status Site Traffic In Site Traffic Out DC-1 DC-2 DC-3
  19. 19. WWW.SELECTOR.AI AIOps – Correlations • Connecting the dots • Detect grey failures • Examine impacts Port Availability BGP Status
  20. 20. WWW.SELECTOR.AI Summary • Operational complexity of a large, diverse, evolving cloud environment • Correlation of multiple data sources – The full picture • Kafka empowers AIOps by collecting and normalizing data at scale • Selector AIOps • Provides simple & powerful insights • Enhances decision making • Reduces downtime • Improves performance
  21. 21. Questions & Answers Thank you for your time!

×