![2
Legal Disclaimer
General Disclaimer:
© Copyright 2019 Intel Corporation. All rights reserved. Intel, the Intel logo, Intel Inside, the Intel Inside logo, Intel.
Experience What’s Inside are trademarks of Intel. Corporation in the U.S. and/or other countries. *Other names and
brands may be claimed as the property of others.
Technology Disclaimer:
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software
or service activation. Performance varies depending on system configuration. No computer system can be absolutely
secure. Check with your system manufacturer or retailer or learn more at [intel.com].
Performance Disclaimers:
Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified
circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does
not guarantee any costs or cost reduction.
Results have been estimated or simulated using internal Intel analysis or architecture simulation or modeling, and
provided to you for informational purposes. Any differences in your system hardware, software or configuration may
affect your actual performance.](https://image.slidesharecdn.com/closedloopplatformautomation-tongzhong-emmacollins-190909230609/85/Closed-Loop-Platform-Automation-Tong-Zhong-Emma-Collins-2-320.jpg)
Closed Loop Platform Automation - Tong Zhong & Emma Collins
- 2. 2 Legal Disclaimer General Disclaimer: © Copyright 2019 Intel Corporation. All rights reserved. Intel, the Intel logo, Intel Inside, the Intel Inside logo, Intel. Experience What’s Inside are trademarks of Intel. Corporation in the U.S. and/or other countries. *Other names and brands may be claimed as the property of others. Technology Disclaimer: Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. No computer system can be absolutely secure. Check with your system manufacturer or retailer or learn more at [intel.com]. Performance Disclaimers: Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction. Results have been estimated or simulated using internal Intel analysis or architecture simulation or modeling, and provided to you for informational purposes. Any differences in your system hardware, software or configuration may affect your actual performance.
- 3. ScaleEfficiencywithData-Driven,ClosedLoopAutomation Intel Platform Features are part of intelligent, closed loop solutions that are reactive, proactive and predictive, delivering new levels of efficiency for IT and network infrastructure. Automated Action Telemetry Analysis Software and Services Telemetry IA Platform Telemetry Fine-grained Hardware and software insights feeding operational intelligence and automation
- 4. Intel's Ingredients for Closed Loop Automation Intel driven Analytics solutions using IA feature data OR Integrate with proprietary/commercial network monitoring and analytics solutionsOrchestration IA Platform Telemetry Analytics Intel driven MANO solutions to provide scale/heal/placement actions that incorporate IA features OR Integrate with proprietary/commercial MANO solutions IA Feature Metrics and Events Exposure IA Feature Detection and Provisioning Power PMU RDT RAS Other… The Closed Loop Intel Platform
- 5. Inteltelemetrycollectionandpublication Platform Compute Networking Memory Storage Acceleration Collectd South Bound Plugins Collectd Collectd North Bound Plugins Open stack MANO Platform/ NFVI Monitoring /Analytics Systems Telemetry Publication Telemetry Consumption Kubernetes ONAP Telemetry Collection Application
- 6. 6 IntelTelemetryCoverage Compute Network Storage Hypervisor NFVIVirtualised Compute Virtualised Network Virtualised Storage Collectd PMU counters NIC counters vSwitch counters Common / Standard Open APIs VM Stall Detection/ RT Stall Detection Enterprise and Network Management Tools RAS Hypervisor/Container Counters Intel® Node Manager Open Platform Collector Intel® Run Sure Technology MCA* PCIe AER Resilient System Technology Resilient Memory Technology SDDC DDDC+1 Mirroring RAID Intel® Rapid Storage Technology Intel® Management Engine IPMI C M T Intel® RDT C A T M B M C D P P O W E R VIM Intel® Infrastructure Management Technologies Redfish SYSLOG KafkaSNMP API VES Plugin Prometheus OpenStack Kubernetes
- 7. TyingtheIntelPlatformtotheCustomerBusinessCase Customer Business Use Case Platform Slicing (Part of 5G Network Slicing) Platform Resiliency Security (side channel attacks) Power Optimisation Intel Platform Power PMU RDT RAS Other… MANO Analytics
- 8. MachineLearningWillCompletetheEvolution Learn Learn Learn Learn Watch Decide Collect Act App Server/Storage/Network Watch Expose Platform and App data through APIs Act Mechanism for Policy Activation and Enforcement Learn Machine Learning for continuous improvement Decide Analytics to define correct response Moving From Automation to Self-OptimizationTaking the manpower out of networking, cloud…
- 9. DeliveringClosedLoopAutomationforNFV Automated Action IA Platform Telemetry Telemetry Analysis 1. Enable IA Service Assurance Telemetry through instrumentation and exposure in an industry standard manner 2. Enable IA Telemetry Analytics through Telemetry Compaction, KPI identification and prediction 3. Enable Closed Loop Automation through Orchestration enabling and industry proof points
- 10. Casestudy:MLforNFVserviceassurance 10 • Efficient dynamic network management is one major challenge for NFV • Machine learning plays an important role in addressing this challenge by analyzing gathered data for various purposes: • dynamic resource allocation • security threats alert • performance degradation detection • demand prediction “Cognet: A Network Management Architecture Featuring Cognitive Capabilities,” Proc. Euro. Conf. Networks and Commun., June 2016 1. Data pre-processing Feature Engineering/Reduction 2. KPI prediction/forecasting Regression/Classification 3. Closed loop optimization Reinforcement Learning
- 11. NFVtestsystems 11 vEPC – virtual Evolved Packet Core vCMTS – virtual Cable Modem Termination System • Telemetry data dumped through Collectd includes CPU, PMU, Memory, Load, etc. • Total number of telemetry data hundreds to thousands sampled at configured interval (1s or 10s) • Target KPI (Key Performance Indicator): packet drop rate
- 12. Datapre-processing 12 • Data filtering – remove irrelevant data (e.g. control plane data for this case) 1065 features remaining • Data alignment and interpolation • Feature selection – remove features with no change over time 726 features remaining • Data normalization • Data splitting to training, validation and testing sets Tens of thousands of samples split into 8:1:1
- 13. Telemetryfeaturecompaction • Feature Selection • The process of selecting a subset of relevant features for use in the model construction • Filter methods – Select features based on scoring from statistical measures e.g. SelectKBest • Wrapper methods – search for optimal feature combination that results in best predictive results e.g. Recursive Feature Elimination, Boruta • Unsupervised learning methods – group features that behave similarly e.g. FeatureAgglomeration • Feature Transformation (dimension reduction) • Convert the feature vector into lower dimension space with learned transformations • Supervised: PLS (Partial Least Squares), CCA (Canonical Correlation Analysis), LDA (Linear Discriminant Analysis) • Unsupervised: PCA (Principal Component Analysis) 13
- 14. 10 20 40 Feature 10 20 40 Feature 10 20 40 Feature • cpu_value_idle_18 • • intel_rdt_value_bytes_llc_20 • • • intel_rdt_value_memory_bandwidth_local_12 • cpu_value_idle_49 • • intel_rdt_value_bytes_llc_27 • intel_rdt_value_memory_bandwidth_local_3 • • cpu_value_interrupt_18 • intel_rdt_value_bytes_llc_41 • intel_rdt_value_memory_bandwidth_local_33 • cpu_value_interrupt_20 • • intel_rdt_value_bytes_llc_50 • intel_rdt_value_memory_bandwidth_local_53 • cpu_value_system_39 • intel_rdt_value_bytes_llc_53 • • intel_rdt_value_memory_bandwidth_local_9 • • cpu_value_user_13 • • intel_rdt_value_bytes_llc_6 • intel_rdt_value_memory_bandwidth_remote_2 • cpu_value_user_3 • intel_rdt_value_bytes_llc_8 • intel_rdt_value_memory_bandwidth_remote_6 • cpu_value_user_46 • intel_rdt_value_ipc_nan_22 • • ipmi_value_fanspeed_System Fan 1 fan_cooling (29.1) • cpu_value_user_9 • • intel_rdt_value_ipc_nan_23 • ipmi_value_temperature_HSBP 1 Temp drive_backplane (15.1) • df_value_free_etc-hosts • intel_rdt_value_ipc_nan_26 • ipmi_value_temperature_LAN NICTemp system_board (7.1) • df_value_used_etc-hosts • intel_rdt_value_ipc_nan_5 • ipmi_value_temperature_P2 DTS Therm Mgn processor (3.2) • disk_read_disk_time_sda • • intel_rdt_value_ipc_nan_50 • • • irq_value_TLB • intel_pmu_value_branches_42 • intel_rdt_value_ipc_nan_53 • • • load_longterm • intel_pmu_value_instructions_14 • intel_rdt_value_ipc_nan_54 • • • load_midterm • intel_pmu_value_page-faults_51 • intel_rdt_value_ipc_nan_9 • load_shortterm • intel_pmu_value_page-faults_all • numa_value_other_node_node1 • • • memory_value_cached • • memory_value_slab_unrecl Top TopTop TopTelemetryFeaturesSelectedByML 14 Recursive feature elimination stepping down the target from 40 20 10
- 15. Featureselectionresults 15 • Applying various feature selection algorithms to reduce the number of telemetry data sampled without compromising the prediction accuracy • Smaller set of telemetry data saves training/inference time • vEPC test data set with original feature dimension ~400 • GradientBoostRegressor algorithm used for packet loss rate prediction • Select top number of features from feature importance output of GradientBoostRegressor
- 16. KPIprediction&forecasting–supervisedlearning 16 • Regression • Train the model to predict target KPI using reduced telemetry data samples • Example: Packet drop rate • Accuracy Measured by MSE (Mean Squared Error) – the smaller the better • Classification • Detect packet drop from telemetry data • Measurement • Accuracy_score = correct_prediction / total_sample • Precision/Recall/F1-score, etc. • Time Series Forecasting • Predict future KPI value based on historical telemetry data and observed KPI trend
- 17. KPIforecastingusingLSTM 17 • vCMTS downlink test data • KPI – scheduling packet loss rate • Input: previous 60 seconds of 20 selected telemetry data + KPI • Output: predicted future value (5 seconds later) of target KPI • Tensorflow BasicLSTMCell: Two layers, each layer 150 neurons
- 18. Closed-loopAutomation Xeon vCMTS - 0 Traffic Generator vCMTS - 0VNFs NIC Machine Learning Modules collectd InfluxDB Traffic forecasting Reinforcement learning HW Optimization (RDT, DVFS, WL consolidation, etc.) Compacted metrics Using ML: • Track/forecast workload, performance • Dynamically adjust resource allocation Benefits: • Reduced TCO through power saving and increased HW utilization
- 19. Frequencytuning
- 20. ENIPoC–NetworkSliceLifecycleManagement 20 • For generating new scale up/down and converting the intent to suggested configuration • LSTM is used for traffic prediction AI-based predictor: TNSM: CNSM: • Provides underlay network control to satisfy the network slice requests • FlexE and a FlexE-based optimization algorithm are used for underlay network slice creation and modification • Provides core network control to satisfy the network slice requests ETSI ISG ENI – Experiential Networked Intelligence
- 21. Trafficpredictionforresourceoptimization 21 Blue: actual traffic Orange: predicted traffic
- 24. Further Resources Learn more from these helpful sites: https://networkbuilders.intel.com/network-technologies/serviceassurance https://wiki.opnfv.org/display/fastpath/Barometer+Home https://wiki.openstack.org/wiki/Telemetry https://01.org/openstack/blogs/2015/openstack-enhanced-platform- awareness-white-paper
- 25. 25 Collectd101materials • Collectd 101 • https://wiki.opnfv.org/display/fastpath/Collectd+101 • Write simple read plugin • https://wiki.opnfv.org/display/fastpath/Collectd+how+to+implement+a+si mple+plugin
- 26. 26 Barometer Strategy: • Ensure platform metrics/events are accessible through open industry standard interfaces. • Demonstrate IA platform technologies can be monitored, consumed and actioned in real time Opnfvbarometer–Intelplatformfeatureplugins One Click Install: • Easy install/configuration for customers • One command to install Collectd/Influxdb/Grafana Three container approach for Collectd: • Stable Container: latest stable branch • Master Container: up to date with master • Experimental Container: cherry pick features of interest
- 27. BarometerLinks Barometer Home: https://wiki.opnfv.org/display/fastpath/Barometer+Home Metrics/Events through Barometer (not on Collectd site): https://wiki.opnfv.org/display/fastpath/Collectd+Metrics+and+Events#CollectdM etricsandEvents-Metrics Barometer “One-click” install: https://wiki.opnfv.org/display/fastpath/One+Click+Install+of+Barometer+Contain ers 27