More Related Content
Similar to An emulation framework for IoT, Fog, and Edge Applications
Similar to An emulation framework for IoT, Fog, and Edge Applications (20)
Recently uploaded
Recently uploaded (20)
An emulation framework for IoT, Fog, and Edge Applications
- 2. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing The Internet of Things 2 ● Transforming the physical world into an information system. ● IDC predicts that by 2025 there will be 41.8B connected IoT devices and data generated from them to be 73.1 ZB by 2025. Analytic Insights Computing Platform ● It only seems “natural” that IoT services offload analytic jobs to the compute resources for processing. https://www.idc.com/getdoc.jsp?containerId=prAP46737220, 2020
- 3. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing The Internet of Things 3 ● Transforming the physical world into an information system. ● IDC predicts that by 2025 there will be 41.8B connected IoT devices and data generated from them to be 73.1 ZB by 2025. Analytic Insights Computing Platform ● It only seems “natural” that IoT services offload analytic jobs to the compute resources for processing. https://www.idc.com/getdoc.jsp?containerId=prAP46737220, 2020
- 4. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Application Use-case 4 A microservice-based application is created, which computes statistics from user (e.g., the average vehicle delay per city segment) Use-case: A taxi company wants to analyse region-based data from its fleet.
- 5. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Partial Aggregator Application Use-case 5 A microservice-based application is created, which computes statistics from user (e.g., the average vehicle delay per city segment) Application Services: ● Partial Aggregator exposes an API on which the IoT devices send their data, performs region-based aggregations, and stores the results in an in-memory buffer Partial Aggregator Partial Aggregator Use-case: A taxi company wants to analyse region-based data from its fleet.
- 6. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Partial Aggregator Application Use-case 6 A microservice-based application is created, which computes statistics from user (e.g., the average vehicle delay per city segment) Application Services: ● Partial Aggregator exposes an API on which the IoT devices send their data, performs region-based aggregations, and stores the results in an in-memory buffer ● Back-end Server requests periodically the data from the partial aggregators, performs the final processing and stores the final results in a Database Back-end Server Overall Aggregations Partial Aggregator Partial Aggregator Use-case: A taxi company wants to analyse region-based data from its fleet.
- 7. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Partial Aggregator Application Use-case 7 A microservice-based application is created, which computes statistics from user (e.g., the average vehicle delay per city segment) Application Services: ● Partial Aggregator exposes an API on which the IoT devices send their data, performs region-based aggregations, and stores the results in an in-memory buffer ● Back-end Server requests periodically the data from the partial aggregators, performs the final processing and stores the final results in a Database Back-end Server Overall Aggregations Partial Aggregator Partial Aggregator Use-case: A taxi company wants to analyse region-based data from its fleet. Users can evaluate the data via a Dashboard and they can submit different queries.
- 8. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Deployment 8 The company will purchase MECs and will place them in different regions inside a city.
- 9. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Deployment 9 The company will purchase MECs and will place them in different regions inside a city. Application Deployment: Partial Aggregator Services are deployed on MECs and the Back-end server on cloud. Back-end Server Partial Aggr Partial Aggr Partial Aggr
- 10. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Deployment 10 The company will purchase MECs and will place them in different regions inside a city. Application Deployment: Partial Aggregator Services are deployed on MECs and the Back-end server on cloud. Back-end Server Partial Aggr Partial Aggr Partial Aggr Taxis send their data to nearby MEC (aka Partial Aggregator Service) and If there is no reachable MEC (suburbs), the data are forwarded to Cloud via mobile internet.
- 11. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Deployment 11 The company will purchase MECs and will place them in different regions inside a city. Application Deployment: Partial Aggregator Services are deployed on MECs and the final Back-end server on cloud. Back-end Server Partial Aggr Partial Aggr Partial Aggr Taxis send their data to nearby MEC (aka Partial Aggregator Service) and If there is no reachable MEC (suburbs), the data are forwarded to Cloud via mobile internet. The involving actors in this scenario like Operators, IoT Developers, Performance Evaluators, face numerous challenges in deployment realization…
- 12. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Plethora of Devices 12 Wide range of devices comes in different shapes, capabilities, sizes and prices. But, not all meet the desired QoS… And if we are wrong… we need more time, resources, money, etc… Raspberry Pi 4 Nvidia Jetson Nano Nvidia AGX Xavier ARM 4 cores @1.5GHz 4GB RAM $54 ARM 4 cores @1.4GHz 4GB RAM + GPU No wifi $69 HiKey 970 ARM 8 cores @2.3GHz 6GB RAM Mali-G72 GPU $299 ARM 8 cores @2.2GHz 32GB RAM 512-core GPU $695 Back-end Server Partial Aggr Partial Aggr Partial Aggr
- 13. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scalable Geo-Distributed Deployments 13 Even if the proper devices are selected… We need to install them in their physical locations… Back-end Server Partial Aggr Partial Aggr Partial Aggr 90ms 10-18ms 2ms ● Installation Efforts ● Manual Configurations ● Time-consuming Deployment ● Network Connectivity evaluation…
- 14. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scalable Geo-Distributed Deployments 14 90ms 10-18ms 2ms Partial Aggr Partial Aggr Partial Aggr Back-end Server Even if the proper devices are selected… We need to install them in their physical locations… In a Geo-distributed environment, scalability increases exponentially the difficulties… ● Installation Efforts ● Manual Configurations ● Time-consuming Deployment ● Network Connectivity evaluation…
- 15. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Positioning and Mobility 15 Different distances and Antenna configurations offer different network QoS. To reproduce similar situations, developers should perform in-place experiments with physical devices which is extremely time-consuming during the development period. Partial Aggr Partial Aggr Partial Aggr Back-end Server
- 16. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Random Generated Reliability Problems 16 The ad-hoc changes on Edge environments and wireless networks are common, so developers need to investigate errors regarding: ● Node disconnections ● Failstops ● Link drops ● Network QoS fluctuations ● Packet drops ● Load fluctuations ● … How do these errors affect the quality of service, running cost, decision-making, overall reliability, etc., of their applications? Partial Aggr Partial Aggr Partial Aggr Back-end Server
- 17. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Monitoring 17 Actors need to generate useful metrics and insights about their deployments in order to make the right decisions maximizing the performance and minimizing the running cost. ● Device & System (e.g., cpu, memory, disk i/o, network i/o, energy consumption, etc.) ● OS & Virtualization Stack (per service) ● Application Metrics ● Network-wise statistics ● … Partial Aggr Partial Aggr Partial Aggr Back-end Server
- 18. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing To this end, the users seek solutions to model and analyze the behavior of Infrastructure and IoT services… Fog Design and Deployment Challenges 18 ● Plethora of Devices ● Scalable Geo-Distributed Deployments ● Positioning and Mobility ● Random Generated Reliability Problems ● Monitoring
- 19. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Simulators & Emulators 19 ● Simulators must have models for every piece of infrastructure, application behavior and their interactions… application does not actually run… ● Emulators mimic production-end environment and application is executed in real-time with relatively low cost. ● Fog Emulators provide resource and network heterogeneity but… what about ad-hoc topology alterations, mobility, 5G concepts, scalability, etc…
- 20. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Emulators 20 Resource Heterogeneity Network Heterogeneity Ad-hoc Changes Mobility 5G-concepts Multi-host scalability MeDICINE MockFog Marvis FogBed Fogify MeDICINE: Rapid prototyping of production-ready network services in multi-PoP environments, M. Peuster et al., IEEE NFV-SDN, 2016 MockFog 2.0: Automated Execution of Fog Application Experiments in the Cloud, J. Hasenburg et al., IEEE Transactions on Cloud Computing, 2021 Towards a Staging Environment for the Internet of Things, J. Beilharz, ΙΕΕΕ PerCom Workshops, 2021 Scalable Fogbed for Fog Computing Emulation, A. Coutinho et al., IEEE Symposium on Computers and Communications (ISCC) 2018
- 21. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fogify A Fog Computing Emulation Framework
- 22. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 22 Outline ● Features ● Key Design Points ● System Overview ● Modeling Abstractions ○ Fog Modeling Abstractions ○ Runtime Evaluation Model ● Implementation Details ● Evaluation ○ Accuracy Evaluation ○ Usability Evaluation ● Conclusion
- 23. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fogify Emulation Framework 23 ● Fog & 5G Modeling Abstractions ● Resource & Network Link Heterogeneity ● Controllable Faults & Alterations ● Geo-positioning and Mobility ● Any-scale Experimentation ● Monitoring Capabilities Fogify: A Fog Computing Emulation Framework, M. Symeonides et al, ACM/IEEE Symposium on Edge Computing (SEC), 2020
- 24. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Key Design Points 24 ● Applications adopt multi-service design (microservices, big-data engines, etc.) ● Application services are containerized adopting Docker containers ● IoT data generators are also virtualized and containerized processes that only generate and transmit simple data points to other services ● The application runs in real-time through the emulation Environment, and ● The emulator takes care of shaping the static and dynamic performance traits of: ○ Compute Resources (physical or virtual) like CPU, RAM, disk, etc. ○ Networks and their Quality of Service like network latency, data rate, error rate, etc.
- 25. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 25 System Overview
- 26. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 26 System Overview
- 27. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 27 System Overview
- 28. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 28
- 29. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 29
- 30. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 30
- 31. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 31
- 32. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 32
- 33. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing System Overview 33
- 34. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 34 Fog Modeling Abstractions
- 35. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Templates 35 The initial Fog Templates of Emulator consist of: ● a set of Services, ● a set of Nodes, ● a set of Networks Partial Aggr Partial Aggr Partial Aggr Back-end Server
- 36. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Templates 36 Partial Aggr Partial Aggr Partial Aggr The initial Fog Templates of Emulator consist of: ● a set of Services, ● a set of Nodes, ● a set of Networks Services are containerized micro-programs and are inherited from specification of docker-compose file. Back-end Server
- 37. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Templates 37 The initial Fog Templates of Emulator consist of: ● a set of Services, ● a set of Nodes, ● a set of Networks Partial Aggr Partial Aggr Partial Aggr Nodes describe the compute resources of the infrastructure and include parameters such as CPU, Memory, and Disk Back-end Server
- 38. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Templates 38 The initial Fog Templates of Emulator consist of: ● a set of Services, ● a set of Nodes, ● a set of Networks Partial Aggr Partial Aggr Partial Aggr Networks are interconnected mesh networks that will connect the emulated nodes. Each network includes a set of QoS, like network latency, data rate, etc., along with VNFs. Back-end Server
- 39. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Templates 39 The initial Fog Templates of Emulator consist of: ● a set of Services, ● a set of Nodes, ● a set of Networks Partial Aggr Partial Aggr Partial Aggr But networks may have more complex structure and QoS like 5G-Networks… Back-end Server
- 40. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Networks Template 40 A 5G Network includes: ● Midhaul Connectivity Characteristics
- 41. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 41 A 5G Network includes: ● Midhaul Connectivity Characteristics ● Backhaul Connectivity Characteristics 5G Networks Template
- 42. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 42 A 5G Network includes: ● Midhaul Connectivity Characteristics ● Backhaul Connectivity Characteristics ● Wireless Connections 5G Networks Template
- 43. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 43 A 5G Network includes: ● Midhaul Connectivity Characteristics ● Backhaul Connectivity Characteristics ● Wireless Connections ● Virtual Network Functions 5G Networks Template
- 44. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 44 A 5G Network includes: ● Midhaul Connectivity Characteristics ● Backhaul Connectivity Characteristics ● Wireless Connections ● Virtual Network Functions ● Radio Units (Basestations) locations 5G Networks Template
- 45. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Topology 45 Partial Aggr Partial Aggr Partial Aggr A Fog Topology consists of Blueprints. ● A Blueprint is a combination of a Node, Service, a set of Networks, a location, a replication factor, and a label Back-end Server
- 46. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Topology 46 Partial Aggr Partial Aggr Partial Aggr A Fog Topology consists of Blueprints. ● A Blueprint is a combination of a Node, Service, a set of Networks, a location, a replication factor, and a label Back-end Server 10 blueprints exist in this topology
- 47. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Fog Topology 47 Partial Aggr Partial Aggr Partial Aggr A Fog Topology consists of Blueprints. ● A Blueprint is a combination of a Node, Service, a set of Networks, a location, a replication factor, and a label Back-end Server
- 48. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 48 Runtime Evaluation Model
- 49. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Runtime Evaluation Model 49 Actions change properties of a running Fog Topology. Actions can be:
- 50. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 50 Actions change properties of a running Fog Topology. Actions can be: ○ Failure Actions Runtime Evaluation Model
- 51. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 51 Actions change properties of a running Fog Topology. Actions can be: ○ Failure Actions ○ Scaling Actions (horizontal or vertical) Runtime Evaluation Model
- 52. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 52 Actions change properties of a running Fog Topology. Actions can be: ○ Failure Actions ○ Scaling Actions (horizontal or vertical) ○ Network Actions Runtime Evaluation Model
- 53. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 53 Actions change properties of a running Fog Topology. Actions can be: ○ Failure Actions ○ Scaling Actions (horizontal or vertical) ○ Network Actions ○ Stress Actions Runtime Evaluation Model
- 54. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 54 Actions change properties of a running Fog Topology. Actions can be: ○ Failure Actions ○ Scaling Actions (horizontal or vertical) ○ Network Actions ○ Stress Actions ○ Positioning Updates Runtime Evaluation Model
- 55. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 55 A Scenario or Trajectory is a sequence of time scheduled actions that are used to emulate more complex user-driven experiments (e.g., multiple errors or moving actions). Runtime Evaluation Model
- 56. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 56 A Scenario or Trajectory is a sequence of time scheduled actions that are used to emulate more complex user-driven experiments (e.g., multiple errors or moving actions). Runtime Evaluation Model
- 57. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 57 A Scenario or Trajectory is a sequence of time scheduled actions that are used to emulate more complex user-driven experiments (e.g., multiple errors or moving actions). Runtime Evaluation Model
- 58. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 58 Implementation Details
- 59. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing FogifySDK 59 Fogify SDK realized as python library and provides: ● Deployment of Topologies, Actions & Scenarios ● Retrieval of the Monitoring data ● Out-of-the-box exploratory analysis ● Functions for building programmably the Fogify’s model Jupyter Integration A Monitoring Agent retrieves: ● Basic Metrics (CPU, RAM, network, disk) ● Application's metrics Monitoring System Monitoring System
- 60. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Computational Resources Translation 60 Fogify utilizes the containers as execution platform while they achieve isolation, through linux namespaces, and resource constraining by using Linux Control Groups (cgroups). Fogify maps the high-level device’s template to the host’s capabilities. For some resources, like memory, the mapping is straightforward but… for CPU, cgroups give us only the ability to set a portion of host CPU. For that: ● Fogify computes the cumulative clock rate (CCR) for both host and model as: ● Then, it specifies the rate of the processing power that the emulated device will occupy from host:
- 61. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Enforcement of the Network QoS & VNFs 61 ● The system realizes a virtual overlay mesh network (Virtual Extensible Lan (VXLAN)) per network description ● Then, emulation agents apply the network functions, so for each emulated Node: ○ Apply firewall rules via iptables tool ○ Inspect the packets and exports statistics ○ Build a tree-base structure by utilizing Classful Queuing Disciplines (qdisc) and apply QoS Rules by utilizing the linux tc-tool
- 62. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 62 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK.
- 63. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 63 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 64. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 64 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 65. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 65 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Nodes are RUs, UEs, Fog devices, and ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 66. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 66 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Nodes are RUs, UEs, Fog devices, and ● Links denote their connectivity with network QoS ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 67. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 67 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Nodes are RUs, UEs, Fog devices, and ● Links denote their connectivity with network QoS ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. , where Shannon’s Formula for wireless MIMO channels Signal-to-noise ratio computation ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 68. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 5G Network Plug-in 68 5G-Slicer: An emulator for mobile IoT applications deployed over 5G network slices, M. Symeonides et al, ACM/IEEE IoTDI 2022 5G-Slice Model FogifySDK Network Conceptual Graph ● Creates the Network Conceptual Graph that is an in-memory structure ● Nodes are RUs, UEs, Fog devices, and ● Links denote their connectivity with network QoS ● Transforms the RU-to-UE distance to respective QoS ● Translates Network Conceptual Graph to Emulated Execution Environment via FogifySDK. , where Shannon’s Formula for wireless MIMO channels Signal-to-noise ratio computation ● 5G-Slicer takes as input 5G-slice models and creates an emulated execution environment via Fogify
- 69. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Evaluation 69
- 70. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 70 Network Accuracy 70 VM Paris VM London wifi i n t e r n e t internet Ethernet We measured the latency and data rate among the devices with ping tool and perf, respectively. Then, we emulated the same infrastructure and captured the same measurements. Emulator achieves near to real-world network link capabilities, only with outliers that are not captured and a slight overhead in low-latency connections
- 71. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 71 Application Accuracy 71 wifi internet Ethernet On the same infrastructure, we deployed a small microservice ML inference application ● Image inference on Paris VM ● Preprocessing service on Edge Server (reduces the image size) ● Laptop (workload generator) sends an image either to Cloud or to Edge preprocessor Preprocessing Workload Gen The emulation results closely follow the real measurements with a 5%-8% deviation of the overall experiment time. VM Paris Cloud Server
- 72. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 72 Emulation Scalability & 5G Accuracy Emulation Scalability: Emulator is only bounded by the underlying resources. When the resource pool expands, the emulation capabilities proportionally increase with a comparable raising of the bootstrapping time. Real vs Emulated MIMO connectivity: The emulation results follow the distribution of the real testbed with the mean absolute percentage error being 11.7% for 23dBm and 5.3% for the 33dBm configuration. “Patras 5g: An open source based end-to-end facility for 5g trials” C. Tranoris and S. G. Denazis, ERCIM News, 2019 “A holistic approach for 5g network slice monitoring” D. Giannopoulos, P. Papaioannou, L. Ntzogani, C. Tranoris, and S. Denazis, MeditCom2021
- 73. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Usability Evaluation 73
- 74. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties 74
- 75. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties We can observe the minimum execution utilization of all services and Fog nodes. Common execution of the infrastructure. 75
- 76. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties 2s network delay at MEC region-1 for 3 minutes 76 We identify a throughput degradation Network traffic increases to the previous level, since the latency does not exceed the pre-defined 10s disconnect limit
- 77. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties A data transfer spike is appeared for the MEC and Car nodes due to queued requests. Common execution for 3 minutes 77
- 78. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties 20s network delay at MEC region-1 for 3 minutes 78 When the network latency is boosted to 20s, every Car node request to the MEC fails The CPU and network effects of directly receiving requests from the taxis operating in region-1
- 79. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties Common execution for 3 minutes 79
- 80. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties Disconnection of the MEC in region-1 80 We observe a similar increase in the cloud CPU load and network traffic as in the previous period (75-110th interval)
- 81. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Scenario - Network Uncertainties In conclusion, with network alterations and fault injections, users: ● evaluate the execution of their services under extreme conditions ● identify unpredictable outcomes of imposed uncertainties to the service behavior 81
- 82. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Other Insights 82 Node profiling: insights are highly beneficial to engineers for capacity planning, optimizing service and resource placement. Mobility Patterns, Scaling Actions & Workload Changes: released insights about service performance and resources utilization. Application-level metrics: Operators can employ Fogify to produce and evaluate analytic insights, implementing adequate app-level metrics.
- 83. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Other Applications 83 Applications A Novel Middleware for Efficiently Implementing Complex Cloud-Native SLOs, T. Pusztai et al., CLOUD2021 EQUALITY: Quality-aware intensive analytics on the edge, A. Michailidou et al., Information Systems 2021 A study on speculative task scheduling for apache spark in fog computing realms, M. Symeonides et al., PCI2019 RAINBOW Project H2020 - A Fog Computing Platform (https://rainbow-h2020.eu/) Human-Robot Collaboration in Industrial Ecosystems (https://rainbow-h2020.eu/use-case-1/) by BIBA GmbH (http://www.biba.uni-bremen.de/) Kind: a tool for running local Kubernetes clusters using Docker container “nodes” (https://kind.sigs.k8s.io/)
- 84. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Conclusion ● We introduced an open-source Emulation suite for IoT, Edge and Fog applications. ○ features a powerful model specification for Fog and 5G Deployments, ○ allows the controllable Faults, Mobility patterns, and Alterations, ○ implements realistically 5G concepts like Slicing, MIMO, Beamforming, VNF, network monitoring, etc. ○ offers enablers for Any-scale Experimentation, and ○ provides Monitoring Capabilities ● We provided a detailed description of the implementation aspects, such as modeling, resource management and network shaping. ● Our future work includes: ○ Emulation of CPU architecture, GPUs, and low-level circuits ○ Emulation of embedded sensors as data generators 84
- 85. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Thank you!!!! 85 Fogify Github: https://github.com/UCY-LINC-LAB/fogify Fogify Documentation: https://ucy-linc-lab.github.io/fogify/ 5G-Slicer Github: https://github.com/UCY-LINC-LAB/5G-Slicer Fogify and 5G-Slicer are an open source projects. Fogify Demo: https://github.com/UCY-LINC-LAB/fogify-demo 5G-Slicer Demo: https://github.com/UCY-LINC-LAB/5G-Slicer-demo Video: https://www.youtube.com/watch?v=PthMM6rC89o
- 86. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Backup Slides 86
- 87. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing 87 Use-cases Single Trace Evaluation In this experiment, we show that the density of the RUs should be at least 100 in the selected bounding box to have fresh and in-time results Experiment Setup: ● Single randomly-selected bus trace (id=1230) ● Different number of RUs (25, 50, 75, 100, 125) in the bus operational area ● Monitor the IoT buffer size and count of timeouts. Network Traffic and IoT density The placement of the Edge service should be in line with the IoT devices density in order to have balanced workload computations Experiment Setup: ● Six MEC servers (Node 0-5) deployed on randomly selected RUs ● 20 IoT devices (buses) operated in the same city area However, during the execution, we realized that nodes 0, 1, and 4 did not receive any traffic so…
- 88. Moysis Symeonides msymeo03@cs.ucy.ac.cy Laboratory for Internet Computing Emulating 5G-Ready Mobile IoT Services 88 Use-cases Dublin’s bus traces sample, https://data.smartdublin.ie/dataset/dublin_bus_sample Dublin’s bus stops, https://data.gov.ie/dataset/bus-stops-served-by-dublin-bus M. Symeonides, Z. Georgiou, D. Trihinas, G. Pallis and M. D. Dikaiakos, "Demo: Emulating 5G-Ready Mobile IoT Services", IoTDI2022 We use the real-world open-access dataset from Dublin's bus traces[9], and bus stops[10] as IoT device traces, and locations for 5G URs, respectively. The datasets include: ● over 950 bus traces ● each bus periodically generates 16 metrics, including bus id, location coordinates, operating city region, etc ● over 4000 bus stops SlicerSDK and Bus Operator Use Case Template Application pipeline: ● When the bus (IoT device) is connected to an RU, it t transmits its data to the nearest MEC. MEC periodically computes region-based analytics and sends them to a Cloud server. If the bus is not connected to an RU, it uses a temporal buffer to store its data.