Slide 1

Slide 1 text

#ONESummit Edge Computing for Connected Cars: Use Case, Requirement and Architecture Toru Furusawa (Toyota) @twitter

Slide 2

Slide 2 text

#ONESummit Speaker Toru Furusawa Senior Researcher Connected Advanced Development Div. Toyota Motor Corporation

Slide 3

Slide 3 text

#ONESummit Agenda • Use Cases and Requirements of Connected Cars • Expectations and Challenges for Edge Computing • Our Approaches and PoC Experiments

Slide 4

Slide 4 text

#ONESummit Global Impact around 2025 100M units Connected Car 1-10 EB/month Communication Source: Our estimate based on PwC and SBD

Slide 5

Slide 5 text

#ONESummit Connected Services ITS Intelligent Transport System ü Local Danger Warning ü Collision Avoidance ü Cooperative Adaptive Cruise Control IVI In-Vehicle Infotainment ü Navigation ü Audio/TV ü Phone, Internet IoT Vehicle IoT ü Location Based Service ü Vehicle Quality Control ü High Definition Map ü Machine Learning

Slide 6

Slide 6 text

#ONESummit Extended Services Insurance and Finance ・Behavior-based Insurance ・Loan and Lease Telematics Mobility as a Service Automotive Services Smart Device Connectivity ・ Search ・Music ・Video ・Game ECU Driver monitoring Camera ・ Feedback to vehicle design R&D / Engineering Maintenance ・ Diagnostic Sales & Marketing Cloud Enhanced Vehicle Feature ・ Intelligent driving ・ High-Definition map ・ Cruise assist Extended Services +Public Data Processed data Open Data Anonymity Filtering AI Etc. Data Analysis ・Traffic Control ・Traffic Info ・Mobility Sharing ・Multimodal Data Collection ・Driving Assistance ・Remote Functionality ・Helpline Other Emerging Services ・ Failure Prediction

Slide 7

Slide 7 text

#ONESummit Diverse Service Requirements ITS IVI • P2P (sidelink) • Low latency & High reliability ☞ DSRC / Cellular V2X Safety • Cloud access (downlink) • High interactivity ☞ Cellular User Experience IoT • Cloud access (uplink) • Big data capacity • Leverage of latency allowance ☞ WLAN and Cellular Capacity

Slide 8

Slide 8 text

#ONESummit Capacity Issue • Regional business forecast around 2025 Enterprise Telco Real-time Non-real-time Core BIG DATA Transaction: 60PB/month Connected car: 3M units* * Assumption: 12% global share and 25% regional ratio Traffic: 20GB /(month*unit) LBS FOT HDMAP Situation Emotion Assumption: 10GB/sec 2 month+ just for storing data Source: Our estimate

Slide 9

Slide 9 text

#ONESummit Potential Solution Enterprise Telco Real-time Non-real-time core BIG DATA Issue: Big data Real-time Non-real-time core Solution: Distribution Distributed

Slide 10

Slide 10 text

#ONESummit Potential Solution edge Enterprise Telco Real-time Non-real-time Solution: Distribution core ③ Opportunistic Data Transfer ② Edge Computing ① Collecting Data in Need edge Load L M H Today’s Focus

Slide 11

Slide 11 text

#ONESummit Edge Computing for Connected Cars Public Cloud Devices Telco Edge Edge Cloud Source: 5GMF

Slide 12

Slide 12 text

#ONESummit Automotive Edge Computing Consortium Cross-industry ecosystem for next generation connected cars and mobility services Launched in January 2018 as a NPO in the U.S. AUTOMOTIVE MOBILE COMMUNICATION CLOUD & BIG DATA ANALYTICS APPLICATION & SERVICES Members as of September 2020

Slide 13

Slide 13 text

#ONESummit Concept of Automotive Edge Computing • Distributed Computing on Localized Network

Slide 14

Slide 14 text

#ONESummit Challenges in Realizing Automotive Edge Computing • Edge Infrastructure • Mobile Network Connectivity • Edge Cloud Infrastructure • etc... • Edge Application • Distributed Application Design • Application Deployment • Edge Resource Management • etc... Operators Interest Lots of Standardization and implementation (ex. Akraino) End Users Interest Not much implementation yet?

Slide 15

Slide 15 text

#ONESummit • Availability and scalability of each edge cloud is limited • Demand for communications and application processing between vehicles and edge cloud drastically changes by time and location • Significant delays and service outages may occur in the event of sudden local spikes in demands Challenges in Edge Cloud Application - Our Perspective

Slide 16

Slide 16 text

#ONESummit Possible Approaches • There are various possible approaches to the problem • In this talk, two approaches are presented • Simple implementation and experimental results to check the feasibility of the approaches (1) Scale-out to adjacent edge clouds (2) Vehicle data upload control

Slide 17

Slide 17 text

#ONESummit clustered Approach (1) Scale-out to Adjacent Edges • Idea • Clustering multiple edge clouds in the same region • Automatically scale-out to adjacent edge if a sudden local spike happens edge edge edge

Slide 18

Slide 18 text

#ONESummit PoC Demo Implementation • Kubernetes and Knative for Serverless Computing for Multiple Adjacent Edge Sites • Knative https://knative.dev/

Slide 19

Slide 19 text

#ONESummit Approach (2) Vehicle data upload control • Idea • Measure the number of vehicles connected to the edge cloud in real time • Adjust the amount of data sent from the vehicles according to the number of connections edge edge High data upload frequency High video resolution Low data upload frequency Low video resolution

Slide 20

Slide 20 text

#ONESummit PoC Demo Implementation UEs eNodeB SPGW-U SPGW-C Video Recognition App Data Upload Control App • Prepared mobile network (LTE) emulation environment to provide MEC • with COMAC (ONF’s open source project) https://www.opennetworking.org/reference-designs/comac/ • Two applications (data upload control app, video recognition app) deployed as MEC applications • Data upload control application • Video recognition app (just as an example) Central Office MME HSS COMAC software based emulation MEC Applications

Slide 21

Slide 21 text

#ONESummit Experiment Scenario UEs eNodeB SPGW-U SPGW-C Video Recognition App Data Upload Control App Central Office MME HSS (1) Notify # of current connected vehicles (2) Request vehicles to adjust upload data size by changing upload frequency and video resolution (3) Send recorded video from all vehicles to video recognition app

Slide 22

Slide 22 text

#ONESummit Experimental Results • The amount of traffic into the edge application (face recognition app) was measured when the number of vehicle connections was increased by one every 40 seconds • It was confirmed that the traffic into the edge app remained constant Note: This graph should be used as just one example only because the emulation software was unstable during the experiment, that might affect data.

Slide 23

Slide 23 text

#ONESummit What We Learned from 2 PoC Experiences • Open Source Accelerates Innovation! • In the past, we needed the help of carriers and telecom equipment vendors to conduct these kind of experiments • With the open networking technologies, end users like us can now build and experiment with mobile networks on our own

Slide 24

Slide 24 text

#ONESummit Summary • Use cases and challenges of connected cars • How edge computing is expected to be the technology to solve the challenges • Challenges in realizing automotive edge computing • Possible approaches and simple implementations

Slide 25

Slide 25 text

#ONESummit Thank you