Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

[D2 CAMPUS] tech meet up(Back-end) - Latency Reduction between mobile users and cloud/edge servers (김명종님)

300 views

Published on

tech meet up (주제 : Back-end) 참가자 발표 자료 입니다.

Published in: Engineering
  • Be the first to comment

  • Be the first to like this

[D2 CAMPUS] tech meet up(Back-end) - Latency Reduction between mobile users and cloud/edge servers (김명종님)

  1. 1. Latency Reduction between Mobile Users and Cloud/Edge Servers or My Journey into Serverless Computing 김명종
  2. 2. Latency Reduction between Mobile Users and Cloud/Edge Servers or My Journey into Serverless Computing 김명종
  3. 3. 1. Preface 2. Research goal 3. Related work 4. Implementation 5. Result
  4. 4. 1. Preface
  5. 5. What: 연세대학교 컴퓨터과학과 졸업 프로젝트
  6. 6. Who: 정재호, 신사무엘, 김명종 (연세대학교 컴퓨터과학과 학부생)
  7. 7. When: 2017년 8월 ~ 2017년 12월
  8. 8. 2. Research goal: Latency reduction between mobile users and cloud/edge servers
  9. 9. - Low processing power - Limited battery life
  10. 10. - Workloads can be “offloaded” to the cloud - Optimized for CPU/GPU-intensive workloads
  11. 11. - Implement techniques to reduce workload offloading latency between mobile devices and cloud/edge servers - Define appropriate situations for different workload offloading techniques Goals:
  12. 12. 3. Related work
  13. 13. COSMOS: Computation Offloading as a Service for Mobile Devices (ACM, 2014.) • Cloud offloading • AWS EC2 instance • Android x86 image • Uses execution predictor for offloading decision • Mantis: Automatic performance prediction for smartphone applications (USENIX, 2013.) Shi, Cong, et al. "Cosmos: computation offloading as a service for mobile devices." Proceedings of the 15th ACM international symposium on Mobile ad hoc networking and computing. ACM, 2014.
  14. 14. Content-Based Scheduling of Virtual Machines (VMs) in the Cloud (IEEE, 2013.) • Similar VM images have many identical disk blocks • Similarity of Ubuntu 12.04 and Ubuntu 11.10 = 60% • Minimize VM provisioning overhead when creating multiple VMs • Schedule workload process to minimize network traffic Bazarbayev, Sobir, et al. "Content-based scheduling of virtual machines (VMs) in the cloud." Distributed Computing Systems (ICDCS), 2013 IEEE 33rd International Conference on. IEEE, 2013.
  15. 15. Just-in-Time Provisioning for Cyber Foraging (ACM, 2013.) • Reduce cloudlet VM provisioning overhead • Deduplication: only transfer overlay information to base VM • Identify and ignore “garbage” disk blocks • Communicate free memory information between VM and VM manager • VM synthesis pipelining Ha, Kiryong, et al. "Just-in-time provisioning for cyber foraging." Proceeding of the 11th annual international conference on Mobile systems, applications, and services. ACM, 2013.
  16. 16. Cloudlet-Based Cyber-Foraging for Mobile Systems in Resource-Constrained Edge Environments (ACM, 2014.) • Introduce multiple cloudlet provisioning strategies • Optimized VM synthesis • Application virtualization • Cached VM • Cloudlet push • On-demand VM provisioning Lewis, Grace A., et al. "Cloudlet-based cyber-foraging for mobile systems in resource-constrained edge environments." Companion Proceedings of the 36th International Conference on Software Engineering. ACM, 2014.
  17. 17. 4. Implementation
  18. 18. Cloudy
  19. 19. 4.1. Simulations
  20. 20. 4.2. Cloud/edge servers
  21. 21. 4.3. Mobile application
  22. 22. 4.4. Latency reduction techniques
  23. 23. Techniques: - Virtualization level - Custom scheduling algorithm - Hop-count minimization
  24. 24. OS-level virtualization vs. Full virtualization
  25. 25. OS-level virtualization vs. Full virtualization
  26. 26. Custom scheduling algorithm
  27. 27. Custom scheduling algorithm
  28. 28. Hop-count minimization
  29. 29. 5. Result
  30. 30. Real-world scenario: 225,702ms Theoretical scenario: 265,400ms Cloudy techniques: 168,043ms Efficiency improvement: 25.6% (real-world) Efficiency improvement: 36.8% (theoretical)
  31. 31. Advantages: - Better performance - Modular design - Scalability - Adaptability
  32. 32. Disadvantages: - Requires server components - Requires additional programming when adding tasks - Less predictable performance
  33. 33. Reflections: - Serverless/edge - Cold boot, keep-alive - Jsch, ntp, cli - Support for multiple platforms
  34. 34. Q&A

×