Offloading: Where and When

Transcript

1. Offloading: Where and When Iwan Setiawan <stwn at unsoed.ac.id>
2. Iwan Setiawan <stwn at unsoed.ac.id> Categories of Edge Nodes[1] • Cloudlets • Micro data centers • Base stations • Other types of edge nodes – Vehicles with onboard computers – IoT gateways – Smartphones Lin, et al., 2019
3. Iwan Setiawan <stwn at unsoed.ac.id> Categories of Edge Nodes[1] • Cloudlets • Micro data centers • Base stations • Other types of edge nodes – Vehicles with onboard computers – IoT gateways – Smartphones Lin, et al., 2019 Fog MEC
4. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Why • Limited computing resources • Energy consumption • Cost? Time Energy Offload or not to offload Money
5. Iwan Setiawan <stwn at unsoed.ac.id> Wu, 2018
6. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Major Challenges • Resource, related to resources from device to cloud • Component, related to program complexity • Intermittence, related to network connectivity • Data, related to the amounts of data – Input data, when the task is offloaded to the cloud – Communication data, between device and remote cloud – Output data, generated from remote execution
7. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] • Target: Cloud service selection +edge and fog • Criteria – Performance: Speed, accuracy, response time – Cost: virtual machine, data, storage, communication – Security: Data integrity, privacy, loss – Bandwidth – Availability
8. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] • Multi-criteria decision making – Matching: finding a matched (cloud) services – Ranking: evaluate (cloud) services, qualitative/quantitative – Selecting: choosing the optimal (cloud) service • Cloudlet-based decision making[2][3] – Two-level offloading: offload to a cloud – Three-level offloading: offload to nearby cloudlet (edge) • Hybrid offloading decision making – Using mobile edge and cloud computing – Taking advantages of both (cloud-edge) (Edge)
9. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] Wu, 2018
10. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] • Multi-criteria decision making – Matching: finding a matched (cloud) services – Ranking: evaluate (cloud) services, qualitative/quantitative – Selecting: choosing the optimal (cloud) service • Cloudlet-based decision making[2][3] – Two-level offloading: offload to a cloud – Three-level offloading: offload to nearby cloudlet (edge) • Hybrid offloading decision making – Using mobile edge and cloud computing – Taking advantages of both (cloud-edge) (Edge)
11. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading – Horizontal offloading Cloud Edge Device 1 2 3 3 1 3 1 2
12. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Direction – Vertical offloading: to higher tier – Horizontal offloading Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2
13. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2
14. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2 Limited resources Limited capacity Limited energy
15. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2 Limited resources Limited capacity Limited energy Compute Storage Network
16. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2
17. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Device 1 2 3 D1 D2 E1 E2 C1 C2
18. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where • Two-tier model – Cloud – Device • Three-tier model – Cloud – Edge – Device Fog • Directions – Vertical offloading: to higher tier, to lower tier? – Horizontal offloading: same tier, federation Cloud Edge Fog 1 2 3 F1 F2 E1 E2 C1 C2 D1 D2 D3
19. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] • Multi-criteria decision making – Matching: finding a matched (cloud) services – Ranking: evaluate (cloud) services, qualitative/quantitative – Selecting: choosing the optimal (cloud) service • Cloudlet-based decision making[2][3] – Two-level offloading: offload to a cloud – Three-level offloading: offload to nearby cloudlet (edge) • Hybrid offloading decision making – Using mobile edge and cloud computing – Taking advantages of both (cloud-edge) (Edge)
20. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: Where[2] Wu, 2018 Latency Capacity Cloud Assisted Mobile Edge
21. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] • Computation vs communication – Time-saving: the computation on cloud better than local – Energy-saving: the energy needs to be less when offload • Interrupted vs uninterrupted – Connection failures: depend on the connection to the cloud – Cloud failures: depend on the cloud reliability • On-the-spot vs delayed offloading – On-the-spot: any wireless network available – Delayed: waiting for a high-quality network
22. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] • Time-saving tm > tm /F + D/B • Energy-saving pm tm > pidle (tm /F) + ptr (D/B) Wu, 2018 tm = Computation time on device F = Speedup factor D = Data exchanged B = Network bandwidth pm = Power on device when compute pidle = Power on device when idle ptr = Power when transfering data
23. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] • Computation vs communication – Time-saving: the computation on cloud better than local – Energy-saving: the energy needs to be less when offload • Interrupted vs uninterrupted – Connection failures: depend on the connection to the cloud – Cloud failures: depend on the cloud reliability • On-the-spot vs delayed offloading – On-the-spot: any wireless network available – Delayed: waiting for a high-quality network
24. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] Wu, 2018
25. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] • Computation vs communication – Time-saving: the computation on cloud better than local – Energy-saving: the energy needs to be less when offload • Interrupted vs uninterrupted – Connection failures: depend on the connection to the cloud – Cloud failures: depend on the cloud reliability • On-the-spot vs delayed offloading – On-the-spot: any wireless network available – Delayed: waiting for a high-quality network
26. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: When[2] Wu, 2018 Energy cost Data queue backlog Estimated rate
27. Iwan Setiawan <stwn at unsoed.ac.id> Offloading: How[2] Wu, 2018 λ = rate of offloadable jobs λ0 = rate of unoffloadable jobs π = probability to offload α = job dispatcher
28. Iwan Setiawan <stwn at unsoed.ac.id> Offloading System Architecture[2][4] Cloud-edge-fog discovery Metrics Profiling Offloading decision Querying across cloud-edge-fog Where Program partitioning What When Offloading Device Remote partition Program Energy Network Time Energy ... Cloud-edge-fog Remote partition Cloud-edge-fog federation Network +Data Why Query service Profiling Remote execution Metrics Energy Storage Cost Perform. How Matching Selecting Ranking
29. Iwan Setiawan <stwn at unsoed.ac.id> References [1] L. Lin, X. Liao, H. Jin and P. Li, "Computation offloading toward edge computing," in Proceedings of the IEEE, vol. 107, no. 8, pp. 1584- 1607, Aug. 2019. [2] H. Wu, "Multi-objective decision-making for mobile cloud offloading: a survey," in IEEE Access, vol. 6, pp. 3962-3976, 2018. [3] J. Wang, J. Pan, F. Esposito, P. Calyam, Z. Yang, and P. Mohapatra, “Edge cloud offloading algorithms: issues, methods, and perspectives,” in ACM Comput. Surv., vol. 52, no. 1, Feb. 2019. [4] K. Yang, S. Ou and H. Chen, "On effective offloading services for resource-constrained mobile devices running heavier mobile Internet applications," in IEEE Commun. Magazine, vol. 46, no. 1, pp. 56-63, Jan. 2008.
30. Iwan Setiawan <stwn at unsoed.ac.id> Offloading Decisions • Where question related to the right place for offloading • When question related to the right time for offloading