Adaptive Computation Offloading for Mobile Edge Computing - PowerPoint PPT Presentation

Adaptive Computation Offloading for Mobile Edge Computing Environment Houssemeddine MAZOUZI Direction Nadjib ACHIR, Khaled BOUSSETTA L2TI, Institut Galilée, Université Paris 13 Journée MAGI Calcul scientifique 3 juillet 2018 1/21

Outline 1. Context 2. Mobile Edge Computing (MEC) 3. Computation offloading in MEC 4. Our offloading approach 5. Conclusion 2/21

Nowadays Mobile Devices 3/21

What is the problem? ⇒ how to extend the capacity of mobile device? User satisfaction on Galaxy S5. Rating system: (1) Very Dissatisfactory (5) Very Satisfactory [1] . [1] M. Halpern, Y. Zhu, and V. J. Reddi, “Mobile cpu’s rise to power: Quantifying the impact of generational mobile cpu design trends on performance, energy, and user satisfaction”, in High Performance Computer Architecture (HPCA), 2016 IEEE International Symposium on , IEEE, 2016, pp. 64–76 4/21

What is the problem? ⇒ how to extend the capacity of mobile device? User satisfaction on Galaxy S5. Rating system: (1) Very Dissatisfactory (5) Very Satisfactory [1] . [1] M. Halpern, Y. Zhu, and V. J. Reddi, “Mobile cpu’s rise to power: Quantifying the impact of generational mobile cpu design trends on performance, energy, and user satisfaction”, in High Performance Computer Architecture (HPCA), 2016 IEEE International Symposium on , IEEE, 2016, pp. 64–76 4/21

The new emerging computing paradigm: Mobile Cloud Computing ◮ End-to-end network latency to the closest AWS data center using wired network 20-30 ms, up to 50-150 ms on 4G mobile network. ◮ Ambiant occlusion requires end-to-end delays under 20 ms !!!!! ◮ What the hell !! Even the cloud is not enough! 5/21

The new emerging computing paradigm: Mobile Cloud Computing ◮ End-to-end network latency to the closest AWS data center using wired network 20-30 ms, up to 50-150 ms on 4G mobile network. ◮ Ambiant occlusion requires end-to-end delays under 20 ms !!!!! ◮ What the hell !! Even the cloud is not enough! 5/21

The new emerging computing paradigm: Mobile Cloud Computing ◮ End-to-end network latency to the closest AWS data center using wired network 20-30 ms, up to 50-150 ms on 4G mobile network. ◮ Ambiant occlusion requires end-to-end delays under 20 ms !!!!! ◮ What the hell !! Even the cloud is not enough! 5/21

The new emerging computing paradigm: extension Mobile Edge Computing Environment Cloud Virtual Reality World ⇒ Ultra-low latency. Internet Operator Network Edge Node ⇒ Small capacity . Network Latency Reduced Latency through Mobile Edge Computing 6/21

The new emerging computing paradigm: extension Mobile Edge Computing Environment Cloud Virtual Reality World ⇒ Ultra-low latency. Internet Operator Network Edge Node ⇒ Small capacity . Network Latency Reduced Latency through Mobile Edge Computing 6/21

The new emerging computing paradigm: MEC Challenges 1. Placement of the Edge Server (cloudlet) in the network 2. Selection of the Edge Server for whom a user offloads its computation 3. Model of the mobile application: define the offloadable parts, offloading condition, virtualization technology 4. Computing resource allocation at the edge server 5. Bandwidth allocation 7/21

The new emerging computing paradigm: MEC Challenges 1. Placement of the Edge Server (cloudlet) in the network 2. Selection of the Edge Server for whom a user offloads its computation 3. Model of the mobile application: define the offloadable parts, offloading condition, virtualization technology 4. Computing resource allocation at the edge server 5. Bandwidth allocation 7/21

Computation offloading: model of the application The app computation requirement Local part Remote part (task) Determine the remote part: Dependencies: data, ⇒ At the design time: parameters, ... The amount of the computation to offload static offloading decision app Mobile Device Edge Server ⇒ At the runtime: Task transmitted from dynamic offloading to mobile device decision app Remote part (task) 8/21

Computation offloading: model of the application The app computation requirement Local part Remote part (task) Determine the remote part: Dependencies: data, ⇒ At the design time: parameters, ... The amount of the computation to offload static offloading decision app Mobile Device Edge Server ⇒ At the runtime: Task transmitted from dynamic offloading to mobile device decision app Remote part (task) 8/21

Large MEC: Computation offloading edge server A c c e s s P o i n t ( W i F i ) 9/21

Large MEC: Computation offloading Edge server Static o oading decision app Dynamic O oading decision app A c c e s s P o i n t ( W i F i ) 9/21

Large MEC: Computation offloading Which user should o oad? How much computation? And to which edge server? A c c e s s P o i n t ( W i F i ) 9/21

Our Offloading Policy ⇒ Goal: Determine which user should offload, select an edge server and the amount of the computation to offload. ◮ Allocate the bandwidth to each user. ◮ minimize the offloading cost: cost = β ∗ Energy + (1 − β ) ∗ Time ◮ assumptions: ⇒ For static offloading decision: a u m , n = 1, the whole computation must be offloaded to MEC. ⇒ For dynamic offloading decision: a u m , n ∈ [0 , 1], we must find its optimal value. 10/21

Problem Formulation: multi-user multi-edge server offloading Minimize � M � N m Z u m , n m n C 1 : � K k =1 x u m , n , k ≤ 1 , ∀ m ∈ M , u m , n ∈ N m ⇒ Each task can be offload to at most one Edge server C 2 : y u m , n − � K k =1 x u m , n , k ≤ 0 , ∀ m ∈ M , u m , n ∈ N m ⇒ Static offloading Decision app must be offloaded C 3 : T u m , n ≤ t u m , n , ∀ m ∈ M , u m , n ∈ N m ⇒ QoS constraint C 4 : x u m , n , k ≤ g u m , n , k , ∀ m ∈ M , u m , n ∈ N m , k ∈ K ⇒ Edge server support Constraint C 5 : � M m ( � N m x u m , n , k ∗ c k ) ≤ F k , ∀ k ∈ K ⇒ Edge server capacity n C 6 : x u m , n , k ∈ { 0 , 1 } , ∀ m ∈ M , u m , n ∈ N m , k ∈ K C 7 : a u m , n ∈ [0 , 1] , a u m , n ≥ y u m , n , ∀ m ∈ M , u m , n ∈ N m This problem is NP-hard . 11/21

Problem Formulation: multi-user multi-edge server offloading Minimize � M � N m Z u m , n m n C 1 : � K k =1 x u m , n , k ≤ 1 , ∀ m ∈ M , u m , n ∈ N m ⇒ Each task can be offload to at most one Edge server C 2 : y u m , n − � K k =1 x u m , n , k ≤ 0 , ∀ m ∈ M , u m , n ∈ N m ⇒ Static offloading Decision app must be offloaded C 3 : T u m , n ≤ t u m , n , ∀ m ∈ M , u m , n ∈ N m ⇒ QoS constraint C 4 : x u m , n , k ≤ g u m , n , k , ∀ m ∈ M , u m , n ∈ N m , k ∈ K ⇒ Edge server support Constraint C 5 : � M m ( � N m x u m , n , k ∗ c k ) ≤ F k , ∀ k ∈ K ⇒ Edge server capacity n C 6 : x u m , n , k ∈ { 0 , 1 } , ∀ m ∈ M , u m , n ∈ N m , k ∈ K C 7 : a u m , n ∈ [0 , 1] , a u m , n ≥ y u m , n , ∀ m ∈ M , u m , n ∈ N m This problem is NP-hard . 11/21

Our proposal: DM2-ECOP algorithm 12/21

Our proposal: DM2-ECOP algorithm 12/21

Our proposal: DM2-ECOP algorithm MEC Computation Offloading manager Lagrangian Lagrangian multipliers multipliers subproblem M: subproblem 1: offloading decision offloading decision and cloudlet selection and cloudlet selection Local offloading manager 1 Local offloading manager M Local Offloading Requests Local Offloading Requests 12/21

DM2-ECOP: Local offloading manager 1- Estimate the bandwidth allocation to each user using Bianchi model: w u m , n = B m ( π m ) π m ◮ B m : is the estimated bandwidth at the AP m ◮ π m : is the number of users that offload 2- For each Static offloading decision task, select the cloudlet that minimizes Z e u m , n , k + λ k c k . 13/21

DM2-ECOP: Local offloading manager 3- For each Dynamic offloading decision task, compute the offloading priority: ξ u m , n = Z l k ∈K ( Z e u m , n − min u m , n , k ); under a u m , n = 1 4- Sort dynamic offloading decision apps in decreasing order of ξ u m , n 5- Select the cloudlet k that minimizes Z e u m , n , k + λ k c k 6- Compute the optimal value of a u m , n 7- when the offloaded task is equal to π m , all the remaining apps will be performed locally 14/21

DM2-ECOP: find the optimal amount of computation to offload ⇒ For each user, the optimal a u m , n is the solution of: min ( Z e u m , n , k + Z l u m , n ) Subject to: a u m , n ∈ [0 , 1] . ⇒ the optimal value of a u m , n is 1 if and only if : ψ u m , n < µ u m , n ⇒ Where: ◮ ψ u m , n = up u m , n γ u m , n w u m , n · [ κ · f 3 u m , n · c k · β u m , n + (1 − β u m , n ) · ( c k − f u m , n ) − β u m , n · P idle u m , n · f u m , n ] ◮ µ u m , n = c k · f u m , n · ( P tx / rx u m , n · β u m , n + 1 − β u m , n ) 15/21

DM2-ECOP: find the optimal amount of computation to offload ⇒ For each user, the optimal a u m , n is the solution of: min ( Z e u m , n , k + Z l u m , n ) Subject to: a u m , n ∈ [0 , 1] . ⇒ the optimal value of a u m , n is 1 if and only if : ψ u m , n < µ u m , n ⇒ Where: ◮ ψ u m , n = up u m , n γ u m , n w u m , n · [ κ · f 3 u m , n · c k · β u m , n + (1 − β u m , n ) · ( c k − f u m , n ) − β u m , n · P idle u m , n · f u m , n ] ◮ µ u m , n = c k · f u m , n · ( P tx / rx u m , n · β u m , n + 1 − β u m , n ) 15/21