1 Introduction
Kubernetes
, it will be easy to manage services (e.g. the data pre-processing services) in the MEC environment. However, these advantages cannot be the excuse of the carelessness in planning the multi-source AIoT sensing and analysing tasks — if the related services are not assigned to appropriate hosts, it may even obtain lower-quality result with much higher cost. More critically, as the edge servers are all resource-constrained [7, 8] and energy-consuming [9‐12], there would be no enough resources for them to run if the data pre-processing services are not deployed on appropriate edge servers. Thus, it becomes more and more important to design a service deployment scheme as well as a resource allocation scheme to balance the quality and cost. The main contributions are summarized as follows: 2 Motivation scenario
3 Related work
3.1 Service placement in MEC
3.2 Resource allocation in MEC
Symbols | The physical meaning of the notations |
---|---|
H | the set of edge server, |H| = n |
\(h_j\) | the j-th server in H |
\(U_j\) | the set of IoT devices in the serving area of \(h_j\) |
\(v^e_j\) | the average transmission rate between \(h_j\) and devices in \(U_j\) |
\(v^c_j\) | the average transmission rate between \(h_j\) and the cloud |
\(b_{j,k}\) | the average transmission rate between \(h_j\) and \(h_k\) |
\(\mu ^{\star }_j\) | the available computing resource of \(h_j\) |
\({S}^{\mathbb {R}}\) | the service set of the ECC system, \(|{S}^{\mathbb {R}}|\) = m |
\(s_i\) | the i-th service in \(\mathcal {S}\) |
\({S}^{\mathbb {V}}\) | the virtual services that collect context-aware data around different edge servers, \(|{S}^{\mathbb {V}}|\) = n |
\(S = S^{\mathbb {R}}\cup S^{\mathbb {V}}\) | the set of real services and virtual services |
\(c_i\) | the i-th service in \({S}^{\mathbb {V}}\) |
\({I}_{i}\) | the average input data size of \(s_i\) |
\({O}_{i}\) | the average output data size of \(s_i\) |
\(w_{i}\) | the average workload of \(s_i\) |
\(\mu ^{k}_{j,i}\) | the resource that \(h_j\) allocates to \(s_i\) |
\(\varvec{G}\) | the AIoT application set of the system, \(|\varvec{G}|\) = K |
\(G_k = (S_k, E_k)\) | the k-th AIoT application in \(|\varvec{G}|\) |
\(p^{k}_i\) | the edge server index where the service \(s_i\) in AIoT application \(G_k\) is placed on |
\(\mathcal {F}_k(s_i)\) | the precursor set of service \(s_i\) in AIoT application \(G_k\) |
\(\eta _{j}\) | the energy conversion rate of \(h_j\) |
4 System model and problem description
4.1 Server and network
4.2 DAG-based AIoT application
4.3 AIoT application deployment scheme
4.4 AIoT application performance evaluation
4.5 Energy consumption model
4.6 Problem definition and formulation
5 Approach
6 Experiments and analysis
Param | Value | Param | Value |
---|---|---|---|
(m, n) | (9, 10) | K | 8 |
\(d^{O}_{i}\) | \(\mathcal {N}\)(5,\(1^2\)) MB | \(v^e_{j}\) | \(\mathcal {N}\)(20,\(10^2\)) MB/s |
\(w_{i}\) | \(\mathcal {N}\)(20,\(10^2\))\(\times 10^4\) MI | \(B_{j,k}\) | \(\mathcal {N}\)(20,\(10^2)\) kMB/s |
\(\eta _j\) | \(\mathcal {N}\)(10,\(1^2\)) | \(\mu _{j}\) | \(\mathcal {N}\)(2,\(0.2^2\))\(\times 10^4\) MIPS |