# Mobile Computing Workshop 2017

[](http://creativecommons.org/licenses/by-nc-sa/4.0/)


###### tags: `workshop`

---
[TOC]

---
## 1. Exploring Network S & V by Applying SDN/NFV to 5G and IoT

> 講者：國立交通大學電子與資訊研究中心主任-林寶樹教授
> 地點：行政大樓-國際會議廳

### 1.1 Major Trends of 5G and IoT

* Open Networking
* SDN/NFV
* Disaggregation

### 1.2 Open Networking

* ==SDN + NFV + Cloud Open Source==
* SDN Controller
* NFV Orchestrator
* Cloud OS
* SDN Layer System View
* Keywords
* **P4 (Programming Protocol-Independent Packet Processors)**
* High-level programming language
* **ONOS (Open Network Operating System)**
* **CORD (Central Office Re-architected as a Datacenter)**
* Network architecture based on SDN
* **Key Elements: P4 and CORD**
* P4 an important processor of SDN switch programmability.
* P4 as an SDN programming language.

### 1.3 SDN/NFV 5G Services

* SDN/NFV Base Services
* Above SDN Controller.
* Basic CORD Concept
* ==**CORD = SDN + NFV + Cloud**==

### 1.4 Why 5G & IoT?

* SDN, NFV, and mm-wave claimed as three key ingredients for 5G by Domain 2.0.
* Three apporaches for S&V for 5G
* Radio Access Network (RAN) approach
* Virtualized RAN
* Core Network(CN) approach
* Virtualized EPC
* System-level approach
* **Prototype of Open 5G Core**
* The relationships between IoT and 5G
* Dynamic mobility management
* Be sliced into many virtualized networks
* NGMN purposes three categories of network slicing
* Different network slicing provides different network applications.

### 1.5 Challenges

* Performance
* Software approach vs. Hardware approach
* Data rate transmission (throughput)
* Latency
* Load balance (Traffic flow)
* Interoperability
* Choice of SDN Controller OS
* ONOS
* ODL (OpenDayLight)
* RYU
* Others
* Deployment of various SDN switches.
* Depolyment
* Provide operators configence in SDN/NFV solutions
* Depolyment in the global scale
* Integration
* Vertical Integrated Stack as SDN Distribution
* Peering Router
* ONOS
* ToR
* OCP
* OF-DPA
* Broadcom
* Incompatible
* Open source development method.
* Short of Talents
* Collaboration with topnotch open networking software leaders.

---
## 2. 5G and IoT Make better Smart Cities

> 講者：臺灣聯合大學系統副校長-林一平教授
> 地點：行政大樓-國際會議廳

### 2.1 Mobile is Everything

* 5G is more than a faster 4G, and application is the main driving force of 5G.
* 因應頻寬需求，演進下一代通訊技術

### 2.2 What is Network Slicing?

* Smartphones
* Autonomous driving
* Massive IoT
* Other slices


### 2.3 LWA (Local Web Authentication) Services

### 2.4 Massive MTC

* 3GPP Solution: NB-IoT
* Enhance Coverage
* Low-cost Devices
* Low Power Consumption
* Massive Connection
* LoRa 未來將大量用於密閉空間中

---
## 3. Mutlicell Clustering in Green Heterogeneous Wireless Networks

> 講者：科技部工程科技發展司司長-廖婉君教授
> 地點：行政大樓-國際會議廳

### 3.1 Power Consumption Breakdown

* BSs and their power amplifier are the most dominant ones.
* About 1/3 from a single factor
* Two Basic Energy-Saving Approaches
* BS Sleeping Control
* Operational powers
* Traffic Offloading (to small cells)
* Save transmit powers
* Is Network Energy Efficiency Important?
* The largest technlogoical gap is in the network energy efficiency.
* Massive machine type communications
* Ultra-reliable and low latency communications

### 3.2 Green Multi-cell Clustering in HetNets

* Heterogeneous Network (HetNets)
* Green Multicell Clustering = BS Sleeping Control + CoMP Clustering

### 3.3 Coordinated Multipoint (CoMP) Clustering

* CoMP Joint Transmission
* **How can we from clusters in an energy-efficient manner?**

### 3.4 Literature on Green Multi-cell Clustering

* The analogy to facility location problems
* Each UE will be satisfactory as long as it is served by any one active BS.
* Facility Location Problem (FLP) is a well-known NP-hard Problem.

### 3.5 Three Phases of This Talk

* Phase-1: (GMC) Green Multicell Clustering in HetNets
* Phase-2: GMC with Independent Caching
* Phase-3: GMC with Sharable Caching

### 3.6 Power Consumption Model

$$
p^{\rm{TOT}}=p^{\rm{BSC}}(y_1) + p^{\rm{MCC}}(x_{ij})+p^{\rm{CST}}
$$

* TOT: Total Power Consumption
* BSC: BS Sleeping Control
* MCC: Multicell clustering
* CST: Constant Consumption

$$
p^{\rm{TOT}}=p^{\rm{BSC}}(y_1) + p^{\rm{MCC}}(x_{ij}) + p^{\rm{BH}}(y_{i} ,z_{fi}) +p^{\rm{CST}}
$$

### 3.7 A Sketch of the Phase-1 Problem

* We consider a "simple" optimization problem to characterize green multicell clustering in HetNets.
* In fact, it is not "simple" at all.

### 3.8 How Can We Solve It Efficiency?

* Local-Search Based Multi-cell Clustering (LSMC)
* The algorithm requires an initial solution for local search.
* Form clusters for each UE simply based on the channel coefficient.
* Local search begins, Keep finding improving local moves.

### 3.9 Cache-Enabled Green Multi-cell Clustering

* To alleviate the induced backhaul power consumption caused by fetching contents remotely, caching popular contents at local storages is regarded as a viable solution.

---
## 4. 科技部資訊工程/智慧計算學門業務說明

> 講者：國立交通大學資訊工程學系-陳志成教授 / 國立中央大學資訊工程學系-范國清教授
> 地點：行政大樓-國際會議廳

### 4.1 資訊學門的業務

* 科技部簡介
* 綜合規劃司
* 自然科學及永續研究發展司
* 工程技術研究發展司
* 生命科學研究發展司
* 人文及社會科學研究發展司
* 科教發展及國際合作司
* 前瞻及應用科技司
* 產學及園區業務司
* 學門研究領域 (三年一換複審委員、審查委員四選二)
* 計算機結構與計算機系統
* 計算機網路與網際網路
* 平行與分散處理
* 程式語言與軟體工程
* 計算機理論與演算法
* 資訊安全
* 申請案類別
* 一般型研究計畫-專題計畫
* 隨到隨審計畫
* 新進人員計畫
* 延攬人才及兩岸交流
* 邀請科技人士短期訪問
* 補助研究生出席國際會議
* 國內成果發表會及研討會
* 鼓勵申請整合型計畫
* 鼓勵參與需求導向之專案計畫
* 專案計畫
* **資訊學門計畫成果發表會**
* 記得申請學門
* ==一定要來喔！==

### 4.2 我對計劃提案之觀察

* 常犯錯誤
* 題目無創新、過時
* **取自己發表過的文章**
* 內容重點不清楚
* 計畫執行方法內容不清楚
* **中英文夾雜**
* **無頁碼**
* 圖片、數學式字體不一
* **主持人專長不符**
* 可能選項
* 學門
* 主題
* 主持人/共同主持人
* 單年/**多年期(推薦)**
* 優先順序
* 經費額度(合理性)
* 申請輔助之設備
* 計畫應強調之事項
* 可行性 (***Hope is not strategy***)
* 可能方向
* 初步成果
* 清楚定義問題
* 學會寫「本計劃貢獻」
* 多年期計畫

---
## 5. DRX Scheduling for Cloud Radio Access Networks (C-RAN) with Dynamic Point Selection

### 5.1 Introduction

* **C-RAN (Cloud Radio Access Networks; Cloud-RAN)**
- 將計算元件 BBU (Baseband Unit) 從基地台 (Base Stations; BSs) 分離並放置於集中式的雲端網路。
- 使用者裝置 (User Equipments; UEs) 透過 C-RAN 可以從多個協同運作的 cell 接收資料，並利用動態擇點 (Dynamic Point Selection; DPS) 技術提升網路效率。
* **DRX (Discontinuous Reception; DRX)**
- 3GPP LTE-A 定義非連續接收機制，允許 UE 週期性地關閉無線介面並進入睡眠模式。
- DRX 機制是被設計來節省 UE 的能耗，其參數會根據 UE 設定並由 eNB 決定。
- 一個基本時間單位表示一個子框 (subframe)，長度為 1 毫秒。
- DRX 參數定義
- DRX-Cycle-Length 是 UE 從 eNB 接收資料的週期。
- DRX-Start-Offset 是 DRX-Cycle-Length 的起始子框。
- On-Duration 是 UE 必須清醒並偵測 eNB 是否有資料傳進來的時間。
- 如果有接收到資料，UE 會啟動 Inactivity-Timer 並在其終止前保持清醒。
- 如果有接收到資料，則重設 Inactivity-Timer。
- 當 Inactivity-Timer 終止，UE 進入睡眠並關閉無線介面。
- 在 UE 睡眠的期間，任何要傳進來的資料會先貯存在 eNB 直到下一個 On-Duration 到來。
* **Quality-of-Service (QoS)**
- 探討 C-RAN 的資源分配問題，根據 DRX 機制節能來最佳化 UE，有效利用頻寬。
- 實作 DPS 的 C-RAN 架構下 UE 的服務品質
- 節能的 DPS 排程方法

### 5.2 Problem Definition

* 如何透過選擇每個 UE$_i$ 適當的 DRX 參數，包含 DRX-Cycle-Length ($T_i$) 、DRX-Start-Offset ($Z_i$)、On-Duration ($O_i$) 與 Inactivity-Timer ($I_i$)，[电邮支援](https://www.emailhelpdesk.us/) $M$ 個 Cell 服務 $N$ 個 UE 的資源排程，以**讓愈多 UE 的服務品質可被滿足**。
- 延遲限制 ($D_i$)
- 預期的資料傳輸速率 ($R_i$)
* **==最小化 UE 的清醒子框的個數==**

### 5.3 Solution

* **==可協調 RRH 的節能 DPS 排程演算法==**
- 讓所有 UE 的 DRX-Cycle-Length 是其他 UE 的整數倍，以減少因資源競爭產生的非必要清醒時間。
- 安排傳輸順序，以減少 UE 必須清醒的時間
- 最小化預期的清醒時間比例來最佳化 On-Duration 與 Inactivity-Timer。
* 第一階段：**決定 DRX-Cycle-Length ($T_i$)**
* 第二階段：**決定 Data Schedules 與 DRX-Start-Offset ($Z_i$)**
* 第三階段：**最佳化 DRX 參數 ($O_i$ 與 $I_i$)**

---
## 6. Resource-Aware Consistent Updates in Software-Defined Networking

### 6.1 Introduction

- **Software-defined networking (SDN)** brings more reliable and flexible consistent network update mechanisms.
- Making the update process independent of the underlying routing protocols
- ==When the network resource is deficient, the resource competition phenomenon may occur, which prolongs the total update completion time.==

### 6.2 Problem Definition

- Based on traffic delivery constraint and resource constraint, we formulate the resource competition problem as an update time minimization problem.
- Proposing a greedy-heuristic algorithm to minimize the total update completion time
- **==The Update Time Minimization Problem==**
- Input instance: Consider a directed graph $G = (V, E)$. The maximum capacity of each link is $L_{xy}$. There is a set of network states $S$ from initial state to final state.
- Objective: The output is a sequence modification operations $O^k$ to minimize the total update time $T$ from initial state to final state. The objective function is represented as follows.
- Minimize $T = \sum\limits_{k = 1}^{|S| - 1}(T^k)$

### 6.3 Solution - ==Efficient Consistent Network Update==

- Phase 1 - **Detection of Resource Competition**
- Create a competitive graph to describe the competition relationship between flows
- Nodes are flows
- Edges are the competitive relationship between flows
- The initial resources of each flows are preoccupied.
- The requiring resources are also accumulated.
- If the total traffic demand on each resource exceeds the maximum available capacity, all flows requiring this resource will have an edge connecting to all flows releasing this resource.
- Phase 2 - **Flow Picking**
- Pick some flows in the competitive graph, and change the update plan for these flows from "initial path to final path” to “initial path to alternative path” and “alternative path to final path”.
- The path with length longer than or equal to 1 as Chain.
- If there does not exist any Chain in the graph, then no resource competition exists.
- How to pick some of the flows which break all existing Chains?
- **==Greedy-Heuristic Method==**
- Assume 99 flows requiring 1 unit and 1 flow requiring 100 unit.
- If the total available capacity for this resource is 100, then all 100 flows will have an edge from themselves to all flows releasing it.
- We can pick 99 flows following their original plan and find out an alternative path for the flow requiring 100 units of resource, which saves 99 alternative paths.
- Iteratively reserve the resource for each flow if all of its requiring resource do not exceed current available capacity.
- Phase 3 - **Alternative Path**

### 6.4 Performance Evaluation

- Simulation Environment
- Fattree topology
- There are 500 switches with a three-layer architecture, including 100 core switches, 200 aggregate switches and 200 ToR switches.
- Assume that the maximum available link capacity of each link is 1 Gbps.
- The switch memory has available capacity of 1500 rule slots by default.
- Set modifying rule as 10 ms and installing/deleting rules as 5 ms each according to the measurement
- Scenario Design
- The scenario is used to simulated the traffic migration when resources are deficient.
- Heavy Loaded Chain

---
## 7. GPS: SDN based Seamless Packet Delivery in Fog assisted VANET

### 7.1 Introduction

- 基於 **SDN** 與 **霧端運算 (Fog Computing)** 輔助架構的 ==Greedy-based Packet Scheduling (GPS) 演算法==，以支援車在網路內的即時資料傳輸。
- 將 GPS 演算法實現於具有 SDN 控制器的車載網路模擬環境，以評估 GPS 支援即時資料傳輸的效能。
- 霧端運算的負載
- 基地台與霧端間的傳遞延遲
- 霧端與控制器間的傳遞延遲
- 使用者在移動過程中根據 GPS 演算法規劃好在移動路徑上接收資料所需要的頻寬，並能在使用者與基地台換手時不需要重新發送請求到霧端，提升使用者在移動中的資料接收量。
- **霧端運算 (Fog Computing)** 是由 Cisco 所提出的架構，基於雲端運算 (Cloud Computing) 所提出的一種全新架構，能==有效減少回應時間==並能==降低雲端負載過重==的問題。

### 7.2 **==Greedy-based Packet Scheduling==**

### 7.3 控制器的運作

- 透過 OpenFlow 實現 SDN 的功能支援 SDN 控制器，SDN 控制器與 RSU、霧端之間的訊息傳遞。
- 當 RSU 接收到使用者的封包時，RSU 會以 Packet-in 訊息傳送至 SDN 控制器。
- 當 SDN 控制器接收到 Packet-in 訊息會進行解析，取出所需要的參數資訊 (包含載具目前的位置、速度等)，執行 GPS 演算法來挑選出與 RSU 所連接的最佳霧端，最後選擇適當的時間傳送 Packet-out 訊息給該霧端。

---
## 8. An Energy Efficient Radio-over-Fiber Network for High-speed Trains

### 8.1 Introduction

- Proposing an energy efficient **Radio-over-Fiber (RoF) network** architecture for high-speed trains as a contribution to the 5G vision.
- The proposed architecture incorporates software-defined networking (SDN) for network management and machine-to-machine for train-to-cloud communication.
- **Self-Organizing Network (SON)**
- SON provides a collection of functions for automatic configuration and optimization.
- Reduce both CapEx (Capital Expenditure) and OpEx (Operational Expenditure)
- The context information of physical cell ID (PCI), obtained from the mobility of trains traversing the tracks, is exploited to develop a self-organizing SDN-enabled energy management solution for HSR network.
- **Software-defined Networking (SDN)**
- Network programmability
- Centralization of the logic control
- Provide global view of the network
- Network virtualization
- **Machine-to-Machine (M2M)**
- Making the communication between devices easier
- Connect a device to another device and enables them to communicate without human interference.
- Enable a device to handle interoperability between different networks and connected systems.
- Reducing the ecosystem cost expenses regardless the diversity of the application, systems, and devices.
- **Simple Network Management Protocol (SNMP)**
- An application layer protocol
- Facilitating the exchange of management information among network devices
- Enables administrators to remotely manage

### 8.2 Problem Definition

- Consider a real-life scenario where the power consumption is not optimized in mobile network.
- **==Reducing the energy consumption by the RoF network.==**

### 8.3 System Architecture

- Context Information Collection and Processing
- Gathers the context information related to the train mobility
- Determines the approximate location of the train
- Centralized Control
- Coordinates the functionality of the entire architecture
- The SDN controller provides both northbound interface to interact with a management application and southbound interface to manage the RoF power status.
- Management Application
- The actual control of RoF nodes
- Decides when to switch on or off the RoF nodes via SNMP protocol based on the received context information.

---
## 9. 作為無線網路基地台的無人飛行載具自動化部署

### 9.1 Introduction

- 將無人機作為無線網路基地台，有效率地部署無線網路基地台的無人機。
- 提出二個基於**賽局理論**的無限飛行載具自動化部屬的方法。
- **Gruber's Method**
- 以無人機作為無線網路基地台的自動化導航方式。
- 限制：
- **需要無人機間互相協調**
- **未考慮無人機所消耗的能源**

### 9.2 Problem Definition

- **Gruber’s Method Assumption**
- 假設無人機會移動而終端機為靜止或準靜止。
- 終端設備若偵測不到任何無人機的訊號干擾比大於或等於 0.1，則不與任何無人機連結。
- 若至少有一個無人機的訊號干擾比大於或等於 0.1，則終端設備與最大訊號干擾比的無人機連結。
- ==最小化無人機的能源消耗，最大化下行頻寬。==
- 自動化佈署
- 獨立尋找並移動至特定位置
- 提供接收服務

### 9.3 Solution
- 無人機終端設備賽局
- 若無法與終端設備連結，則選擇任意方向移動
- 選擇有提升與終端聯結數量的方向
- ==柏拉圖改善方法 (Pareto Improvement Approach; PIA)==
- 最小化至終端設備的平均距離 (MAD2T)

### 9.4 Experiment (Simulation)

- 在 $60 \times 60$ $km^2$ 的部屬範圍內以均勻分布的方式隨機放置 360 至 3600 台無人機。
- 以均勻分布與群集分佈進行
- 指標
- 平均頻寬效率
- 總移動成本

---
## 10. Highway Travel Time Prediction with Spatiotemporal Mobility Exploration Based on the Internet of Vehicles

### 10.1 Introduction

- **高速公路的旅行時間估計與預測**
- 提出 ==Spatiotemporal Weighted-RMSS (Root-Mean-Square Similarity) 預測法==分析資料相似性，找出最適當的訓練樣本數量進行資料建模，以預測當日旅行時間；而不在當日已知資料的條件下。
- 利用高速公路 ETC 單向資料傳輸之車聯網通訊技術所搜集的資料，分別透過時空、樣本數測試與相似度函數的階段性改進。
- 透過高公局 ETC 資料驗證預測精準度高達平均 98%。

### 10.2 Solution

- 旅行時間估計與預測
- **時間可歸納性**
- **空間可歸納性**
- 具有當日之部分旅行時間資料
- Weighted-RMSS (Root-Mean-Square Similarity)
- 改進 k-NN (k-Nearest Neighbors) 演算法進行旅行時間預測。
- Temporal Weighted-RMSS
- 計算出相同路徑下，不同時間點出發的旅行時間值。
- 將當下時間點的旅行時間值進行方均根相似度計算。
- **Spatiotemporal Weighted-RMSS**
- 解決 Temporal Weighted-RMSS 在地域性相異性對預測精準度之影響。
- 考量空間歸納性。
- 不具當日之部分旅行時間資料
- Multi-SBLR (Slope-Based Linear Regression)
- 整體資料集的趨勢斜率
- 資料在相異週別之同一時間點的點趨勢斜率
- 與前一週別資料間的正負差異
- Temporal Multi-SBLR
- Multi-SBLR 的預測值與實際值有相同甚至相似的變化趨勢，甚至在壅塞時段更為明顯。
- **Spatiotemporal Multi-SBLR**
- 在不同地域涵蓋的資料通常內部相依性較高。
- Temporal Multi-SBLR 嚴重降低旅行時間預測值與地域的交通的關聯性。

---