Skip to content

Latest commit

 

History

History

tutorial3

README.md

Tutorial 3: Workload Orchestration and Server Selection

Overview

This tutorial demonstrates the performance evaluation of different workload orchestration policies in multi-tier edge-cloud computing environments. The focus is on comparing various algorithms that determine where to offload computational tasks - whether to edge servers or cloud servers based on system conditions and resource availability.

Scenario Description

Simulation Scenario

Figure 1: Tutorial 3 multi-tier workload orchestration scenario.

The simulation environment is designed with the following characteristics:

  • Multi-Tier Architecture: Mobile devices can offload tasks to either edge servers or cloud servers
  • VM Provisioning Strategy: Worst-fit VM provisioning algorithm (least loaded first) is used across all server tiers
  • Inter-Edge Communication: Tasks can be transmitted to remote edge servers via Metropolitan Area Network (MAN)
  • Independent Network Modeling: WLAN and WAN delays are modeled independently, ensuring WLAN performance is not affected when tasks are sent to remote servers

Key Features

  • Flexible Offloading Options: Tasks can be directed to local edge servers, remote edge servers, or cloud servers
  • Network-Aware Routing: MAN connectivity enables communication between geographically distributed edge servers
  • Isolated Network Performance: WLAN operations remain unaffected by WAN traffic to cloud servers
  • Dynamic Resource Allocation: Real-time monitoring of edge server utilization and network bandwidth

Network Infrastructure

WLAN (Wireless Local Area Network)

  • Purpose: Connects mobile devices to local edge servers
  • Performance: Independent of WAN operations
  • Characteristics: Low latency, limited coverage area
  • Usage: Primary interface for edge server access

MAN (Metropolitan Area Network)

  • Purpose: Interconnects distributed edge servers within metropolitan area
  • Performance: Medium latency, high bandwidth
  • Characteristics: Enables load balancing across edge infrastructure
  • Usage: Inter-edge server communication and task forwarding

WAN (Wide Area Network)

  • Purpose: Connects edge infrastructure to cloud servers
  • Performance: Higher latency, variable bandwidth
  • Characteristics: Internet-based connectivity with bandwidth fluctuations
  • Usage: Cloud server access for resource-intensive tasks

Workload Orchestration Algorithms

This tutorial implements and compares three distinct approaches for server selection and workload orchestration:

1. Random Server Selection (RND)

  • Strategy: Randomly selects a server (edge or cloud) to offload tasks
  • Characteristics:
    • No optimization criteria or intelligence
    • Provides baseline performance metrics for comparison
    • Equal probability for edge and cloud server selection
    • Useful for evaluating the impact of intelligent orchestration policies

2. Edge Server Utilization Heuristic (ESU)

  • Strategy: Makes server selection decisions based on edge server CPU utilization
  • Decision Logic:
    • IF average edge servers CPU utilization > 75%
    • THEN offload task to cloud server
    • ELSE offload task to edge servers
  • Characteristics:
    • Edge-centric approach prioritizing local resource availability
    • Prevents edge server overload by redirecting to cloud when necessary
    • Balances edge processing capabilities with cloud scalability

3. Network Utilization Heuristic (NWU)

  • Strategy: Makes server selection decisions based on WAN bandwidth availability
  • Decision Logic:
    • IF WAN bandwidth > 5 Mbps
    • THEN offload task to cloud server
    • ELSE offload task to edge servers
  • Characteristics:
    • Network-centric approach considering connectivity quality
    • Leverages high-bandwidth periods for cloud processing
    • Ensures edge processing during network congestion

Note: This tutorial provides fundamental insights into workload orchestration and server selection strategies. Real-world deployments should consider additional factors such as data locality, security policies, service level agreements, and cost optimization.