Performance Evaluation of Wireless Ad-hoc Network ...

2 downloads 32153 Views 17MB Size Report
Light-weight and Small: Laptop, USB Flash Drive and Wi-Fi stick. • Basic pre-installed software: Notes, Spreadsheet, etc. • Inexpensive: Given Laptop can be provided, USBs are negrigible. • Easy to .... best to reach more than 1 hop away ...
Performance Evaluation of Wireless Ad-hoc Network for Post-Disaster Recovery using Linux Live USB Nodes

Vaibhav Garg, Kotaro Kataoka Indian Institute of Technology Hyderabad, India Siva Subramanya Rohith Talluri Manipal Institute of Technology, India

Doubts about Big System (1/3) • Cost of Equipment and Preparation • Size of Equipment

Doubts about Big System (2/3) • Risk and Cost of Sending Engineers • System consumes spaces, needs proper setups and professional skills

Doubts about Big System (3/3) • Sustainability of operation and support activity • Internet is crucial, but what to do before Internet comes?

Linux Live USB Light-weight and Small: Laptop, USB Flash Drive and Wi-Fi stick Basic pre-installed software: Notes, Spreadsheet, etc. Inexpensive: Given Laptop can be provided, USBs are negrigible Easy to setup: Boot Guest OS from Linux Live USB, then GUI helps Wireless: Wireless Mesh for interconnecting Live USB nodes (can work as Wi-Fi APs for sharing Internet connection if available. • Local Online Services: Web, proxy, video/audio communication, etc. within the local network without waiting for the recovery of Internet connection. • Other Benefits: OS Maintenance at one place, Isolation from the existing OS, Automated health check from remote • • • • •

Post-Disaster Recovery Network

Points to be addressed • Planning the network – Multi-channel Hop-by-Hop vs. Single Channel Mesh – Addressing and Routing

• Automating Network Setup – DHCP-like address delegation – Configuration of Dynamic Routing Protocol

• Evaluating the network through Field Trial

Networking in Single Channel Scenario • • •

All Live USB Clients join a single mesh 2-tier IP Addressing: Backbone and Wi-Fi Hotspots Routing

– Static Default Gateway with NAT on Live USB Client. NAT traversal problem exists for communication between end users. – Dynamic without NAT on Live USB Client. Many control messages will flow on the mesh network.



Broadcast message travels much

Networking in Multi Channel Scenario • • • • • •

IP Subnet per channel Requires subnet coordination No reason to use NAT Requires Dynamic routing Tends to form a line shape
 instead of a tree depending on 
 the available number of WI-Fi interfaces Broadcast messages are isolated

DHCP Pool Allocator at Server 1, 2: Client 1 uses DHCP for to join Server’s mesh 3, 4: Client 1 used DHCP-PA for creating a new mesh 5, 6: Client 2 uses DHCP for joining Client 1’s mesh. 7, 8: Client 2 uses DHCP-PA for creating a new mesh 9, 10: Client 2 uses DHCP-PA for creating a WI-Fi Hotspot User Terminals is served by Client 2

Field Trial Setup (1/2) Features

Specifications

USB Flash Drive

SanDisk Extreme USB 3.0 32GB USB Flash Drive (Read: 245MB/s, Write: 100MB/s)

USB Wi-Fi Adapter

WLI-UC-AG300N (Supporting 8012.11n 5GHz)

Guest OS

Linux Ubuntu 14.04.1 LTS

Wireless Ad-hoc Networking Protocol

batman-adv-2014.3.0, olsrd-0.6.8

Routing Protocol

RIPv2 (Quagga)

Wi-Fi Hotspot

hostapd and dnsmasq

Field Trial Setup (2/2) •

2 weeks to collect all data from 2 setups in IIT Hyderabad

– Single Channel Mesh and Multi-Channel Hop-by-Hop – 802.11n 5GHz, Noisy condition (other 5GHz signal was seen during trials)

Single-Channel Mesh

Multi-Channel Hop-by-Hop

Performance Evaluation • Network Setups – – – –

Single-channel + OLSR Multi-channel + OLSR Single-channel + BATMAN-ADV Multi-channel + BATMAN-ADV + RIPv2

• Tested Performance Factors

– UDP Throughput from Clients 1, 2 and 3 to Server • Average Jitter over 10 trials < 4 ms • Average Packet Loss over 10 trials < 5%

– Packet Loss Rate vs. UDP Throughput – TCP Throughput from Clients 1,2 and 3 to Server

Results of Field Trial (UDP Throughput) • Minimum, maximum and average UDP throughput values from C1, C2 and C3 to S with the following conditions – Average Jitter over 10 trials < 4 ms – Average Packet Loss over 10 trials < 5%

Results of Field Trial (Packet Loss Rate) > 5% Loss, 10Mbps > 1% Loss, 28Mbps

• OSLR performs better for 1 hop • Multi Channel + Batman-adv performs better
 for multiple hops 5% Loss,
 3Mbps

Results of Field Trial (TCP Throughput) 14.76Mbps, 1 session, 1 hop away from Server

> 2.5Mbps, 1 session, 3 hops away from Server

> 3Mbps,
 1 session, 2 hops away from Server

• Multi Channel + Batman-adv performed the best to reach more than 1 hop away

Discussion and Future Work • Wi-Fi USB Adapters performed OK • What about OpenWRT or DD-WRT?

– They are Wi-Fi Aps by nature (better antenna gain) – Maintenance and compatibility of Custom Firmware with Wi-Fi Aps maybe problematic – By the way, they are not Live USB anymore…

• • • •

More 5GHz Frequency Channels What if we use 801.11ac for mesh? Larger-scale Field Trials? DTN-oriented networking

– Support intermittent connections – Takes long time to propagate date, but reaches by 100%

Conclusion • Linux Live USB as alternative solutions for PostDisaster Recovery – Small, inexpensive, works with laptop battery, fully wireless, local online service without Internet

• Questions about Network Performance • Hanson-oriented approach – System Implementation – Field Trials in a small area of University Campus

• Our primary choice – Multi Channel batman-adv to cover multiple rooms

Thank you. Q&A?

Suggest Documents