Liveblogging LCDNets 2013

Good Morning from Miami, I'm here today with Arjuna Sathiaseelan at ACM MobiCom Workshop on Lowest Cost Denominator Networking for Universal Access (LCDNet 2013)

ACM proceedings are now available at dl.arm.org.

Workshop Chairs Welcome: Arjuna Sathiaseelan, University of Cambridge

 

Its a full day, 9 presentations and posters. Introductions by the attendants.

Why are we here ?

Internet is essential: information is pervasive: MOOS, e-gov, e-health. Internet access is seen by many as a human right.

Challenges to Universal Access: Geographic & physical limitations, Economic challenges (in some countries internet is 40x the national average income.

LCDNets:  How do combine techs to provide low cost access to all, pooling resources at different network layers and securing these new networks

Agenda for today: (see below)

 

KEYNOTE The Challenge of Deployment - Prof Srinivasan Keshav, University of Waterloo (via Skype)

 

Challenges and learning from mistakes

Problem: Technology operates in a social, culture and economic context. Its difficult to determine the context and context changes with time

Low cost networks: Constraints such as Environment (temperature), Infrastructure (power, phone lines), Social (training), Culture (do poeple actually want what your offering?), Legal

SoftBridge Stack Model from Tucker and Blake: People, UI. Synch, Media, Device, Network, Power - normally as network engineers we ignore the applications, media and people

How to proceed ?: Iterative process, context describes the requirements, constraints and alternatives.

Design - Deploy - Monitor - Refine

Design: Consider every layer of the SoftBridge stack, DTN offers some solutions but its not a silver bullet, people don't want a delay that are just will to trade-off latency for cost, don't underestimate cellular networks, femtocells for example

Implementation: Options are limited, your limited free technology, don't build on unproven technology, "error 33" don't depend on the success of another research project, monitoring is critical, need to support for years not just weeks

Test: Build in test harnesses, always investigate unexpected results, this is the heart of new knowledge

Deploy: Always start with a pilot, where is OK to fail, you will need on the ground support

Monitor: monitor everything, visualize data graphically, don't skip on analysis and insist on formal reports from collaborators

Refine: Use monitoring data to refine assumptions, constraints and requirements, be prepared to for failure

Rule of 3: 1st too simple, 2nd too complex and 3rd is simple is just the right ways. Strive to achieve the second simplicity

Project Example: KioskNet

Internet cafe with connections to the internet via a bus (a real bus) that drives between the cafe and the internet gateway. Opportunistic networking in the wild. Trading delay for costs, see mobicom2006 paper on it.

Comparing the KioskNet to the SoftBridge network stack with pictures of the deployment.

Single-board linux based computer are tough to debug, vehicular environment is harsh and difficult to get agreements from bus operators

VLink: Reuses existing windows and Linux desktops, software only solution, leverages USB memory sticks and SMS

Outcome ? Failure, NGO don't want to unsupported software, people don't want delay, 3G is widespread

Conclusion: Its tough out there, 3 lessons learned: Contextualize using SoftBridge stack, iterate and simplify

 

Q: What are the big challenges ?

A: The issues that would not occur to you before the project: We are trying to use technology to solve problems where its not suitable, vast cultural differences

Q: Why did the project failure ? Did you fail to assess the requirements ?

A:  No, we had lots of people working on requirements, but the failure was due to availability of cellular technology

Q: Could the project still work in a new area, if the area lacked cellular technology ?

A: Low cost femtocell, with second hand cell phones and long range wifi would make more sense in most situations

Q: What that you suggest to DTN research ?

A: There's some interesting optimal path work in DTN but people don't want delays, they will tolerate them due to the cost limitations. Abstract DTN is great but make sure the DTN is really the best solution

Q: If people are not involved, is the delay ok ?

A: Its application specific

Q: But there are some great DTN projects ?

A: LTE femocells could achieve the same goals

Q: How helpful are governments ?

A: There are willing to help if the problem serious

Q: How can I apply the rule of 3 ? does it find your experience ?

A: First was mobicom2006 paper (publishable but impossible to deploy), 2nd was kisok and the 3rd was VLInk but Vlink was too late to be successful

Q: But what about offloading from 3G ?

A: We can only watch TV 24 hours a day, development of new spectrum use is faster, I don't think its a technology problem but a social/government problem

Binder: A System to Aggregate Multiple Internet Gateways in Community Networks - Luca Boccassi (Cisco Systems), Marwan Fayed (Univ of Stirling), Mahesh Marina (Univ of Edinburgh) [slides: pdf]

 

Motivation: it to make a difficult, not research challenges

Context here matters a lot, this began as a small wireless test bed in Scotland. Very remote areas

Q: why do poeple live there if its so remote ?

A: People looking for a quiet life

We have residents in the areas doing the deployment instead of amarys of grad students. Community are taking responsibility for there own broadband.

HUBS: Rural broadband Co-op ISU, no real technical obstacles we have the tools that we need but plent of techincal opportunties. A bunch of academics creating a ISP. The postive reason of this is data, data, data

The villages are connection: 4 ways in and out, ASDL and 40Mbps to an academic network. This is the 3/5 Mbps ASDL.

Gateway Aggregation (link boding): Suffer from packet re-ordering issues for TCP so in practical flows are allocated to pipes, so you can have more users but a single user doesnt benfit.

Goals: Useful bandwidth aggregation, load-balances and fault-tolerances

Constraints: Simple, Deployable, Resilient

Binder Architecture: 3 components: packet capture (at the village AP)  & redirection (a server at uni of Edinburgh),

Client -> Capture traffic with OpenVPN client-> MPTCP to OpenVPN server -> Server

We are tunnelling IP on IP, conventional wisdon says that losses on the internal tunnel cause other tunnels to try to remover. this is not the case in our senerio

Q: Where's the Proxy location ?

A: at the university (high speed and free)

Q: Why don't you do MPTCP from the client devices

A: Devices (such as windows) don't support it yet

Initial results: Loss probability verse goodput, latency verse goodput, time verse googdput (see paper for graphs)

Summary: pragmatic gateway aggregation, implementation builds on well-developed software, establishes a new potential use-case

Q: Doesn't modern application run multiple TCP connections ?

A: Usage like Skype and BBC iPlayer are tending towards single flow

Q: What is a user runs a VPN ?

A: No problem the VPN is totally transparent

Q: How long are the WiFi links

A: On average 3-15  km, the longest is 40 km. most of this is over water and therefore line of sight is easy to achieve

How Much Can We Carry? A Capacity Analysis of Delay Tolerant Networking in Developing Countries - Abdullah Alhussainy (ETH), Karin Anna Hummel (ETH), Panayotis Antoniadis (ETH)

Raabta: Low-cost Video Conferencing for the Developing World - Zubair Nabi (ITU)

 

Motivation: Internet access is a human right, strongly linked to development, demand is high, barriers are cost and infrastructure

DTN - Cheap phones, no infrastructure, lacking instantaneous end-to-end path so we using store-carry-forward. Application: rescue, disaster relief and anti-censorship, extending 3G infrastructure

Study the potential of DTN to extend the network coverage in terms of capacity ?

Approach: Real world cellular data, D4D anonymized mobile call details records, phone calls and SMS from 5 million orange customers. We are using this data to map locations using antenna.

We are mapping the popularity of cellular towers, and movement of users.

Assumptions: Local hubs, not just rely on people

Graphs of estimated capacity (see paper)

Assumed probablly of deliver: 0.2, data capacity: 100MB and video size 500 MB

Conclusion: Cellular dataset to capture mobility in Ivory coast, simple formula for capacity, Low cost APs

Q: Is the video sharing for just one user ?

A: Yeah, its just a light example, we haven't considered many factors like how we slice up the video

Raabta: Low-cost Video Conferencing for the Developing World - Zubair Nabi (ITU)

 

Leverage analog cable TV networks using Rasberry-Pi instances with a video conferences application and network codesign using multi-layered encoding

Motivation: Internet has the power to improve human condition, standard of life, resource utilisation

Challenges: high costs, intermittent power, extreme weather conditions, frequent equipment, lack of trained workforce and rolling investment.

Solutions: WiMAC, satelite, Long-distance WiFI, ZigBee, optical, cellular

Observation: backbone access is expensive, basic community-wide communication required

Design Cost: low cost, low power, fault tolentent (P2P), user interface, interent connectionvity is not a requirements, no infastructure,

Analog cable TV networks: broadcast, shared medium, cable modem as MAC layer bridge

Raspberry Pi: ARM, low cost, ethernet optional

Video conferences: 3 functional aspects: Overlay topology, encoding and packet loss recovery.

Raabta: UI -> database -> encoding -> RTP and SIP -> IP -> CSMA/CD or FDMA -> Coaxial

Costs: total end host $ 92 (custom hardware can reduce this greatly)

Network: use non-TV channels for video conferencing, same band for upstream.downstead

MAC layer: FDMA: supports 15000 end hosts or CSMA/CD: dyn allocation

Network layer: unmodified IP layer, opening potential to wider internet connectvity,

Trasport layer: RTP on UDP

Video Conferencing: User database (broadcast regularly to discover hosts), Mutli-layer encoding, UI

Beyond Rabbt: easy to extend to get internet connectivity, no additional operotional or power cost, simple mangement due to decentralizations, leverage existing infrastructure.

Q: What about power for the Raspberry Pi, the broadcast to discover host means it may take time ?

A: We need to have some super-hosts, that are highly connected to coordinate discovery

Q: Future work ?

A: I'm co-coordinating students to work on it

 

KEYNOTE Nanosatellites for Universal Network Access - Scott Burleigh, JPL/NASA

 

Motivation: The right to internet access, etc... as before

Challenges: economic, geographical and political obstacles

Communication Satellites: earth orbiting satellites can rely radio communications, visible for all points

Geostationary satellites - costly, provides communication to a limited part of the earth surface, high latency, single point of failure

LWO satellites - broad coverage area, low latencies, not satelites and a lower capacity then geostationary

Nano-satellites - broad coverage, no single point of falure but they normally have a single radio and limited movement

DNT and nano-satellites have the answer !! using base-stations on earth which store data and forward to satalities when in contact

The proposed size of the set of nanosatelite is 150, latency is still a problem

Many application are inadely delay tolerant like email, news feeds, backups and archiving, file transfer, disaster recovery, social networking. We were able to do things over post so we can like with the

Costs: estimated assembly cost $15 million, estimated launch cost for network $10 mill, estimated base station cost $3 million, total lifetime cost of the network: $33 million. Mean cost of transition $.011 per MB.

The network would be functional from just one satellite, so incremental deployment is easy and network degrades gracefully in the loss of satelities

Caveats: Tools are still needs like route computation and basic networking security measures

Q: How does this compare to the Loon from Google

A:  This should be cheap and simpler

 

CARPOOL: Extending Free Internet Access over DTN in Urban Environments -Ioannis Komnios (DUTH), Vassilis Tsaoussisdis (DUTH)

 

Why are we working in well connected countries ? There's digital excluded people within the first world, Its still seen as a lunxary by many even working people.

Extend internet coverage provided by public hotspots through buses and other pubic transport. The routing information for buses can be used for routing data.

Algorithm: create a connectivity tables (gateway, ferry, contact time) where gateway is a bus stop and and ferry is the bus. Breath first search of possible paths so the first valid path to the destination will be the shortest

 

SocialDTN: A DTN implementation for Digital and Social Inclusion -Waldir Moreira (Univ of Lusofona), Ronedo Ferreira, (Federal Univ of Para), Douglas Cirqueira (Federal Univ of Para), Paulo Mendes (Univ of Lusofona), Eduardo Cerqueira (Federal Univ of Para)

[paper: pdf]

 

Motivation: digital divide and social exclusion (same as every other talk)

SocialDTN, android implementation of DTN

Scenario: mobile nodes, Disruptive and intermittent links, low node density,

First Experiments: Deployment test with Android devices carried by 7 students and a traffic generator. Disseminate educational information

Future Steps: release the trace and using social board routing

 

Providing Security for Wireless Community Networks - Milena Radenkovic (Univ of Nottingham), Heidi Howard (Univ of Cambridge), Jon Crowcroft (University of Cambridge), Murray Goulden (Univ of Nottingham), Christian Greiffenhagen (Loughborough Univ), Derek Mcauley (Univ of Nottingham), Richard Mortier (Univ of Nottingham)

 

[thats me :) I'll be right back]

Filling the Gaps of Unused Capacity through a Fountain Coded Dissemination of Information - George Parisis (Univ of Cambridge), Dirk Trossen (Univ of Cambridge)

 

Motivation: Internet access as a human right (same as the other talks)

Fountain coding transport in our ICN architecure, focus on information dissemination

Digital Fountains: PubSub model, data from pub is fragmented, recombined into symbols and XOR. Retransmission free method.

Fountaining coding matches ICN as the PubSub model fits well, Multicast tree management - delivery of infromation

[be right back]

Q: What is the main reason for using fountain code ?

A: We needed a reliable transport mechanism

Traffic Shaping for Enabling Less-than-Best Effort Services at the Edges of Broadband Connections - Sotirios-Angelos Lenas (DUTH), Vassilis Tsaoussidis (DUTH)

 

Gneral: "Free internet for all", exploitation of resource pooling technology, user provided networks (UPNs), two parties: micro-providers and guest users, less than best effect to guest users.

Low-expected quality, but it still needs to be service. We need to balance BE and LBE traffic. Explotation of DTN network management benefits in UPNs, incoropration of non-volatile storage into AP's architectures

Numerically assessed the impact of LBE traffic on BE traffic: different packet-sized distributions and vary network speeds.

Hybrid Packet scheduling scheme: Broadband connection sharing (FON,OpenSpark, Wifi,com, Open-Mesh) not of these support the free internet for all. Reference to PAWS [:)]

LBE service deployment on AP: User -Provided Network Queueing (UPNQ): exclusively allocation of resources may not be acceptable, instead use a non-preepmtive priority queuing scheme (PQ)

Two traffic classes: Home traffic verse Guest Traffic

Aim: Evaluate the impact of guest-traffic on home-traffic in terms of queuing delay and average system time

[see graphs in paper]