Transcript Diapositive 1 - EU-MESH

Deploying Wifi on Lampposts
The Ozone way…
Nicolas MECHIN
Ozone
EU-Mesh’s Heraklion meeting 07/08
Playground

« free » access to lampposts in Paris city center

Permanent power available

Aesthetic constraints
 Devices deployed need to be very discrete
- No big directionnals or patch antennas…

No land lines for internet backhaul
 Backhaul through Ozone’s wireless 5GHz network

Idea is to offer seamless wifi roaming (sort of…) within the
coverage area
2
Area to be covered
1,4 km long
3
Architecture
OZONE
Lamppost n+1
Lamppost n
Lamppost n+2…
Radio Link
Radio Link
Ethernet
POE
POE
220 V Powerlines
About 100 m
4
Hardware

Mikrotik RB532
 MIPS architecture
 3 Mini-PCI slots

3 * CM9 wifi cards
 Atheros chipset

Dual-band antennas
 4,5 dB @ 2.4GHz
 7 dB @ 5GHz
5
What it looks like
6
Actual Deployment

10 LampPosts
 4 directly connected to Ozone’s
network
 6 through other lamppost
 All links use 5 GHz (5480 Hz – 5700 Hz)

3 different connecting point to
ozone’s 5GHz network
7
Actual Deployment
500 m
730 m
8
Coverage achieved

Over 160 000 m² covered
9
Coverage achieved
10
Coverage achieved

Focus on the « Hotel de
Ville » area

« seamless » wifi mobility
within this area
 14 000 m²
11
First impressions
cf. EU-MESH Benefits and Performance Metrics

Low cost deployment : Yes
 Wireless backhaul : no heavy cost for deploying fixed lines
 Relatively cheap hardware
- 200 € / AP (all included)
 Still have to climb the lamppost…

Fast deployment : Not so much
 we missed easy to use tools to know whether the signal was
good enough, what was the best position for antennas. Mass
deployment not ready
 Radio settings and IP addressing had no be prepared

Reliability, flexibility, reliability ...No
 Static routing and static radio configuration
- No self healing…
12
First impressions
cf. EU-MESH Benefits and Performance Metrics

EU-Mesh should provide us with an answer to these
shortcomings
 Easy deployment : auto-configuration, tools for controlling radio
signal at deployment
- Considering « low skills » technicians are to deploy the network
 Mesh technologies can provide self healing networks
- Wheread today we have a static network with static radio
configuration, static IP adressing and routing…
13
Tests planned

Performances of radio backhaul

Performances of inter lamppost radio hops
 Considering « dedicated » radio interface and not single radio
scenario
- Throughput
- RTTs
14
Results
Focus on radio performances
15
Results
Focus on radio performances

Relatively poor radio signal in NoLOS scenario
 Trees do affect a lot the radio link
 The higher the frequency, the higher the attenuation

Very good results in LOS situations
 19 Mb/s with a 700m radio link

Throughput degradation at each hop
 Despite the use of separate dedicated radios
- Degradation isn’t 50% as in a 1 radio scenario, but still around 30 to
40%
16
Results
Within Eu-Mesh

Tests already done : benchmark for EU-Mesh enhancements

Many further tests can be done
 Playing with frequencies allocations
 Changing power settings
 Whatever you want to test …and see the impact on
performances…
17
Mobility
The problem

Goal is to offer the client a seamless experience
 Even if he is moving
- pedestrian or « car-in-paris-traffic-jam » speed
 he doesn’t care which AP he is connected to, and even less
which backhaul this AP is using

Lamppost may be backhauled by different Ozone’s aggregation
point
 3 in our case, with IP addressing specific to each of them

Lampposts may be backhauled by other technology
 DSL, Fiber, 3G, Wimax (all of these from Neuf Cegetel / SFR)

Lampposts may be backhauled by other providers
18
Mobility
Our solution : tunneling
+ Transparent to the user
No need to deploy third party software on clients devices
+ Simplify provisionning of AAA and captive portal
Only need to allow each AP to connect to the network (affect IP
address and gateway) : tunnels go live automaticaly
- Has an impact on available throughput
Overhead of tunnels : from 10% to 40% today, depending on
adopted solution
19
Mobility
Our Tunnel solution : OpenVPN
+ Very easy to configure
Available for a very large number of hardware platforms and OSs
+ Can handle NAT without any problem
GRE can’t, IP-Sec needs Nat-traversal
+ Can handle No MTU Problems
- ??
20
Mobility
Architecture

All tunnels are bridged at the controller
 No IP change for the client when
changing association

Roaming possible between various
lampposts
21
Mobility
Architecture

Roaming also possible between
lampposts and DSL « box »
22
Mobility
First conclusions

Roaming times depend on client’s wifi driver implementation
 When they consider a signal is too weak and switch to a better
one

Tunnels may deliver this feature
 Efficiency depends on Wifi reassociation times
 …but has a non negligeable impact on performances
- 8 to 10 % in our current solution (OpenVPN on UDP, no encryption)
 And is dependent on network architecture
- Tunnels will be shut down if mesh architecture changes ; need to be
able to go live very fast after such a change

… just a transitionary solution to « real mobility » features
 To be developped within EU-Mesh (Wioptimo…)
23
Monitoring
Tools and how-to exploit them

Classical monitoring tools
 Cacti
 Nagios

Netflow exploitation
 Nfsen
- Able to compare usage of « fixed-wireless » clients vs « Mobile »
clients
- On protocol, ports, type of application…
- Might be interesting within EU-Mesh as an input on what usage is
made of a wireless (mesh) network, and impact this usage has on
the network
24