Mesh Networking: Building, managing, and the works Suman Banerjee

advertisement
Mesh Networking:
Building, managing, and the works
Suman Banerjee
Wisconsin Wireless and NetworkinG Systems (WiNGS) Laboratory
Mesh in the press
Outdoor metro meshes
Many cities, urban
downtowns
A set of mesh points
connected to gateways
Goal: provide Internet access
to users
Business model?
Serve the taxpayer
Run it as an ISP
Single “expert” administrator
of the network, and
homogenous nodes
But there are other scenarios!
Madison, WI downtown
60 mesh points
Scenarios I want to talk about
Rural networking
An urban hub connecting a set of villages (say 40 mile range)
Goal: provide Internet access to users
Integrate applications: Distance education, tele-medicine,
expert advice
Cost needs to be low
“Not-so-expert” administrator, possibly homogeneous nodes
Indoor (home) meshes
Extend the notion of home-networking
HDTV over wireless: from set-top box to 2nd floor TV
Phone base unit to handset in kitchen
Single “not-so-expert” administrator, heterogeneous nodes
Rural networking: Issues
Why meshes make sense?
Can use WiFi (unlicensed)
Most of equipment is low cost and widely
available
Technology is getting there to meet the
demands
Rural networking: Issues
Start at the very beginning
How do we deploy a mesh?
Manage
How do we monitor and manage it?
Improve
If we detect performance problems, what are the right changes
to make?
Security
A perennial problem in any domain
Deployment
Where to place
the mesh nodes
+ how many
nodes
such that:
40 km
Budget
constraints are
met
Good fault
tolerance and
quality of service
• 384 Kbps to each village
• Also multicast
Deployment
Parameter choices:
Many possible locations (> few 100s)
Candidates for deploying mesh nodes
Directional nodes vs. omni-directional nodes
Cost vary depending on nature of antenna systems in use
Interference patterns
Buildings, other hotspots
Gateway locations
Choice of channels in multi-radio nodes
Deployment
Current state-of-the-art:
Manual inspection based human
judgement
Example : MadCity broadband uses
consultants
WFI Networks
Neither cost effective nor scalable
Management
All of the network management headaches that occur in enterprise
Plus
Those due to multi-hop wireless nature of the network
A control wired backplane does not exist
Control and data on the same “flaky” wireless interface
Performance debugging
User calls up:
“Network is too slow at University Ave. and Randall St.”
Can we find the bottlenecks?
Can we detect route mis-configurations?
Management
Current solutions:
Very basic, SNMP based
Essentially monitoring done by the mesh nodes themselves
Can be extremely inaccurate
Data is of low fidelity
Maybe a specialized (low-cost) monitoring
infrastructure
Improve
How to upgrade the network?
Where do we deploy new nodes?
It is possible to spend money and degrade
performance
How many nodes?
What kind of nodes?
Where do we place them?
Security
Current planned model
Secure each link
End-to-end security obtained through
composition of secure links
What if a mesh node is compromised?
Indoor meshes: Issues
Devices manufactured by different
vendors
TVs, set-top boxes, phones/handsets, etc.
Interoperability is key
Should be virtually un-managed
Things should just work out of the box!
Security
Finally, the applications
Multi-hop wireless has many interactions
that reduce end-to-end throughput
HDTV, voice, tele-medicine requires some
QoS guarantees
Wireless links are very diverse and have
different properties
Some of the MAC protocols adapt poorly
How to manage priorities of these traffic on
the mesh
Thanks!
Suman Banerjee
Wisconsin Wireless and NetworkinG Systems (WiNGS) Laboratory
Download