Download PPT

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

IEEE 1355 wikipedia , lookup

Deep packet inspection wikipedia , lookup

Recursive InterNetwork Architecture (RINA) wikipedia , lookup

Distributed operating system wikipedia , lookup

IEEE 802.1aq wikipedia , lookup

List of wireless community networks by region wikipedia , lookup

CAN bus wikipedia , lookup

Airborne Networking wikipedia , lookup

UniPro protocol stack wikipedia , lookup

Routing in delay-tolerant networking wikipedia , lookup

Transcript
Wireless Capacity
A lot of hype




Self-organizing sensor networks
reporting on everything everywhere
Bluetooth personal networks connecting
devices
City wide 802.11 networks run by
individuals and companies
No more Cat5 in homes/businesses
Capacity


As systems researchers, the most
glaring question is “Does this scale?”
What do we mean by scaling?


What is the aggregate network capacity?
What is the per-node capacity for nodeoriginated data
Observed capacity

Das et al. simulation of 100 nodes




2Mbps base throughput
7 simultaneous transmissions
Per-node bandwidth few kbps
Others see similar capacity
Physical limit

Competition for physical bandwidth

Signal power degrades with distance as 1/ra for
some a>2
Pi
| X i  X j |a
Pk
N 
a
|
X

X
|
ktransmitting
k
j

As an order of magnitude, in ns transmission range
~250 meters, interference ~550 meters
Network capacity

Upper bound total capacity,arbitrary
destination
(Wn )


Why? Intuitively, assuming constant
density: total area/capacity ~n,
diameter/average path length ~n
Global scheduling can achieve:
1
(
)
n log( n)
What is the limit?

As density increases, the number of
nodes a packet interferes with increases


Constant power, nodes per unit area larger
Lower power/more hops, total
transmissions increase
802.11 Chain propagation (simulation)


Achieve 1/7 of maximum 1.7Mbps
Expected ¼ of maximum 1.7Mbps
MAC inefficiency?



802.11 works until
offered load
exceeds capacity
Waste bandwidth at
first node
Waste time backed
off
Simulation vs. Reality
Solutions?

Smaller networks?



Add extra repeater nodes



Requires exorbitant number of nodes
Factor of k repeaters, k extra per-node capacity
Local communication patterns?



Suggested in papers
Only helpful if lower overall use
Widespread base stations
Local data processing
Be sneaky
Traffic pattern
Power law traffic pattern
p ( x) =
xa

A
t a dt
Per-node capacity
a<2 Approaches constant
a=2 O(1/log(n)): GLS uses this
a>1 O(1/n)
Be sneaky

If we achieve three properties, we
should be able to get scalability




All direct communication is local
Message paths are short (preferably O(1))
Squander no opportunities to send
Can we still achieve full connectivity?

Maybe: Mobility
Mobility

Nodes move randomly



Persistent communication patterns



Ergodic (uniform space filling) motion
No proof that this is NECESSARY
Random source/destination patterns
Unlimited data
Buffering

Nodes can buffer data
Mobility

To achieve scalability, we want three
properties

All direct communication is local




Send messages only to nearest neighbor
Distant communication depends on chance
movement
Message paths are short (preferably O(1))
Squander no opportunities
Mobility

To achieve scalability, we want three
properties


All direct communication is local
Message paths are short (preferably O(1))


Never forward along paths longer than 2 hops
Squander no opportunities
Mobility

To achieve scalability, we want three
properties



All direct communication is local
Message paths are short (preferably O(1))
Squander no opportunities Send data through
everyone


Whenever you are near any node, give it a (new) packet
for the destination.
On average should have data for every possible
destination
Requirements


Know closest node/range
Schedule local transmissions


They found the standard MAC may be ok
Buffering


Scales with radio bandwidth?
Scales with expected time to see a
destination node?
Model

Is this useful?



Potentially very long time to delivery
Potentially wide variance in delivery times
Unknown dependence on movement model





Space filling unrealistic(destructive to homes)
Another submission claims that travel along random line
segments also works
Unclear generalization to multiple hops
Static population model/bounded movement model
unrealistic for many random movement models
Existing applications seem unlikely consumers
What next?

Radio people



MAC layers tuned to ad hoc mode
Wasn’t clear from results presented this is
more than a moderate constant factor
Systems/applications people


Communication patterns with good locality
Take advantage of external sources of
bandwidth (fiber optics or station wagons
of tapes)