Transcript Systems Issues in Wireless Networks

Systems Issues in Wireless
Networks
Henning Schulzrinne
Columbia University
NSF wireless workshop – July 2003
Overview
•
•
•
•
•
•
•
Applications
Autoconfiguration
Evolution
Hybrid networks
Non-data applications
New wireless models?
New types of networks?
What makes wireless networks
different?
• Network probabilistic throughout
– wired generally hides this at lowest layer
•
•
•
•
Node and link availability
Identifier collisions
Packet loss
Link bandwidth availability
Applications
• Transition from general-purpose networks to
niche networks
• Need honesty in scoping (cf. active networks)
• What are different types of networks good for
(and what not)?
• Applications for ad-hoc and sensor networks
– Beyond the (dubious) emergency and military
applications
– Clusters of vehicles
– Short-term ad-hoc networks
– Beer glasses waiting to be filled
Autoconfiguration & Discovery
• With open spectrum, how to find
– Available network services  can’t just scan 3 GHz of bandwidth
• Might be hundreds of ad-hoc and infrastructure networks
– Available services that take my credentials
– That offer best/cheapest/… service
• Geographic service discovery
• Ad-hoc and disconnected network services
• How to automatically deploy networks
– In simulations, they just magically connect…
• Not much usable development
–
–
–
–
LDAP doesn’t scale (administratively)
SLP has severe functionality limitations (no hierarchy)
UDDI, etc. too narrow
Zero-conf too low level and not suitable for all networks
Evolution
•
Implicit assumption: natural evolution
–
–
–
–
–
•
from specialized to general
from circuit to packets
from ATM to IP
from large to small cells
from single network access to global AAA with roaming
However, even IP networks can look rather crufty:
– GPRS Russian-doll stack
– WAP & 3G “walled garden” bias
•
Longer lifetime of networks
– analog wireless may not disappear for a decade+
•
Specialization and large-enough eco system niches
– restricted applications simplify security (DOS), billing, lower cost
•
Stuck with glueing together 30 years of wireless technology at the
application layer?
– very little research on evolution, co-existence and gateway models
– economics, security and technical issues
Non-data applications
• Almost exclusive research focus on data
(and multimedia) applications
– higher bandwidth, general-purpose networks
• Other wireless applications:
– location sensing (GPS, location beacons)
– awareness of other people and resources
– remote control (garage door opener…)
• Layer on top of general-purpose networks
or construct special-purpose networks?
Hybrid networks
• Not just vertical hand-off
• Software-defined radios: use diverse
networks at (almost) the same time
– e.g., high-speed ad hoc network + low
bandwidth WWAN
– separate control and data channels?
New wireless models
• Have we explored the architecture space?
– infrastructure access
– ad-hoc as infrastructure “funnel”
– ad-hoc networks
– intermittent-connectivity networks
• Do things change with highly directional
networks that are no longer broadcast?
New types of networks
• Roughly now have tools for ubiquitous and highbandwidth data delivery
• But still far too expensive for many applications
– BlueTooth USB: $50 – requires “real” processor
– only useful for very small groups, not clusters of O(100)
• New sets of applications for wide area, very low
bandwidth, ultra low complexity devices
– many applications only need 1 bit/hour (light switch, thermostat)
– many sensor applications (toothbrush, scale, toaster, traffic light
controlled by bus, …)
• 802.15 status?
• Parasitic networks?
Protocol design challenges
•
Wireless starting to reach up the stack – cross-layer design, but need to be
precise
– Applications and transport have only limited repertoire of actions: send packet,
drop packet, delay packet, fragment packet
•
•
Currently, usually don’t even know outgoing/incoming interface for a socket
(and its current bandwidth)
Trade-off error rate vs. bit rate more explicit in wireless
– not just binary  another example of probabilistic network behavior
•
Examples:
– packet loss 
• wait (fade)
• reduce rate (congestion)
• retransmit (but probably better delegated down)
•
•
Unequal error protection (UEP)
Delivery of errored PDU to application
– application needs control
– may need stronger e2e checksum if done inadvertently