Systems Issues in Wireless Networks Henning Schulzrinne Columbia

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 & 3 G “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 – Blue. Tooth 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 e 2 e checksum if done inadvertently