An Architecture for Tetherless Computing S. K eshav University of Waterloo October 21, 2004 Joint work with Aaditeshwar Seth and Patrick Darragh

Server DC MS DC What if… MS

Applications Healthcare tell a doctor about latest test results Environmental monitoring fly a plane over a site to collect data Rural development bus on a rural route becomes a data mule Personal productivity Almost always in-sync iPod Image sharing for migrant workers

Tetherless computing Smart mobile devices that opportunistically communicate with resource-rich data centers over heterogeneously administered wireless and wireline networks Internet cloud Server Data center

1. Computing costs are plummeting From Processor costs have come down by six orders of magnitude in three decades CMOS allows on-chip logic, memory, imaging and RF components Devices will merge computing, audio, and video Cell phone Still camera Video camera MP3 player PDA

2. Batteries are lasting longer

3. Wireless networks are proliferating © Intel

4. Data Centers aggregate resources

Requirements Must build on existing Internet architecture Mobility transparency Address changes as a mobile host moves from one subnet to another How to locate a mobile? Disconnection resilience TCP cannot sustain long duration disconnections Should not require both ends of a connection to be simultaneously present Identity management Mobile should have the same identity no matter where it goes Mobile and infrastructure should mutually authenticate Access sensing Is a mobile in a hot spot in the first place? Important to make the right decision Need low control overhead To maximize use of opportunistic communication

Existing solutions have limitations Network layer solutions: Mobile IP/HMIPv6 Provide mobility transparency But do not provide disconnection transparency Transport layer solutions: TCP Migrate/Rocks and Racks Provide both mobility and limited disconnection transparency Both endpoints need to be simultaneously present Only support short disconnections Only work with TCP Session layer solutions: DHARMA, PCMP Provide both mobility and disconnection transparency Both endpoints need to be simultaneously present Only work with TCP Application layer solutions: Message ferrying, Gnutella Application specific Do not deal with control overheads

Tetherless Computing Architecture

Disconnection tolerance using DTN Sender Receiver DTN Overlay DTN router

DTN is just like but… Built-in support for disconnection Reactive fragmentation Both ends can move Better routing Dynamic, instead of MX records Can be based on opportunistic or scheduled links General purpose API Send()/Recv() like sockets Support for multiple classes of service Better identity management We hope… Avoid spam

Mobility transparency using a DHT Sender Internet Region DHT Overlay Receiver at R-1 Receiver at R-2 Mobile AP

Overview Bundles sent to DTN addresses with GUIDs Location management DHT stores translations from GUID to DTN region Within a region store translation from a GUID to custodian Late binding from GUID to destination region InternetRegion (R)Custodian (C)Mobile (I) (I, R)(I, C)TCP Local DTN router (M) TCP HLRVLR L LR Fig. 1: Three tiered hierarchy of lookups in TCA

Details When a bus drives past a mobile, it uses GUID to deliver bundles to the mobile, and picks up late- bound bundles from the mobile DTN routers with an interface on the Internet resolve late-bound addresses to a destination region gateway using DHT Destination region gateway maps from GUID to the custodian and forwards bundles to the custodian Mobile picks up bundles from the custodian After a move, tables have to be carefully updated ‘Make then break’

Features Sender can inject all its data into the network, whenever it gets a chance Receiver can extract all waiting data from the network, whenever it gets a chance Far notifications not needed in near movements Short latency stretch

Innovations Unbound destination address allows disconnected endpoints Lookup in the forwarding path GUID is MD5 hash of address Provides security without PKI or DNS DHT for HLR Makes lookup scaleable and robust Compatible with DTN routing No extra overhead on headers or routing tables

Ongoing research Solving underlying point problems Building a prototype Building real-world tetherless applications

1. Fundamental underlying problems How to manage identities of mobile disconnected users? How do you know you’re in a hotspot? How to boost priority of communicating processes?

Identity management (Seth) Use Hierarchical Identity Based Encryption Public Key = Hash (identity) Each user must get Private Key from a PK server Allows disconnected operation Can limit need for revocation using timed keys Working on a secure architecture for tetherless computing

Hot spot detection (Lifchits and Liang) Monitor signal/noise strength of incoming beacons sent out once every 100 ms Hard to make the right choice We are still collecting raw data sensitivity to position sensitivity to orientation sensitivity to motion Is signal strength the right indicator?

Priority boosting (Liang) Simple idea Boost priority of communicating processes when in a hotspot Complex implementation! We have implemented in Linux 2.4, Linux 2.6 and FreeBSD Each OS has its own quirks and needs a lot of careful kernel hacking Can show some improvement for each OS by using a combination of priority boosting, quantum size boosting, and memory locking

2. Building a prototype Have successfully installed DTN on a variety of platforms (Darragh and Thomas) Have successfully integrated DTN with Internet Indirection Infrastructure (Darragh) First app is a ‘mobile blog’ (Fung and Darragh) Working on rewriting DTN bundle daemon (Ahmed)

Hardware platforms Soekris single board computers x86 compatible, running Pebble Linux powered by lead acid cell runs DTN bundle daemon and I3 client Sensor motes integrated with DTN mote proxy Sharp Zaurus 5500 PDAs Somewhat flaky hardware, but works OK most of the time IPAQ 4150 nice platform, but Linux support is missing investigating CELib

3. Applications Medical applications (Hilliker) working with Guelph General Hospital early stages -- still learning context Plant operations (Hilliker) simplifying workflow UW Formula SAE racing team (Hilliker and Salmon) working with on-board DAS to send information to the pit

Course projects Integration with Jabber Formula SAE Mobility VMPod

Conclusions Tetherless vision will take years to realize Attacking on several fronts Progress is encouraging More at