Services for Mobile Users • Mobility was the requirement of the 90’s, first in communications and then in computing. • rapidly growing demand by users • many interested players: • equipment manufacturers, infrastructure and service providers • Current technology (devices, access) makes mobile computing feasible, but present support for it is limited.
Introduction to Mobile Computing • Various definitions of mobile computing: • not the same as wireless computing • nomadic (orlocation-independent) computing • Our interest is in supporting users who work from multiple locations, and whose means of “connection” to their home system may take different forms at different times. • Emphasis to date has been on functionality, with little attention to performance.
What Mobile Users Want • Seamless mobility • “connect” from any location, at any time • convenience of use (no extra setup, “plug and play”) • same computing environment, same services, consistent interfaces, regardless of location • Mobile users may be willing to sacrifice some performance for mobility, but only some.
Application System Services Network Services Transport Medium The Mobile Computing Stack Mobile User
Technical Challenges • Networking Challenges • communications issues: protocols (old and new), technologies (old and new) • accommodating host relocations • network services to mobile users (e.g., mobile multicast) • Operating System Challenges • OS support for mobility oriented devices (e.g. intermittently powered hard drives, limited resources) • OS services for mobile clients (e.g., to ensure data availability, data integrity)
Technical Challenges (continued) • Other Challenges • device design: size and weight, usability • energy conservation • security, authentication, authorization • application development • . . .
Recent Research Projects • Accommodating mobile host relocations • with Carey Williamson, Vineet Chikarmane, Wayne Mackrell • Multicast support for mobile hosts • with Carey Williamson , Tim Harrison, Wayne Mackrell • TCP over wireless links • with Venkat Josyula • File system support for mobile users • with Kevin Froese
Accommodating Mobile Host Relocations • The problem: • IP routing is based on the network component of a host’s IP address, which is bound inextricably with its location. • Moving to a new location means acquiring a new IP address and then informing all “correspondents”. • Roaming must be handled on an ad hoc case-by-case basis (by individual users, system administrators, or both). • Mobile IP aims to provide for seamless relocation by providing services to mobile users as if they were at their home network.
Mobile IP: An Emerging Standard • Features of Mobile IP: • Separates “location” from “address”. • No new IP addresses or address formats required. • Only “mobile aware” routers and mobile units need new software. Other routers and hosts use current IP. • Impact of Mobile IP on users: • Can take any computer to any location; routing of communications from correspondents is done “automatically”. • Services provided as if at home network.
Mobile IP: How It Works • Mobile unit registers with the foreign network upon arrival. • Home Agent and Foreign Agent cooperate to deliver IP datagrams to the mobile unit. • forwarding caches at both agents • IP-in-IP encapsulation • Mobile unit deregisters (explicitly or implicitly) upon leaving foreign network.
IP header To: care-of address IP header To: mobile node datagram payload Datagram forwarding • HA tells local nodes and routers to send MN’s datagrams to it • HA intercepts datagrams intended for MN, then encapsulates and forwards them to MN’s care-of address • FA receives encapsulated datagrams, then decapsulates them and delivers them to MN
Integrating Wireless Access • What are the implications of integrating wireless connections into the internetworking fabric? • Our focus was TCP, with emphasis on short range connections: • tests of functionality and performance by experiment and simulation.
Wireless Computing • Existing wireless technologies (such as infrared, radio or cellular) can be employed for signal propagation • Can provide for tetherless computing • Wireless links are characterized by • higher error rates, more lost packets, longer delays • For wireless links to integrate seamlessly into the internet, TCP must work well over wireless connections since TCP/IP is the basis for many current network applications
TCP in a Wireless Environment • Problems with TCP in a wireless environment: • TCP congestion management: uses loss as congestion indicator • TCP timers: use delays for timeouts and retransmissions • Proposed solution: • sender manages end-to-end packet transmission • a (transparent)proxy looks after loss on the wireless link • caches packets from sender for transmission over wireless link • performs retransmissions of dropped packets • ACKS from receiver flow through to sender • sender retransmissions reduced • TCP semantics preserved
Sample Measurement Results • Retransmission Time-out Behaviour
The Proxy Model Sender Receiver Proxy
Sample Simulation Results • Impact of proxy on end-to-end throughput Proxy ON Proxy OFF
Summary of Findings • Design decisions within TCP present problems when applications run over wireless (lossy) links. These problems have a profound impact on end-to-end performance of the application. • While proxy solutions cannot affect the loss, they can control TCP’s response to it and thus improve end-to-end performance.
File System Support for Mobile Clients • Location-independent computing characterised by • disconnection, movement to a new working location, reconnection • type and quality of connection (to home file server) varies • Mobile users want access to remotely stored files, regardless of current type of connection. • this research is focused on maintaining acceptable file access performance across different forms of connection. • The ultimate distributed file system • File caching at the client is fundamental to any solution.
File Caching for Mobile Computing • Goal is to provide effective file system service to mobile clients. • Optimistic caching of file replicas at the client is a key to availability. • Project considered impact on performance of • configuration issues, at the client and on the network • cache management strategies • demand characteristics
Strongly Connected Operation • a high-bandwidth connection is available, over which read and write operations are serviced • file caching can improve performance (by reducing latency) • the conventional distributed file system
Disconnected Operation(CODA file system) • no connection to home file server • users optimisticallyhoard replicas of desired files prior to disconnection • all file operations processed in the cache • read misses are fatal • updates to file system are logged at the client • upon reconnection, replay of logged events reintegrates changes with home file system
Weakly Connected Operation • a low-bandwidth connection is available • read misses are no longer fatal • asynchronous write backs provide for reintegration of logged changes with home file system, but must share the bandwidth available with reads • reads should have priority
Project Objectives • To investigate performance issues relating to mobility-aware file caching using trace-driven simulations. • configuration parameters: • cache unit, cache size, bandwidth available • policy parameters: • what to write, when to write, read/write scheduling • performance measures
Lotto NoWB, PWB Effect of write-back policy (10 MB cache) Resource tradeoff Sample Results
Summary of Findings • It is possible to provide quite acceptable remote file service to weakly connected mobile clients, even when very little bandwidth is available. • Reads can be serviced in a timely manner. • Even very simple write-back policies can provide timely reintegration. • Requires only reasonably sized caches at the mobile client.
The Future: Wearable Computers http://www.media.mit.edu/wearables/ A whole new meaning to the term “smarty pants”
Concluding Remarks • Mobile functionality is available now, but performance remains an issue. • What the future holds: • Better devices for mobile users. • Seamless and transparent mobility. • Better mobility infrastructure. • Mobile IP everywhere: foreign agent capabilities at conference sites, hotels, airports, ... • widespread support for wireless access: base stations on many networks • But, there’s still much work to be done to get us there.