Authors: David Garlan, Daniel Siewiorek, Asim Smailagic ...

Authors: David Garlan, Daniel Siewiorek,

Asim Smailagic and Peter Steenkiste

Slides by: Marco Maurier

New problem: limited human attention

Limited human attention refers to users ability to attend to primary tasks, ignoring system-generated distractions such as poor performance and failure

Project Aura is a system whose effectiveness is considerably greater than other systems today

Designed for pervasive computer environment involving wireless communication, wearable or handheld computers, and smart spaces

Research spans every system level: from hardware to operating system

(1) Uses proactivity which is a system layer’s ability to anticipate requests from a higher layer

Today’s systems merely react to the layer above it

(2) Self-tunning which is when layers adapt by observing the demands made on them and adjusting accordingly

Current systems are staticBoth of this techniques will lower

demand for human attention

The hardware technologies are readily available

Research is needed on building blocks of pervasive computing and seamless integration

Project Aura is two years old and the development is on system architecture, algorithms, interfaces, and evaluation techniques

Odyssey – resource monitoring and application-aware adaptation

Coda – nomadic, disconnectable, and bandwidth-adaptive file access

Spectra – adaptive remote execution that uses context to best execute remote call

Prism – new system layer that captures and manages user intent

Amplify the capabilities of mobile clientThese compute and data-staging

servers are called surrogatesData-staging reduces Internet latency

on interactive file intensive applicationsCache misses can cause delays due to

latency limitation, and also because: (1) cache may be small (2) periods of disconnect(3) uncached files may become needed

Coda extended to use prefetchingStages data in coarse-grained

snapshots of the file system dataEach snapshot corresponds to a

volume, which is a partial subtree of the file system name

Data-staging split across server, surrogate, and client

Orange components represent preexisting distributed system

Yellow represent the modification implemented to support data-staging

Staging server on the Surrogate is an unmodified Apache HTTP/1.1 Web server. The proxy intercepts and redirects file system traffic

Staging manager on the Coda server oversees snapshot creation

Network aware applications can use reasonable estimates of future available to make informed decisions

Application affected by:(1) competing traffic, noise, and

interference(2) device physical location

Monitoring(1) and Prediction(2) use SNMP(simple network management protocol) to gather information

(1) AP Segment Service collects incoming and outgoing traffic rates, including errors and collision rates

(2) AP Device Service gathers cell population information by querying each access point bridge table

PPREV – predicts future values to be the same as the most recent value observed

AV – (average value) uses evenly weighted average of the several previous observations

Arfima – computes future values as a dynamically weighted average of past values

Arfima – works well in low-utilization cells, not high-utilization cells

PPREV – works well in bursty traffic, high-utilization cells

The fact that different data rates are used depending on noise conditions, complicates predictions

Based on signal strength and access point information from IEEE 802.11

Two algorithms:(1) CMU-PM – pattern matching

algorithm(2) CMU-TMI – triangulation-based

remapped interpolated algorithm

Determines the location by measuring the signal strength from device to all available access points

Compares these measurements to a table with reading of signals strength for each location

Training requires entering the user’s location into the system

Radar algorithm gathers signal stregth from access points

Algorithm takes the signal strength and stores in a table

During use, compares measure values to those in the table and computes the difference

It assumes the entry with the smallest difference is the current position

Accuracy almost always within15 feet with dense table of patterns

Goals in design: accuracy within a few feet, and scalability so as to not require a trained data point but interpolation between trained values

Calculates client position on a continuous coordinate grid assuming that the signal strength map directly to distance

This is interpolated to trained values and produces more accurate results

CMU-TMI requires less training points are is more accurate at higher distances

CMU-PM give larger errors because of incorrect training point being returned

CMU-TMI returns continuous small errors but return location not far form user’s actual location

Environment can affect output such as reduction of signal strength from user’s antenna, or office walls

When user moves from one environment to another, Aura attempts to reconfigure the new environment so that the user can continue working on tasks started elsewhere

Aura needs a representation of user intent, a new layer of system abstraction called the task layer

This allows the system to anticipate user needs

Task Manager – explicit representation of tasks as coalition of abstract services

Context Observer – configures tasks in a way appropriate to the environment

Environment Manager – assists with resource monitoring and adaptation

Service Suppliers – supports user ‘s tasks

This is built on top of middleware such as RPC or Corba

Represents user tasks explicitly and as a service coalition

Providing an abstract characterization of the services in a task

Providing an environment with self-monitoring capabilities

Current prototypes support migrating tasks between Windows and Linux

One application that uses the Aura infrastructure is Handy Andy

It has been design with the goal of supporting on-campus collaboration

Its core applications are Portable Help Desk (PHD) and Idealink

Context aware application built on two fundamental services:

(1) spatial awareness – includes users relative and absolute positions an orientation

(2) temporal awareness – schedule time of public and private events

It has both a visual and an audio interface

Each interface supports users in different contexts

If user is walking around campus is less likely to be distracted with a hands-free interface

While a stationary user might want a richer visual interface

It delivers proactive and user-driven information to users when they are interacting with Aura infrastructure resources, such as a printer

When a user starts a print job, PHD alerts the user if there is a large print queue and suggests a nearby printer with a shorter one

Virtual collaboration environment that facilitates planned and ad-hoc collaboration among mobile users

It provides users with features that let them communicate their ideas to others via a shared distributed whiteboard

Supports standard pen and text tools, multiple channels, and simultaneous collaborative sessions

Integrated to PHD to retrieve information related to user’s preferences and schedules

Combines each user’s additions to the session and distributes these updates to each client.

At the end of the meeting archives the session

Goal to develop a non-trivial prototype that the campus community can use

Focus on:(1) set of contextual information

services(2) applications that exploit the Aura

infrastructureUse dynamic information for Aura to

perform self-tuning and adaptation