[attachment=8385]

Abstract
This paper introduces a new class of application called monitoring applications, These
applications show dier substantially from conventional data processing. The fact that
a software system should process and react to, inputs in continuous manner from many
sources like sensors rather than from operators requires one to rethink about the fun-
damental architecture of a DBMS for this area of application. In this paper, I present
Aurora, a new model for DBMS that is constructed at Brown University,Brandeis
University , and M.I.T. I describe the basic system architecture, set of operators,
optimization methods, and support for real- time operation.

Chapter 1
Introduction
The focus of Traditional DBMS on business data processing. They are designed to
meet these businesss reqiurements. Aurora is a new model for DBMS which is used
manage data streams. This model is mainly useful for monitoring applications. The
question comes if there are any particular reasons why we need a new system instead
of using the traditional DBMSs. This paper gives ve reasons as following:
1. Wrong basic computation model. Traditional DBMSs are a human-active, DBMS-
passive (HADP) model. A DBMS as a passive repository storing a large amount
of data elements. Aurora system is a DBMS-active, human-passive (DAHP)
model. The monitoring application gets the data from external data source.
The system noties humans when abnormal activity is detected.
Traditional DBMSs store time-series information is a challenge. Traditional
DBMSs care about only current value. Monitoring applications require both
current records and some history of values.
2. Most monitoring applications are trigger-oriented.They cannot scale a large
amount of triggers per table. Traditional DBMSs treat triggers and alerters
as second-class citizens.
3. Traditional DBMSs can't answer approximate query.It assume that data are
synchronized. Therefore, the results of querying are exact. But in data stream-
oriented applications, data comes asynchronously and are often lost, stale, or
intentionally dropping for processing reasons (e.g. load shed).
4. Monitoring applications also have real-time requirements. Traditional DBMSs
assume that the applications have no real-time requirements.

Chapter 2
Aurora system model
Mainly there are two kinds of data sources in Aurora. Incoming valuse at regular or
irregular coming from program of a computer. The other is from hardware devices
like sensors. A unique source identier is used for every source. A timestamp is given
by Aurora system for incoming tuple. A data stream is made up with several such
data sources. They look like as following:
StockID, Time,Price ID, Time, Position
(MSF, 1:00, 20) (1; 2 : 00; 3)
(IBM; 1 : 00;16) (2, 2:00, 5)
Aurora system is mainly used to process incoming data streams in the way dened
by an application administrator. The fundamental elements in Aurora system are
boxes and arrows that are similar to those in process
ow and work
ow systems.
Boxes represent processing operations. Data stream, in terms of tuples,
ow through
a loop-free, directed graph. Output streams are sent to applications, which must
be implemented to process the asynchronous tuples in an output stream. Arrows
represent a collection of streams with common schema. Fig.1 illustrated the high-
level system model.

Aurora: a new model and architecture for data stream management
SHAFEEK T K
2K7744

Overview
 Monitoring applications
 Why not traditional database?
 Outline of a new model-AURORA

One-line statement
 A new DBMS model and system
specialized in data stream management
for monitoring applications

Emerging Applications
 Military application
 Financial data analysis
 Sensor networks monitoring
 GPS Tracking applications
 Network traffic management.
 Web logs

Sensors of a Motors car
Telematics agent
collects and
transmits data to
center All parts are
RFID tagged
Armed various sensors
RPM,
temperature,
pressure, oil
status, …
Pressure,
exchange
date, …
Brightness,
…
Telematics agent can
test the car and report
malfunctioning part ids

Data stream
 Continuous dataflow  streams
 Unbounded sequences of data
 Autonomous Data Sources
Eg: sensors, mobile devices , system

[attachment=8387]