1. Multimedia Data Stream Management System
By David Kleinman

2. Outline Definition Motivating Examples Nine Requirements
Current Systems
Comparison
Brief Overview of current Stream Systems
Preview of My Project

3. What is it? Stream of multimedia data from a source (video camera)
Query stored in a system (This query may itself change
Process high volumes of data in real-time

4. Motivating Examples Security Surveillance Baby Sitting Traffic Reports
Crowd Security
Air Security
Burglary
Baby Sitting
Traffic Reports
Science
Animal behavior
Ocean
Sensors can be expensive – especially when applied across a large area. It is also difficult to install (i.e. Washington D.C. do not have right to install on other’s buildings.
Sensors are not 100% dependable. Sensors can be disabled easily. They are often located on the ground and are easily accessible. A video camera can be stored in one safe location high above ground. Sensors cannot detect items that a video camera can detect.

5. Reqirement #1 - Process Quickly
Low latency
Messages Processed “In-Stream”
No Storage to perform operation
Active System
Avoid Polling
Low latency – the system must be able to perform message processing without having a costly storage operation. Storage adds latency to process. – i.e. writing to a database requires a disk write –
Passive systems wait to be told what to do by an application before beginning processing. Passive systems require applications to continuously poll for conditions. Polling results in additional latency because on average half the polling interval is added to the processing delay.
Active systems avoid this by having built in event/data driven processing capabilities

6. Requirement #2 – Query using SigmaQL for Streams (StreamSigmaQL)
Querying Mechanism
Based on SQL
Express Continuous Streams of Data
Window Construct
Time
Frames
Breakpoints
Merge Operator
SQL has remained the most enduring standard database language for over 30 years because it is very good at expressing complex data transformations. It is based on a set of very powerful data processing primitives that do filtering, merging, correlation, and aggregation.
Also SQL is widely understood and used by programmers. The language should be easy to learn
Windows should be definable over time, number of frames, or breakpoints in other attributes in a message. Windows should be able to slide a variable amount. Depending on slide amount windows can be made disjoint or overlapping
A merge operator is needed to join multiple streams

7. Requirement # 3 –Handle Imperfections
Data might be late delayed, missing, or out-of sequence
Time out individual calculations or computations
Challenges with Dealing with out-of-order data
Mechanism for additional time
Networks aren’t reliable
Let’s say computing average number of people in ten rooms. One of the cameras in a room is broken. You don’t want the system to block waiting for a result that will not come. A time out system is a must
Let’s say you have a time window 9:00 – 9:01; Ordinarily after a timestamp greater than 9:01 is received the window will be closed. However this action assumes that data arrives in timestamp order which is not always the case.
To deal with out of order data, a mechanism must be provided to allow windows to stay open for additional period time.

8. Requirement #4 – Generate Predictable Outcomes
Generate deterministic and repeatable results
Time-ordered deterministic processing throughout entire pipeline
Important for fault tolerance and recovery
A stream processing system must process time-series messages in a predictable manner to ensure that the results of the processing are deterministic and repeatable
Soundtrack with movie – It’s important that the frames of the movie and the sound wav file are processed in the correct order
Time ordering is needed to guarantee correctness
Time ordering is needed for fault tolerance and recovery, as replaying the same stream should yield the same results

9. Requirement #5 – Integrate Stored and Streaming Data
Comparing present with past
Capability to efficiently store, access, and modify state information
A query may wish to include a picture with known terrorists
Finding unusual activity – requires gathering the usual activity patterns and comparing

10. Requirement #6 – Guarantee Data Safety
Must use a high-availability solution
Secondary System
Synchronizes with primary frequently
Takes over in case of failure
Mission critical information needs backup plan. If monitoring can’t have it failing.

11. Requirement #7 – Partition and Scale Automatically
Take advantage of distributed computing
Support multi-threading
Takes advantage of multi-processor
Avoids blocking
Load Balance across machines
Automatic process
Transparent

12. Requirement #8 – Process and Respond Instantaneously
Needs to respond in real – time
Highly optimized, minimal overhead execution path
All system components have high performance

13. Requirement #9 - Adaptability
Change queries without restarting
Accept all different types of multimedia streams
Allow for custom configuration
Work with different systems
API

14. DBMS Widely used Passive Do not keep data moving
Use SQL – but not equipped for Streams
Passive
Do not keep data moving
Difficult to handle out of order data
Difficulty with predictable out comes
Incur latency with seamless integration
Widely used due to their ability to reliably store large data sets and efficiently process human-iniated queries.
Passive – wait to be told what to do. Some have trigger mechanisms but Triggers are not scalable
Moving – require write to disk and then access – not real time
Difficult – trigger systems have no obvious way to time out.
Predicatable outcomes are difficult because they are passive

15. Rule Engine Example – Prolog Active Handle imperfections
Troubles with seamless integration
A rule engine typically accepts condition / action pairs – using if then notation – enforces a collection of rules

16. Stream Processing Engine
Handle all the requirements
Not specifically designed to handle multimedia constraints
Not Specifically designed to handle streams of multimedia

17. Chart DBMS Rule Engine SPE MSPE Keep data moving No Yes SigmaQL
Handle Imperfections
Difficult
Possible
Predictable outcome
High availability
Stored and Streamed data
Distribution and scalability
POssible
Real time
Adaptability

18. Aurora
DSMS developed at MIT and Brown

19. Aurora Query Network QoS .
Consists of operator boxes and connection points – storage points
Use QoS graphs to determine best path
Has built in scheduling, optimization and load shedding
Supports distributed environment

20. Stream Management System
Developed at Stanford
Uses synopsis and queues

21. Simple Query Plan Q1 Q2  ⋈ ⋈ State3 State4 Scheduler State1 State2
Stream3
Consists of queues which connect producer and consumer
Synopses – has tables at operators to store state
Has a scheduler
Stream1
Stream2

22. NiagaraCQ Developed at Wisconsin First DSMS Uses a grouping strategy
Not as complete as other two

23. System Architecture

24. TelegraphCQ Developed at Berkeley Stem – storage point
Eddy – route tuples
Good at handling multiple queries
Adaptive

25. Adaptivity (Telegraph)
Output
Queues
STeMs for join R
grouped
filter (R.A)
EDDY S
grouped
filter (S.B)
R x S x T T
Input Streams R S T
Runtime Adaptivity
Multi-query Optimization
Framework – implements arbitrary schemes

26. My Project Design a multimedia streaming database
Outline the specifications
The Scheduling algorithm
The query structure
The operators
Etc.