More Related Content Similar to Patel-Paper Review Similar to Patel-Paper Review (20) Patel-Paper Review1. Linear Road: A Stream Data Management Benchmark.
A. Arasu, M. Cherniack, E. Galvez, D. Maier, A.
Maskey,E. Ryvkina, M. Stonebraker, R. Tibbetts.
VLDB Conference, Toronto, Canada, 2004.
Your name : Nabilahmed Patel
Email: nabilpatel11@gmail.com
2. References
A. Arasu, M. Cherniack, E. Galvez, D. Maier, A.
Maskey,E. Ryvkina, M. Stonebraker, R. Tibbetts. Linear
Road: A Stream Data Management Benchmark. VLDB
Conference, Toronto, Canada, 2004.
http://www.cs.brandeis.edu/~linearroad/linear-road.pdf
Sharma Chakravarthy 2
4. Talk Outline
Summary of the paper
Challenges in designing benchmark due to stream data.
Benchmark requirements.
Implementation and Experiments.
Strong and weak points
Why you think it got accepted
How to extend/improve the work
conclusions
Others
5/15/2016 © your name 4
5. Summary of the paper
Due to unbounded and continuous nature of stream data, input data
introduces some unique challenges
1. Semantically Valid Input
2. Continuous Query Performance Metrics
3. Many Correct Results
4. No Query Language
There are some specific ways in which Linear Road addresses this
challenges.
1. The input data is generated by traffic simulator, MITSIM.
2. Response Time and Support Query Load (L-rating)
3. Validation for each of queries considers all possible valid
answers.
4. Queries are specified formally in the “predicate calculus”.
5/15/2016 © your name 5
6. Summary of the paper
Understanding of Data
1) Position reports: (Type = 0, Time, VID, Spd, XWay, Lane, Dir, Seg, Pos)
2) Account Balance: A request for the vehicle’s current account balance,
(Type = 2, Time, VID, QID)
3) Daily Expenditure: A request for the vehicle’s total tolls on a specified
expressway, on a specified day in the previous 10 weeks.
(Type = 3, Time, VID, XWay, QID, Day)
4) Travel Time: A request for an estimated toll and travel time for a journey
on a given expressway on a given day of the week, at a given time.
(Type = 4, Time, VID, XWay, QID, Sinit, Send, DOW, TOD)
5/15/2016 © your name 6
7. Summary of the paper
To avoid the complication of unpredictable event delivery order,
The four types of input tuples are multiplexed together into a single stream
of tuples consisting of the union of all fields.
In order, these are:
(Type, Time, VID, Spd, XWay, Lane, Dir, Seg, Pos, QID, Sinit, Send,
DOW, TOD,Day).
Linear Road implementations can use the Type field to determine which
fields are relevant for a given tuple.
5/15/2016 © your name 7
10. Summary of the paper
Historical Queries:
5/15/2016 © your name 10
13. Summary of the paper
Implementation:
The historical data generator is run to generate flat files consisting of
10 weeks worth of historical data.
The traffic simulator is run to generate L flat files, each of which
consists of 3 hours of traffic data and historical query requests from
vehicles reporting from a single expressway during rush hour. The data
driver is then invoked to deliver this data in a manner simulating its
arrival in real-time.
The system running the benchmark is configured to generate a flat file
containing all output tuples in response to the queries defined in the
benchmark.
The validation tool is used to check the response times and accuracy of
generated output to see if they meet the requirements of the benchmark.
5/15/2016 © your name 13
14. Strong points
The problem is defined and explained in detail with the real world
example.
The simulation and presentation of stream data is given extensively,
which makes understanding of queries very easy.
The queries are defined in “predicate calculus”, hence everybody can
understand and can implement in their systems using their own query
languages.
5/15/2016 © your name 14
15. Weak points
Some details are repeated again and again in different sections.
The “Travel Time Estimation query” is not supported by
implementation of Benchmark as it is too complex to expressed.
5/15/2016 © your name 15
16. Why do you think it got accepted?
Linear Road Benchmark was the first attempt to compare the
performance characteristics of SDMS systems.
The Paper simulates the example of real world into the design of
SDMS by creative thoughts, which makes it easy to understand.
It is creative thought on how to meet the challenges of large scale
streaming data applications.
5/15/2016 © your name 16
17. Possible Future Work
Time Travel Query is not supported by current implementation of
benchmark, so it can be improved to response this kind of query.
The paper is comparing SDMS with RDBMS, but by following the
procedure it is possible to compare two different SDMS.
5/15/2016 © your name 17
18. Conclusions
The Paper describes how the challenges introduced due to nature of
stream data can be outlined using design of Linear Road Benchmark.
It also covers how the stream data is simulated and generated using
traffic simulator.
Two continuous and two historical queries are defined in predicate
calculus.
Implementation of Linear Road Benchmark in Aurora (SDMS) and
System X (RDBMS) is covered.
Results of experiments shows that SDMS can outperform a Relational
Database system in processing stream data by at least a factor of 5.
5/15/2016 © your name 18
19. Others
Readability/presentation
It is well explained and extensively described with experimentation
results.
It is easy to read.
It is well presented with detailed explanation.
Technical depth
Queries are in predicate calculus.
Stream data is well defined.
Novelty
STREAM (SDMS) also has implemented Linear Road Benchmark.
This benchmark makes it possible to compare the performance
characteristics of SDMS’ relative to each other and to RDBMS.
Overall comment
Experiments has proven that SDMS outperforms RDBMS at least
by the factor of 5.
5/15/2016 © your name 19