Download 1 - School of Information Technology

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Nonlinear dimensionality reduction wikipedia , lookup

Transcript
Nilanshu Dharma & Shalva Singh
Position Paper
5/4/2017
Databases for Moving Objects
Introduction
There has been a remarkable advancement in several location based services technologies with the growth
of mobile devices. Global Positioning System (GPS) which makes use of a network of satellites, provide
aid to navigation, land surveying, and scientific studies of various kinds by determining receiver’s location,
directions, and speed. These functionalities are used by Location Based Services (LBS) for tourists, mobile
commerce, digital battlefield, and emergency responses which involves tracking of the transient location of
a mobile caller or a vehicle, also termed as Moving Objects (MOs). (Behr, Almeida, & Guting R.H, 2006)
To store and manage the voluminous and incessantly changing data of millions of MOs, it has become
inevitable to devise scalable data management system which would deal with data mining, location
propagation, privacy, and synchronization, efficiently. (Behr, Almeida, & Guting R.H, 2006)
We present here an analysis of three methods to store and query Moving Object data. First we discuss the
issues related to moving object’s databases. Each following section provides a description of the strategy
used and its claims. Finally we try to evaluate each strategy on some common grounds and take a position
supporting one update for Moving Object at a Timestamp. This is followed up by introduction to a new
idea where we propose a new hybrid model for storage of moving objects.
The Challenges
Important issues involved with Data Management of MOs are modeling of location information,
uncertainty management, spatio-temporal data access languages, indexing & scalability, data mining,
location dissemination, privacy of data and location fusion & synchronization. The database of location
networks should support point query as well as point update query. Point query involves locating a MO
with a certain key. Point Update query is used to update the current location of a MO with a key. The
challenges and issues involved with MO’s location management are distributing, replication, and caching
of database for efficient execution. The issues to be addressed also involve- how to search database and
1
Nilanshu Dharma & Shalva Singh
Position Paper
5/4/2017
how frequently the database needs to be updated. If the updates are done via a network then the
resource/bandwidth tradeoffs and factors such as cost also need to be addressed. (Ouri, 2002)
Strategy 1: Moving Object Management System (MOMS) Based on a File
The author has proposed a moving object storage system based on file system. This system stores both the
current location and the past location of the moving object to store and search data efficiently, as location
of MOs change intermittently. MOMS’ architecture consists of three major components, namely Query
Processor Component, Location Storage Component, and Index Component. The additional module like
Gateway is used to acquire current location of MOs with the help of GPS etc.
Location Query Component carries out query depending upon MO’s model and its operator. Index
Component comprises two indexes simultaneously- Current Location Index Component (CLIC), that takes
only current locations into consideration and Past Location Index Component (PLIC) which processes time
interval and trajectories queries. Location Storage Component is used to store MOs and search the ones that
associate with query results of location. CLIC adopts the approach of spatial based indexing on current
location information and object based indexing on MO Identification. On the other hand, PLIC manages
spatiotemporal index about the past location information.
2
Nilanshu Dharma & Shalva Singh
Position Paper
5/4/2017
Figure: Past Location Index and Current Location Index (Cho & Jang, 2007)
The design of the File based system includes Connection Manager that provides the function of connecting
with the client to activate SQP in order to process client’s request. SQP has access to client’s information
therefore it analyzes and processes client’s request query. Buffer Manager connects to the File Manager to
manage most recent location information using MOID. Index Manager creates index file and uses MOID
and MOTIME as keys. (Cho & Jang, 2007)
Strategy 2: One Update for all Moving Objects at a Timestamp
Even with use of indexing techniques, efficient management of large data is still a problem. Also vast data
updates which occur at different times is challenge to manage. The authors suggest an updating technique
applied for indexing methods developed from R-Tree. It proposes to update the indexes at one time. The
authors claim to achieve increased quality of queries.
R-tree as we know is a height balanced external memory data structure. It is an efficient method for
indexing, but requires deletion of obsolete state and then insert new state in top-down manner. The features
of this approach are as follows:

Support for both deletion and update queries.

Updates process for all new states at one timestamp, which means it tries to access a disk block at
most once in a process.

It does not deteriorate the quality of the tree while providing improved performance.

It is not dependent on a specific type of new data distribution.

Capacity of main memory used in algorithm is not large and can be easily estimated.
Let us take a closer look at the deletion and update query processing. The deletion takes place from the leaf
level, i.e. deletes all the obsolete states at leafs using a parent_of pointer. This also saves memory as instead
of loading the entire tree, only the pointer is needed. For the insertion process the rule is, if leaf node is
underflow the process will not reinsert its entries immediately instead it would move them into a stack in
main memory for being inserted together with insertion process. If internal is underflow normal insertion
process is used.
3
Nilanshu Dharma & Shalva Singh
Position Paper
5/4/2017
Figure: Example of use of information table and parent_of pointer (Tung & Ryu, 2006)
The authors have conducted experiments to test the claims they make. The experiments aim to compare
update and insert query performance compared to other R-Tree update methods. The algorithm proposed
outperforms its competitors in two sets of experiments conducted. One was update queries randomly
generated for set of 10,000 cars for timestamps 1 to 4 at rates 1% and 5%. Other experiment was on
different data sizes, 5k, 10k, 20k and 30k cars. Updates were taken at 1% and 5% rates and the algorithm
proved to give most stable results for all loads. (Tung & Ryu, 2006)
Comparative analysis: Both the strategies given above are unique methods. But we take a position that
method using R-Tree is better approach. It gives a detailed organized algorithm to store and retrieve
indexes. The experiment results are quite convincing to convey the claim. The R-Tree model is scalable
and consistent in performance. Also it is less cumbersome in terms of resource use as compared to file
based location storage. (Tung & Ryu, 2006)
Hybrid Model: We also propose a novel approach where a model can be designed which incorporates
features based on heuristics. A problem exists with moving objects exists. If no update is received the
position of the object cannot be declared (Ouri). We want to extend the concept given by Ouri to the one by
Tung, Ryu. We propose a model which would use the past information from a moving object to predict its
current location. Incorporating such “intelligence” would help further reduce the use of database resources
and improve efficiency of the entire system. This model would be implemented on the one update at a
timestamp concept.
4
Nilanshu Dharma & Shalva Singh
Position Paper
5/4/2017
Conclusion: We reiterate that one update at a timestamp is a better database approach than index file
method. We also propose that it would be beneficial if this concept is used under a model which also uses
heuristics to determine the position of an object even if no update is provided. This model would work best
for objects whose path is predetermined.
References:
Behr T, Almeida V.T, & Guting R.H. (2006). Representation of Periodic Moving Objects in Databases.
ACM GIS’06.
Cho D.S, Jong I.S Location Information Storage System Based on File. Retrieved February 15, 2007, from
www.isprs.org/istanbul2004/comm4/papers/544.pdf
Tung H.D.T, Ryu K.H. (2006). One Update for all Moving Objects at a Timestamp, Proceedings
of The Sixth IEEE International Conference on Computer and Information Technology (CIT'06).
IEEE Computer Society.
Ouri W. (2002) Moving Objects Information Management: The Database Challenge (Vision Paper),
Proceedings of the 5th International Workshop on Next Generation Information Technologies and Systems,
pp 75-89.
5