Page 23 - profile2014.indd

periodically publish sensor measurements as updates to create
new versions installed into an embedded database system. Various
types of application queries are submitted to retrieve multiple ver-
sions of data items maintained in the embedded database. Thus,
maintaining multi-versions of data are important to many CPSes for
detecting the operation environment, detecting critical events, or
collecting statistics about the dynamics in the operation environ-
ment. Therefore, it is critical to facilitate access performance to the
current and the previous data versions maintained in an embed-
ded database. At the same time, embedded systems usually adopt
non-volatile memory, instead of hard disks, as their storage devices.
Thus, the characteristics of the embedded storage media/systems
should be also considered.
Our research focuses on proposing index designs to facilitate ac-
cess of multiversion data items in embedded database systems.
Due to the dynamic nature of the entities in CPSes, the proposed
With information on user preference of products, we propose a multiversion index designs can support a large number of update
new optimization problem to select suitable products to configure transactions, and can efficiently support key-range query and
a bundle, such that the number of users adopting this bundle is version-range query transactions, which are critical functions of
maximized. We will continue to explore the effect of social influ- multiversion database systems. We also explore the impacts of us-
ence, and propose efficient algorithms for real applications.
ing different non-volatile memories (e.g., flash memory and phase-
change memory) as the storage media in embedded systems. In
Most of the literature on social graph query focuses on only the particular, we are exploring potential solutions to remove inherent
social dimension, and ignores the real social applications that are problems with hard disks, and to take advantage of the special
necessary to examine multiple dimensions. It is expected that characteristics of these non-volatile memories. Various technolo-
searching for people based on social, temporal, spatial, and prefer- gies, such as mining algorithms and data deduplication, will be
ence dimensions will be very challenging, but has the potential to studied and redesigned to fully utilize the capability of non-volatile
deliver useful applications in everyday life. Nowadays, most social memories. The inherent endurance and reliability issues of these
group activities are still initiated manually via phone, e-mail, mes- non-volatile memories will be also investigated to reinforce their
senger, etc., even with the availability of ripe web collaboration usefulness.
tools (such as Google Calendar) and socio-spatial information web-
sites (such as Facebook Places). To address this issue, we propose a
group query to return a group of familiar individuals and their com- 4. Location-based Data Collection and Application Deployment
mon available time slots. Different social familiarities can be flex- Platform
ibly specified for various types of social activities. The proposed al-
gorithm can find the optimal solutions at least 20 times (100 times Location-based data embeds useful information to be mined to
on average) faster than the commercial parallel CPLEX optimizer support or enhance various applications (such as route selection),
in the same server. Moreover, for location-based social networks, or solve difficult location-based problems (such as city profiling).
such as Loopt, Buddy Beacon, FindMe, and Facebook Place, we for- However, it is challenging to collect large volumes of data from us-
mulate a social spatial group query with a new index structure to ers for various reasons. In this project, we propose the use of the
efficiently find spatially close-by friends with tight social relations. PLASH platform to help location-based service (LBS) providers de-
We have also explored the group query to consider the user pref- ploy their applications conveniently; users can contribute their ef-
erence dimension. We propose a randomized algorithm based on forts and location-related data by using the services, which is the
the concept of Optimal Computational Budget Allocation (OCBA) main difference from other location-aware services.
and Cross-Entropy (CE) to efficiently obtain results. Our algorithms
are implemented in Facebook to validate the effectiveness of que- The PLASH system provides a GUI to allow users to construct their
ry results. We will continue to formulate new query problems and LAS applications, in such a way that the process generates pro-
design efficient query optimization algorithms and techniques for grams on both smartphone and server. It also allows users to do-
finding the optimal or approximate solutions in a small time. nate software components to be mashed up by other users in their
applications. However, it has unavoidable and inherent security
3. Index Designs for Embedded Databases problems, as well as other system risks. PLASH considers scalabil-
ity and compatibility, and we have been using road network route
Embedded database systems are widely adopted in cyber-physical selection application to exemplify the performance characteristics
systems (CPSes) to maintain and monitor the status of entities in and design goals of PLASH.
deployed operation environments. Sensors or devices in CPSes

23