Database Reference
In-Depth Information
In the physical layer, the system is divided into different regions, each of which
includes a set of complete system components and complete copies of tables. The
data table is horizontally segmented into record groups, which are called Tablets.
Tablets are distributed among many servers; every server may have tens of thousands
of Tablets but a Tablet may only be stored in a region of a server. The query language
of PNUTS supports the selection and projection of a signal table. To upgrade or
delete an existing record, the main key words must be specified. The consistency
mode provided by PNUTS has a feature between the general serializability and
eventual consistency.
4.3.2
Design Factors
Of the various database systems, there is not a single system that can achieve the
optimal performance under all workload circumstances. In each database system,
some performance goals have to be compromised to achieve optimized operation
for specific applications.
Cooper et al. in [
20
] discussed the trade-offs confronted by data management
systems based on cloud computing, including reading performance and writing
performance, delay and durability, synchronous and asynchronous copies, and data
segmentation, among others. Some researchers also differentiated and analyzed
other design factors [
21
-
23
]. In the following, we compare several prominent
features of the existing database systems (rather than analyzing the design goals
of a specific system).
Data Model
: this section examined three core data models, i.e. key-value,
column, and document models. In particular, PNUTS uses a row-oriented data
model.
Data Storage
: in some systems data are designed to be stored in RAM and their
snapshots or copies are stored in discs. Other systems store data in discs, with
the cache stored in RAM. A few systems have pluggable background programs
that are allowed to use different data storage media, or standardized underlying
document systems are required.
Concurrency Control
: there are three concurrency control mechanisms used
in the existing systems: lock, MVCC, and non-concurrency control. The lock
mechanism only allows a user to read or modify a real object (i.e., object,
document, or row) at any time. The MVCC mechanism ensures the reading
consistency. However, if several users modify a real object at the same time,
several conflicting editions of a real object may be created. Some systems do not
offer atomicity but allow different users to concurrently modify different parts
of the same object, and may not ensure which edition will be acquired during
reading.
Consistency
: according to the CAP theorem, strict consistency could not be
simultaneously achieved along with availability and partition tolerance. The
Search WWH ::
Custom Search