Database Reference
In-Depth Information
ECOS: Evolutionary Column-Oriented Storage
Syed Saif ur Rahman, Eike Schallehn, and Gunter Saake
Faculty of Computer Science,
Otto-von-Guericke University, Magdeburg, Germany
{
srahman,eike,saake
}
@ovgu.de
Abstract.
As DBMS has grown more powerful over the last decades,
they have also become more complex to manage. To achieve eciency
by DBMS tuning is nowadays a hard task carried out by experts. This
development inspired the ongoing research on self-tuning to make DBMS
more easily manageable. We present a customizable self-tuning storage
manager, we termed as Evolutionary Column-Oriented Storage (ECOS).
The capability of self-tuning data management with minimal human in-
tervention, which is the main design goal for ECOS, is achieved by dy-
namically adjusting the storage structures of a column-oriented storage
manager according to data size and access characteristics. ECOS is based
on the Decomposed Storage Model (DSM). It supports customization at
the table-level using five different variations of DSM. ECOS also proposes
fine-grained customization of storage structures at the column-level. It
uses hierarchically-organized storage structures for each column, which
enables autonomic selection of the suitable storage structure along the
hierarchy using an evolution mechanism (as hierarchy-level increases).
Moreover, for ECOS, we proposed the concept of an evolution path that
provides a reduction of human intervention for database maintenance.
We evaluated ECOS empirically using a custom micro benchmark show-
ing performance improvement.
Keywords:
column-oriented storage, evolving hierarchically-organized
storage structures, customization, autonomy.
1
Introduction
Ecient data management demands continuous tuning of a database and a
DBMS. The need for tuning a DBMS is driven by changes, such as database
size, workloads, schema design, hardware, and application specific data man-
agement needs. Existing DBMS need extensive human intervention for tuning,
which contributes to a major portion of the total cost of ownership for data
management [7]. Self-tuning is the solution to reduce the tuning cost through
minimizing the human intervention [22]. However, researchers are united on one
conclusion that the self-tuning based solutions are the biggest challenge in the
database domain because of the inherent complexity of existing DBMS architec-
tures. Their functionalities are tightly integrated into their monolithic engines,
and it is dicult to assess the impact of tuning of one knob on another [6].
In this paper, we present a customizable and online self-tuning storage man-
ager. As a key design concept, we propose the selection of an appropriate storage
