Database Reference
In-Depth Information
B+-Trees are optimized for both cache and disk performance, which is also a
goal for the ECOS. However, the ECOS concepts do not restrict the use of any
fixed structure; instead it suggests the use of different storage structures in the
hierarchy to support an ecient use of underlying hardware.
An automated tuning system (ATS) [11] is a feedback control mechanism that
automatically adjusts the tuning knobs using the defined tuning policies accord-
ing to the monitoring statistics. ECOS also works in similar fashion as suggested
in ATS. ECOS also monitors and adjust storage structures with changing data
management needs. Malik et al. in [13] suggested the benefit of online phys-
ical design techniques and proposed an online vertical partitioning technique
for physical design tuning. Similarly, ECOS also operates in online fashion. Au-
tomated physical design research focuses on finding the best physical design
structure for a running workload, e.g., indexes, materialized views, partitioning,
clustering, and views [3]. Existing automated physical design tools assume the
workload as a set of SQL statements [3]. These tools use the query optimizer
to identify the appropriate physical design selection from various proposed can-
didate designs [15]. ECOS also performs automated physical design, but at the
different level, i.e., at the storage manager level. It does not rely on a query
optimizer. Furthermore, ECOS is designed with the motivation of exploring new
architectures for developing self-tuning DBMS instead of developing techniques
to self-tune existing ones.
6 Conclusion and Future Work
In this paper, we presented ECOS, a customizable and online self-tuning storage
manager. ECOS and evolution paths enable and use the fine-grained customiza-
tion of storage structures at the table-level and column-level. In addition, ECOS
and evolution paths allow storage structures to autonomically evolve (to more
suitable storage structures) with the change in the data management needs, to
maintain the desirable performance while keeping the human intervention at a
minimum. We also presented a detailed evaluation and discussion of ECOS and
evaluation paths showing the performance improvement and reduced resource
consumption. As future work, we plan to enhance the presented dictionary based
DSM schemes for better performance. ECOS self-tuning design makes it a suit-
able candidate for emerging cloud computing platforms for data services. We
also intend to investigate the ecient utilization of multi-core and many-core
parallel processors using the presented evolution mechanism. Once query pro-
cessing is implemented, we want to integrate the presented evolution mechanism
with query processing, and then we will be able to evaluate the ECOS using
the full TPC-H benchmark. Transaction management is also an implementation
specific future work for our ECOS prototype.
Acknowledgments.
Syed Saif ur Rahman is a HEC-DAAD Scholar funded by
Higher Education Commission of Pakistan and NESCOM, Pakistan.
