Database Reference
In-Depth Information
(i.e., data size, workload, and data access) to keep our discussion simple and
understandable. Table 1 shows only the evolution for dictionary columns for the
LINEITEM table as they utilizes the benefits of evolving hierarchically-organized
storage structures to their full potential. Before we conclude this section, to avoid
any confusion, we want to disclaim that the terms and concepts of evolution,
evolution path, mutation rules, and heredity information used in this paper have
no relevance with their counterpart in evolutionary algorithms or any other non-
relevant domain.
4
Implementation and Empirical Evaluation
In this section, we provide the details of our micro benchmark and the evalu-
ation results for ECOS
2
. The data and index storage structures that we have
implemented in the existing ECOS prototype implementation are the same as we
have discussed in Section 3.2. To simplify our discussion, we present the results
involving sorted array, sorted list, and HLC SL.
4.1 Micro Benchmark Details
For ECOS evaluation, we set up a micro benchmark with repeated insertion,
selection, and deletion of data using API based access methods. The data contain
keys in ascending, descending, and random order, which also represents their
insertion, selection, and deletion order in the database. For different columns, the
number of records (cardinality) is kept different. We defined seven columns with
two unique non-null columns, one of them used as a primary key. We used three
different widths for columns, i.e., 16, 85, and 4096 bytes to assess the impact of
tuple width on performance of different storage schemes. All storage structures
used in a micro benchmark operate in main-memory. For ECOS evaluation,
we used CPU cycles and heap memory as resources. We used OpenSuse 11.2
operating on Intel(R) Core(TM)2 Duo CPU E6750 @ 2.66GHz with four GB
of RAM. We measured execution speed by taking the average of CPU cycles
observed over multiple iterations of the micro benchmark. We used Valgrind
tools suite [21] to measure the heap usage.
4.2 ECOS Performance Improvement
To demonstrate the performance gain using ECOS, we presented our observa-
tion of the effect of an increase in data size on performance of different storage
structures in [18,19]. According to our observation in [18,19], we suggest the
performance gain and reduced resource consumption using the evolving storage
structures because evolving storage structures attempt to use minimal/simple
storage structures (such as sorted array for small data management) as long as
possible using the definitions from evolution paths. To demonstrate the evolv-
ing storage structures evolution, we present the evaluation results for evolving
2
Please refer to the web link for all related publications and prototype evaluation bi-
