Database Reference
In-Depth Information
Table 1.
TPC-H LINEITEM table observed statistics, possible customization, and
anticipated evolution
Column
Distinct Workl oad
Data
Storage Structure Storage Structure Storage Structure
Name
Count
Access Initial
1st Evolution
2nd Evolution
L ORDERKEY
1500000
Sorted Array
Sorted List
B+-Tree
L COMMENT
4501941
Sorted Array
Sorted List
Hash Table
L DISCOUNT
11 Read-Intensive
Sorted Array
L SHIPMODE
7
Heap Array
L SHIPINSTRUCT
4
Heap Array
L RECEIPTDATE
2554
Heap Array
Heap List
L COMMITDATE
2466
Ordered Sorted Array
Sorted List
L SHIPDATE
2526
Ordered Sorted Array
Sorted List
L LINESTATUS
2
Heap Array
L RETURNFLAG
3
Heap Array
L TAX
9 Read-Intensive
Sorted Array
L EXTENDEDPRICE
933900 Read-Intensive
Sorted Array
Sorted List
B+-Tree
L QUANTITY
50 Read-Intensive Ordered Sorted Array
L LINENUMBER
7
Heap Array
L SUPPKEY
10000
Heap Array
Heap List
L PARTKEY
200000
Sorted Array
Sorted List
Hash Table
3 Evolutionary Column-Oriented Storage
In this section, we explain the concepts of ECOS in detail. We introduce and
explain four DSM based schemes proposed to reduce the high storage require-
ment of standard 2-copy DSM. We also discuss the concepts of the table and
the column customization, hierarchical organization and evolution of the storage
structures, and the evolution path.
3.1 Table-Level Customization
ECOS is a customizable and online self-tuning storage manager. We use the
term storage manager in its standard meaning for DBMS, i.e., a component
to physically store and retrieve data. Data storage eciency is assumed to be
the main goal for a storage manager. By storage structure, we mean the data
structure used by the storage manager to physically store data and indexes.
ECOS stores data according to the column-oriented storage model, where each
column stores a key/value pair of data. ECOS suggests two customizations for
each table in a database, i.e., at the table-level and at the column-level. At the
table-level, we customize, how columns are stored physically for a logical schema
design. We use five variations of DSM for table customization, i.e., Standard 2-
copy DSM [10], Key-copy DSM (KDSM), Minimal DSM (MDSM), Dictionary
based Minimal DSM (DMDSM), and Vectorized Dictionary based Minimal DSM
(VDMDSM). The motivation for proposing and testing different variations of
