Database Reference
In-Depth Information
FastBit index
Comp. Data
Q1
Q2A
Q2B
Q3A0
Q3B0
Q4A0
Q4B0
total
1.157
0.540
0.541
1.004
4.995
9.799
12.784
30.819
66.436
28.478
28.367
76.431
128.688
5.235
0.015
333.65
(a)
12
Comp. data
FastBit index
10
8
6
4
2
0
1
10
100
1000
10000
100000
1e+06
1e+07
1e+08
Column Cardinality
(b)
Figure 6.19
Time needed to answer count queries from the Set Query
Benchmark with 100 million records: (a) total query response time; (b) time
to answer-Q1 queries.
lowest cardinality column. This process continues on all columns including the
column of sequence numbers. After this reordering, the first few columns are
very nearly sorted and can be compressed very well. Their compressed sizes
are much smaller than the original data, and Q1 queries (
Select count(*) From
Bench Where :col = 2
, where
:col
is the name of a column in the test dataset)
can be answered very quickly. In these cases, the time required by two test
systems are about the same. However, when higher cardinality columns are
involved in the queries, the indexless system requires much more time. Overall,
the total time required to answer all 13 instances of Q1 is about 1.2 seconds
with the FastBit indexes, but about 66 seconds with the compressed base data.
The indexless system works well on queries Q4A0 and Q4B0 because these
queries only involve low-cardinality columns where the compressed columns
are very small.
