ChunkSim - Data Warehousing Design and Advanced Engineering Applications - page 144

Table 1. Extract of analysis report for replication degree

Time
PL-H (LP)

Replication Degree Time FM (LP) Time (LP) Time Query

10% 240 153.552276 239.607907 247.232275

25% 240 153.552276 238.044278 247.362589

50% 240 153.552276 232.592335 247.362589

75% 240 153.552276 219.842612 247.232275

100% 240 153.552276 207.945652 244.492903

150% 240 153.552276 198.712255 233.492903

200% 240 153.552276 185.513946 223.753531

300% 240 153.552276 179.151584 212.753531

400% 240 153.552276 175.329786 207.623217

500% 240 153.552276 173.505943 207.623217

that every chunk has a copy in the system. These
results allow the user to see what performance
benefit there may be from adding redundancy and
also where the choice lies in the interval between
the worst and best PL-H and FM times.
System Size Planning (SSZP) - this experi-
ment answers the question of how the performance
varies with different system sizes under specific
placement and replication scenario, which can
be used to determine the number of nodes for a
dataset.

Additional inputs: replication Degrees array
(optionally, also an array listing the set of
nodes that is offline, in case AA is used to
characterize the behavior in a particular
scenario in terms of which nodes are of-
fline or fail);
Outputs: a set of tuples (Number Of Nodes
Offline, Replication Degree, FailureRate,
Time (LP), Time Query). The “Time (LP)”
and “Time Query” fields are the average
expected runtime of the Local Processing
part and of the whole Query, respective-
ly. The “Number Of Nodes Offline” and
“Replication Degree” are self-explanatory.
The “FailureRate” parameter is a number
between 0 and 1 and represents the fraction
of runs that could not end because a chunk
was not present in any of the active nodes
(the set of nodes remaining after taking
nodes offline). Given that nodes are het-
erogeneous, may have heterogeneous data
loads, and different nodes may fail, it is
possible for a run to succeed (find all nec-
essary chunks) while another one may fail
(not being able to run for lack of at least
one necessary chunk).

Additional inputs: Number of Nodes array;
Outputs: a set of tuples (Number of Nodes, Time
(LP), Time Query). The “Time (LP)” and
“Time Query” fields are the average ex-
pected runtime of the Local Processing part
and of the whole Query, respectively. The
Number of Nodes array specifies which al-
ternatives should be tested.

Availability Analysis (AA) - this experiment
answers the question of whether and how fast the
system is able to provide answers for varying
replication degrees when different sets of nodes
are offline or fail.