Database H gh Ava lab l ty: An Extended Survey
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission
of Idea Group Inc. is prohibited.
Finally, the last component of the availability-benchmarking methodology specifies the way
that results are collected and presented. Essentially, this component defines the procedural
aspects of carrying out an availability benchmark. First, the system is run under the generated
workload with no faults injected. The quality of service values collected during this
run are statistically processed to produce a 99% confidence interval demarcating the normal
quality of service behavior of the system. Then, the experiments are repeated multiple times
with different combinations of faults injected during those runs; the methodology specifies
both single-fault micro benchmarks in which a single fault is injected and the system is left
untouched until it stabilizes or crashes, and multiple-fault macro benchmarks in which a
series of faults designed to mimic a complex real-world scenario is injected, with human
intervention allowed for system maintenance purposes. The results of these faulty runs
are reported graphically, with quality of service plotted versus time, overlaid with both
an indication of when the faults were injected as well as with the 99% confidence interval
computed from the normal run.
Pages:
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107