Implement Benchmarking Strategy: Heuristic 2 (Binary Search Strategy)

Closes #49 (closed) Please review @she Tested with unit-test with a complex input and with manual test for the following two workloads: (1 = successful experiment, in the sense that the application could handle the load, 0 = unsuccessful experiment):

Manual Workload 1

dim_value/instances	1	2
100	1	1
200	1	1

The result of the execution was as follows (executed experiments marked with 'x', not executed with 'o'):

dim_value/instances	1	2
100	x	o
200	x	o

Manual Workload 2

dim_value/instances	1	2
100	1	1
200.000	0	0

The result of the execution was as follows (executed experiments marked with 'x', not executed with 'o'):

dim_value/instances	1	2
100	x	o
200.000	x	x

execution/README.md

@@ -163,5 +163,5 @@ There are the following benchmarking strategies:
 
 * `default`: Execute a subexperiment for each combination of the number of replicas (N), and the number of workload intensities (M). The amount of executed subexperiments is always N*M.
 * `step`: A heuristic which works as follows: After each subexperiment, it is checked, whether the application could handle the workload. If the workload could be handled, the workload intensity is increased in the next subexperiment. Otherwise, the workload intensity is kept the same and the number of instances is increased. The amount of executed sub-experiments is at most N+M-1.
-* `binary-search`: A heuristic which works as follows: For each number of instances, a binary-search like strategy is used to determine the workload intensity, to which the system handle the workload. The limiting workload intensity is taken into consideration when executing sub-experiments for other number of replicas. The number of executed sub-experiments is at most M*log(N).
+* `binary-search`: A heuristic which works as follows: For each number of instances, a binary-search like strategy is used to determine the workload intensity, to which the system handle the workload. The limiting workload intensity is taken into consideration when executing sub-experiments for other number of replicas. The number of executed sub-experiments is at most N*log(M).