bart.sched.SchedMatrix module

The SchedMatrix provides an ability to compare two executions of benchmarks with multiple processes.

For example, consider a benchmark that spawns 4 identical threads and any two threads should exhibit a certain behaviours and the remaining another identical but different behaviour.

SchedMatrix creates a Matrix of Scheduler Waveform Correlations

A = Reference Execution B = Execution to be Evaluated

           +---+  +---+
           |   |  |   |
A1, B3 +---+   +--+   +--------------+
                      +---+  +---+
                      |   |  |   |
A2, B4 +--------------+   +--+   +---+
           +---+  +---+
           |   |  |   |
A3, B1 +---+   +--+   +--------------+
                      +---+  +---+
                      |   |  |   |
A4, B2 +--------------+   +--+   +---+

Correlation Matrix

A1	1	0	1	0
A2	0	1	0	1
A3	1	0	1	0
A4	0	1	0	1

Thus a success criteria can be defined as A1 having two similar threads in the evaluated execution

assertSiblings(A1, 2, operator.eq)
assertSiblings(A2, 2, operator.eq)
assertSiblings(A3, 2, operator.eq)
assertSiblings(A4, 2, operator.eq)

class bart.sched.SchedMatrix.SchedMatrix(reference_trace, trace, topology, execnames, aggfunc=<function csum>)

Bases: object

Parameters:

• reference_trace (str, trappy.ftrace.FTrace) – The trace file path/ftrace object to be used as a reference
• trace (str, trappy.ftrace.FTrace) – The trace file path/ftrace object to be verified
• topology (trappy.stats.Topology.Topology) – A topology that describes the arrangement of CPU’s on a system. This is useful for multi-cluster systems where data needs to be aggregated at different topological levels
• execnames –

  The execnames of the task to be analysed

  A single execname or a list of execnames can be passed. There can be multiple processes associated with a single execname parameter. The execnames are searched using a prefix match.

Consider the following processes which need to be analysed:

• Reference Trace

  PID	execname
  11	task_1
  22	task_2
  33	task_3

• Trace to be verified

  PID	execname
  77	task_1
  88	task_2
  99	task_3

A bart.sched.SchedMatrix.SchedMatrix instance be created following different ways:

• Using execname prefix match

  SchedMatrix(r_trace, trace, topology,
              execnames="task_")
  

• Individual Task names

  SchedMatrix(r_trace, trace, topology,
              execnames=["task_1", "task_2", "task_3"])
  

assertSiblings(pid, expected_value, operator, tolerance=0.8)

Assert that the number of siblings in the reference trace match the expected value and the operator

Parameters:

• pid (int) – The PID of the process in the reference trace
• operator –

  A binary operator function that returns a boolean. For example:

  import operator
  op = operator.eq
  getSiblings(pid, expected_value, op)
  

  Will do the following check:

  getSiblings(pid) == expected_value
  

• tolerance (float) – A correlation value > tolerance will classify the resultant process as a sibling

See also

bart.sched.SchedMatrix.SchedMatrix.getSiblings

getSiblings(pid, tolerance=0.8)

Return the number of processes in the reference trace that have a correlation greater than tolerance

Parameters:

• pid (int) – The PID of the process in the reference trace
• tolerance (float) – A correlation value > tolerance will classify the resultant process as a sibling

See also

bart.sched.SchedMatrix.SchedMatrix.assertSiblings

print_matrix()

Print the correlation matrix