bart.sched.functions module

Scheduler specific Functionality for the stats framework

The Scheduler stats aggregation is based on a signal which is generated by the combination of two triggers from the events with the following parameters

EVENT	VALUE	FILTERS
sched_switch()	1	next_pid
sched_switch()	-1	prev_pid

Both these Triggers are provided by the event trappy.sched.SchedSwitch which correspond to the sched_switch unique word in the trace

See also

trappy.stats.Trigger.Trigger

Using the above information the following signals are generated.

EVENT SERIES

This is a combination of the two triggers as specified above and has alternating +/- 1 values and is merely a representation of the position in time when the process started or stopped running on a CPU

RESIDENCY SERIES

This series is a cumulative sum of the event series and is a representation of the continuous residency of the process on a CPU

The pivot for the aggregators is the CPU on which the event occurred on. If N is the number of CPUs in the system, N signal for each CPU are generated. These signals can then be aggregated by specifying a Topology

See also

trappy.stats.Topology.Topology

bart.sched.functions.CPU_FIELD = '__cpu'

The column in the sched_switch event that indicates the CPU on which the event occurred

bart.sched.functions.NEXT_PID_FIELD = 'next_pid'

The column in the sched_switch event that indicates the PID of the next process to be scheduled

bart.sched.functions.NO_EVENT = 0

Signifies no event on an event trace

bart.sched.functions.PREV_PID_FIELD = 'prev_pid'

The column in the sched_switch event that indicates the PID of the process that was scheduled in

bart.sched.functions.SCHED_SWITCH_IN = 1

Value of the event when a task is switch in or scheduled on a CPU

bart.sched.functions.SCHED_SWITCH_OUT = -1

Value of the event when a task is switched out or relinquishes a CPU

bart.sched.functions.SERIES_SANTIZED = '_sched_sanitized'

A memoized flag which is set when an event series is checked for boundary conditions

bart.sched.functions.TASK_NOT_RUNNING = 0

In a residency series, a zero indicates that the task is not running

bart.sched.functions.TASK_RUNNING = 1

The column in the sched_switch event that indicates the CPU on which the event occurred

bart.sched.functions.TIME_INVAL = -1

Standard Value to indicate invalid time data

bart.sched.functions.WINDOW_SIZE = 0.0001

A control config for filter events. Some analyses may require ignoring of small interruptions

bart.sched.functions.binary_correlate(series_x, series_y)

Helper function to Correlate binary Data

Both the series should have same indices

For binary time series data:

\[\alpha_{corr} = \frac{N_{agree} - N_{disagree}}{N}\]

Parameters:

• series_x (pandas.Series) – First time Series data
• series_y (pandas.Series) – Second time Series data

bart.sched.functions.csum(series, window=None, filter_gaps=False)

aggfunc() for the cumulative sum of the input series data

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window
• filter_gaps (bool) – If set, a process being switched out for bart.sched.functions.WINDOW_SIZE is ignored. This is helpful when small interruptions need to be ignored to compare overall correlation

bart.sched.functions.filter_small_gaps(series)

A helper function that does filtering of gaps in residency series < bart.sched.functions.WINDOW_SIZE

Parameters:series (pandas.Series) – Input Time Series data

bart.sched.functions.first_cpu(series, window=None)

aggfunc() to calculate the time of the first switch in event in the series This is returned as a vector of unit length so that it can be aggregated and reduced across nodes to find the first cpu of a task

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

bart.sched.functions.first_time(series, value, window=None)

aggfunc() to:

• Return the first index where the series == value
• If no such index is found +inf is returned

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

Returns:

A vector of Unit Length

bart.sched.functions.get_pids_for_process(ftrace, execname, cls=None)

Get the PIDs for a given process

Parameters:

• ftrace (trappy.ftrace.FTrace) – A ftrace object with a sched_switch event
• execname (str) – The name of the process
• cls (trappy.base.Base) – The SchedSwitch event class (required if a different event is to be used)

Returns:

The set of PIDs for the execname

bart.sched.functions.get_task_name(ftrace, pid, cls=None)

Returns the execname for pid

Parameters:

• ftrace (trappy.ftrace.FTrace) – A ftrace object with a sched_switch event
• pid (int) – The PID of the process
• cls (trappy.base.Base) – The SchedSwitch event class (required if a different event is to be used)

Returns:

The execname for the PID

bart.sched.functions.last_cpu(series, window=None)

aggfunc() to calculate the time of the last switch out event in the series This is returned as a vector of unit length so that it can be aggregated and reduced across nodes to find the last cpu of a task

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

bart.sched.functions.last_time(series, value, window=None)

aggfunc() to:

• The first index where the series == value
• If no such index is found bart.sched.functions.TIME_INVAL is returned

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

Returns:

A vector of Unit Length

bart.sched.functions.period(series, align='start', window=None)

This aggfunc() returns a tuple of the average duration between two triggers:

• When align=start the SCHED_IN trigger is used
• When align=end the SCHED_OUT trigger is used

\[\begin{split}E = e_{1}, e_{2}...e_{N} \\ T_p = \frac{\sum_{j}^{\lfloor N/2 \rfloor}(e_{2j + 1} - e_{2j})}{N}\end{split}\]

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

Returns:

A list of deltas of successive starts/stops of a task

bart.sched.functions.residency_sum(series, window=None)

aggfunc() to calculate the total residency

The input series is processed for intervals between a bart.sched.functions.SCHED_SWITCH_OUT and bart.sched.functions.SCHED_SWITCH_IN to track additive residency of a task

\[\begin{split}S_{in} = i_{1}, i_{2}...i_{N} \\ S_{out} = o_{1}, o_{2}...o_{N} \\ R_{total} = \sum_{k}^{N}\Delta_k = \sum_{k}^{N}(o_{k} - i_{k})\end{split}\]

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

Returns:

A scalar float value

bart.sched.functions.sanitize_asymmetry(series, window=None)

Sanitize the cases when a SWITCH_OUT happens before a SWITCH_IN. (The case when a process is already running before the trace started)

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

bart.sched.functions.sched_switch_in_trigger(ftrace, pid, sched_switch_class)

Parameters:

• ftrace (trappy.ftrace.FTrace) – A ftrace object with a sched_switch event
• pid (int) – The PID of the associated process
• sched_switch_class (trappy.base.Base) – The SchedSwitch event class

Returns:

trappy.stats.Trigger.Trigger on the SchedSwitch: IN for the given PID

bart.sched.functions.sched_switch_out_trigger(ftrace, pid, sched_switch_class)

Parameters:

• ftrace (trappy.ftrace.FTrace) – A ftrace object with a sched_switch event
• pid (int) – The PID of the associated process
• sched_switch_class (trappy.base.Base) – The SchedSwitch event class

Returns:

trappy.stats.Trigger.Trigger on the SchedSwitch: OUT for the given PID

bart.sched.functions.sched_triggers(ftrace, pid, sched_switch_class)

Returns the list of sched_switch triggers

Parameters:

• ftrace (trappy.ftrace.FTrace) – A ftrace object with a sched_switch event
• pid (int) – The PID of the associated process
• sched_switch_class (trappy.base.Base) – The SchedSwitch event class

Returns:

List of triggers, such that

triggers[0] = switch_in_trigger
triggers[1] = switch_out_trigger

bart.sched.functions.select_window(series, window)

Helper Function to select a portion of pandas time series

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

bart.sched.functions.trace_event(series, window=None)

aggfunc() to be used for plotting the process residency data using trappy.plotter.EventPlot

Parameters:

• series (pandas.Series) – Input Time Series data
• window (tuple) – A tuple indicating a time window

Returns:

A list of events of the type:

[
    [start_time_1, stop_time_1],
    [start_time_2, stop_time_2],
    #
    #
    [start_time_N, stop_time_N],
]