report work

Claudio Maggioni 2 years ago
parent 0c908dbeca
commit d7071adb87

@ -638,24 +638,58 @@ number of termination events per task, broke down by task termination. In
addition, the table shows the mean number of \texttt{EVICT}, \texttt{FAIL},
\texttt{FINISH}, and \texttt{KILL} for each task event termination.
The mean number of events per task is an order of magnitude higher
than in the 2011 traces
Generally speaking, the event type with higher mean is the termination
event for the task
The \# evts mean is higher than the sum of all other event type means,
since it appears there are a lot more non-termination events in the
2019 traces.
\subsubsection{Conditional Probability of Task Success}
The first observation we make is that the mean number of events per
\texttt{EVICT}ed and \texttt{FAIL}ed tasks increased more than 5-fold (namely
from 2.372 to 78.710 and from 3.130 to 24.962 respectively). Also observing the
95-th percentile we can say that the number of events per task has generally
increased overall.
As observed in 2011, 2019 Borg tasks have all a multitude of events with
different types, with \texttt{FINISH}ed tasks experiencing almost always
\texttt{FINISH} events and unsuccessful tasks and the same observation holding
for \texttt{KILL}ed tasks and their \texttt{KILL} events. Differently from the
2011 data, \texttt{EVICT}ed tasks seem to experience an high number of
\texttt{KILL} events as well (25.795 on average per task, over 78.710 overall
events on average). A similar phenomena can be observed with \texttt{KILL}ed
jobs and their \texttt{EVICT} events (1.876 on average per task with a 8.763
event overall average).
Considering cluster-by-cluster behaviour in the 2019 traces (as reported in
figure~\ref{fig:tableIII-csts}) the general observations still hold for each
cluster, albeit with event count averages having different magnitudes. Notably,
cluster E registers the highest per-event average, with \texttt{FAIL}ed tasks
experiencing 111.471 \texttt{FAIL} events out of \texttt{112.384}.
\subsection{Conditional Probability of Task Success}
In this analysis we measure the conditional probability of task success given a
number of specific unsuccessful (i.e. \texttt{EVICT}, \texttt{FAIL} and
\texttt{KILL}) events. This analysis was conducted to better understand how a
given number of unsuccessful events could affect the termination of the task it
belongs to.
Conditional probabilities of each unsuccessful event type are shown in the form
of a plot in figure~\ref{fig:figureV}, comparing the 2011 traces with the
overall data from the 2019 ones, and in figure~\ref{fig:figureV-csts}, as a
cluster-by-cluster breakdown of the same data for the 2019 traces.
In figure~\ref{fig:figureV} the 2011 and 2019 plots differ in their x-axis:
for 2011 data conditional probabilities are computed for a maximum event coun
t of 30, while for 2019 data are computed for up to 50 events of a specific
kind. Nevertheless, another quite striking difference between the two plots can
be seen: while 2011 data has relatively smooth decreasing curves for all event
types, the curves in the 2019 data almost immediately plateau with no
significant change easily observed after 5 events of any kind.
The presence of even one \texttt{KILL} event almost surely causes the
corresponding task to terminate in an unsuccessful way: a task with no
\texttt{KILL} events has 97.16\% probability of success, but tasks with 1 to 5
\texttt{KILL} events have 0.02\%, 0.20\%, 0.44\%, 0.04\%, and
0.07\% probabilities of success respectively. The same effect can be observed,
albeit in a less drastic fashion, for the \texttt{EVICT} and \texttt{FAIL}
curves. The \texttt{EVICT} curve has for 0 to 5
Refer to figure \ref{fig:figureV}.

File diff suppressed because one or more lines are too long