performance bottleneck analysis

GiriprasadSridhara

Hello,

I read this on finding bottlenecks (https://www.win.tue.nl/promforum/discussion/365/finding-the-bottleneck).

I have some further doubts. Is it always necessary to mine a control flow (say in the form of a petri net) to do performance bottleneck analysis? Cannot it be done using the log data itself without extracting a work/control flow?

for example, suppose my event log has <ID,EventName,Timestamp> as follows:
<1, A, t1> <1, B, t2> <1, C, t3> <1, D, t4>
<2, A, t5> <2, C, t6> <2, B, t7> <2, D, t8>

Can i not simply subtract the times for consecutive steps based on the ID such as t2-t1 for events, B,A for trace 1? The maximum differences should give the bottleneck(s) right? Or is it too simplistic?

Where can I find more about why exactly the two steps of petrinet/workflow extraction and replay required?

Thanks.

hverbeek

Hi,

You are assuming that the entire process is sequential. Your example suggests that B and C run concurrently, as both the order BC (trace 1) and CB (trace 2) are there. Then, for C you would need to compute t3-t1 instead of t3-t2, as C could start as soon as A had finished.

From the first trace alone, you cannot know whether or not B and C run concurrently. Fort his, you need information from the other traces as well.

To make it a bit more complex, it can also be that your process contains the activity C in multiple places. There is not a single activity C, but there may be more. So, when you encounter a C in the trace, to which activity does it correspond? Process mining can help here by finding the most plausible activity, but you need a process model for that.

I hope this helps a bit.

Kind regards,

Eric.

GiriprasadSridhara

thank you very much for your detailed response.