nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

As we transition to the new Hyland Community, some post links may be temporarily disrupted as we finalize redirects. We apologize for the inconvenience and are working to resolve this as quickly as possible. Details.

How to optimize process bottleneck?

gromar

Hello.

I tried to optimize process from my previous post (How to predict process bottleneck?).
The way, how is it done, is described here Optimize the process 1.

Questions

Does anybody have real case where simulation could be used to optimize process? (I would like to use real case to extend simulation framework)In fact, is simulation needed to optimize processes?As usual I would appreciate any feedback and interesting suggestions to extend simulation framework.

Find more posts tagged with

Comments

jbarrez

Nice!

One way that typically 'solves' these bottlenecks is escalation: put a timer on the slow task, and escalate it to someone else (or a group) when the timer fires.

Does anybody have real case where simulation could be used to optimize process? (I would like to use real case to extend simulation framework)
In fact, is simulation needed to optimize processes?

I think you must separate two different phases of optimization:

- Optimization during design: here simulation is usable. Use the simulation to test-proof the process and see potential bottlenecks. Typically, you'd see group allocation in simulation tool. Ie user task A can be done by group X with 5 people and task B can be done by group Y with only 2 people. Then, you can take a bunch of processes and run them through the simulator. You'd probably see bottlenecks arise once group Y has more and more work to do.
- Optimization during runtime: here simulation is not usable, as this must be done based on the historical data produced by the engine

gromar

One way that typically 'solves' these bottlenecks is escalation: put a timer on the slow task, and escalate it to someone else (or a group) when the timer fires.

Yes. Escalation 'solves' bottleneck. I would say, it is 'Fireman extinguishes fires' approach, and it works in many cases.

I think you must separate two different phases of optimization:

- Optimization during design

Yes exactly, simulation during design can decrease pressure on the 'fireman'.

- Optimization during runtime: here simulation is not usable, as this must be done based on the historical data produced by the engine

I hope simulation is usable in runtime too. From my point of view, simulation is something like a scrying ball. (another question is how much scrying ball costs and how precise is.

). That's why I used historical data to simulate users activities.
I made small examples ( DueDate Monitor, Are we able to finish all started process instances? )

The first example was done some time ago - it is not based on the simulation process engine. It uses simulation on the process tasks. Migration is simple.

I hope somebody will have real case where simulation (in design/runtime/optimization phase) can be successfully applied.

jbarrez

Yeah, I think you are right: the simulation engine could be used for runtime optimization too. Or at least there is a big shared piece.

I don't have any 'real' business processes unfortunately. As framework builders, we're the last one to see them …

gromar

The comment on the margin
I do not have any real example but I tried to generate report about percentage of tasks finished after dueDate. Does not matter whether simulation was used or not.
To read more visit DueDateGenerator page.

jbarrez

Cool. What are your plans with this? Do yo envision a toolset around Activiti?

gromar

Yes, I would like to provide tool which can support process life cycle.

The tool (let say scrying ball) should provide:

1. reports to monitor activiti process engine (running, finished processes, tasks…..) - it should be independent from "live" engine or "simulation" engine. Currently implemented reports are:

a. Amount of task instances - process bottleneck b. Tasks after due date -dueDate generator 2. simulation engine - to simulate process behavior and to help in process monitoring and optimization.To achieve that I would like to have real example, on which I can prove the concept.
Any ideas are welcome.

Currently I am working on process model extension ( to cover simulation model too).

gromar

I added new functionality to simulator.

The idea is to use the real load and user behavior to optimize processes. I used following steps:

1. record load and users behavior (History service full audit from live engine)2. prepare simulation engine for playback3. playback engine execution4. evaluate simulation run ( if acceptance requirements were not met change process and playback again)To read more visit a link or try following test

Help:

Could somebody help me to find good example where process simulation and optimization can be applied?Any feedback is appreciated.Thank you in advance.

jbarrez

The link you posted isnt working anymore, where can I find it?

gromar

Sorry - I've renamed the project

gromar

https://github.com/gro-mar/activiti-crystalball/wiki/Optimize-the-process-2
https://github.com/gro-mar/activiti-crystalball/blob/master/simulator/src/test/java/org/activiti/crystalball/simulator/PlaybackTest.java

jbarrez

Hmmm I'm not sure about that extra namespaced attributes on those elements…. feels pretty invasive (but I can see the need for it).

Can't you use a BpmnParseHandler (http://activiti.org/userguide/index.html#advanced_parseHandlers) that you configure on the process engine config to add those custom elements dynamically? Would make it prettier, I think.

But overall, again nice progress!

gromar

BpmnParseHandler

are part of activiti 5.12 - I did not upgrade dependency from 5.11 yet. 😞

add those custom elements dynamically

Could you explain little bit more?
Should I store simulation model data outside of process definition? (it can make sense too - because of different simulation models, and simulation experiments….)

jbarrez

are part of activiti 5.12 - I did not upgrade dependency from 5.11 yet.

In Activiti you can use a BPMNParseListener. The mechanism is similar and you can upgrade easily once you switch to 5.12

Should I store simulation model data outside of process definition? (it can make sense too - because of different simulation models, and simulation experiments….)

You could do that, and that's certainly a valid thing to think about.
But that was not where I was hinting add. I meant that adding the custom elements with your own namespace might be too cumbersome to do, and you might add them dynamically to your process definition when it is parsed. That way, you can enrich your processes dynamically when they are loaded into memory before execution.

But I don't know if there are general rules to apply those elements. Eg. is it always the same class for a given type? Or does that change? If it changes depending on the author of the process definition, your approach probably is more suitable.

gromar

But I don't know if there are general rules to apply those elements. Eg. is it always the same class for a given type?

I do not think so. It depends on the simulation experiment. How far in the simulation do we want to go. In that case it is not always the same class for given element type. Simulation experiment has to be driven by modified process model. Simulation process model and "real" process model are tighten together now. But there can be several simulation experiments defined on the one process model. May be good way is to specify multiple simulation experiments in the one process model. I do not want to over complicate process definition for now.

BPMNParseListener

Thanks. I used it in jobs and timers support in simulation engine (details). So currently we are able to support asynchronous events in the simulation.

I started to work on simulation application examples as you propose in the following post. Thank you once again for the good example. When I finish example series I will consider to make the first release of activiti-crystalball
[img]https://raw.github.com/gro-mar/activiti-crystalball/master/examples/src/main/process/org/activiti/crystalball/examples/mortages/MortageDemo-0.png[/img]

gromar

Continue in reading here
http://forums.activiti.org/en/viewtopic.php?f=6&t=6266