Some of my demo flows which run locally and have a maximum of thirty actions run fine. However, we are currently developing much longer processes on VM's with the specifications mentioned in the table below. Unfortunately we encounter the following performance issues:
- Designer responds really slow and sometimes freezes for a while (for example when adding new actions to the flow).
- When running process flows, some steps take very long time or the process stops without any error.
- It seems to help to completely sign off for the software to perform a bit better for a while.
- If we analyze the task manager, both CPU and memory are peaking while running flows before they get stuck.
4 cores at 2.8 GHZ
What are the recommended specifications for machines to run Power Automate Desktop on? And are there any known performance issues or performance improvements the development team is working on?
Thanks in advance,
The specifications of the VM are sufficient for most usual cases, however it also depends on the size of your desktop flows and if other processes are running on the VM or not.
Could you let us know how many Actions do those flows have approximately?
Also, are there other processes running on the VM while PAD is working?
Thanks for your quick response! Example process is around 350 actions divided over 25 subflows. We're planning on other processes that will be longer. No other significant processes are running as the VM is specifically used for RPA with PAD. The flow is mainly using Excel actions, many desktop UI navigation (in ERP and explorer), some keystrokes and Outlook actions.
In addition to the issues mentioned previously we noticed two things today;
- The designer get's stuck for a few seconds while switching subflows ('program not responding'). Most of the time it will release and proceed after a few seconds.
- As we were testing I'm not sure whether this has to do with the process ending in an error or it also happens when successfully ending a flow but we noticed in the task manager background processes are stacking up. When we found out we've had tens of Excel background processes running.
- We keep having the feeling that when we completely sign off our session which closes all applications the designer is much quicker for a while when developing and running flows after a new logon.
Good to know, besides the challenges that come with new tooling we're also very excited about PAD!
If you need any further information please let me know,
Thank you for your feedback.
The many background processes may be one of the reasons why we have performance issues with PAD.
In case those processes are starting through a Desktop Flow, please make sure that every time an instance of a process is starting through an Action, then there is also an Action that closes the instance.
For example, when you launch Excel in the background through the Launch Excel Action, please make sure that a Close Excel Action will follow. Since it's running in the background it's difficult to identify that.
So could you please check, if there are no redundant background processes running, are you still facing performance issues with PAD?
I'm having a lot of performance issues as well... After 30min to 1 hour of usage, PAD, becomes unusable and too slow. I close it and restart it and seems okay for a little while. Very painful...
We have the same problem. On a server that have 2 cpu cores, 8gb ram we can no longer edit the flow because it is too slow to work with. On a laptop whit 8gb ram and 4 cpu cores (8 threads) it runs ok, but it starts to get slow to edit. The process hav 10 subflows with a total of 145 actions. we also have the problem that it is ok in the begining but after a while it gets slower and slower.
We have no other runing programs except one instans of chrome whit one tab.
We have checked our flows and made some improvements to prevent redundant background processes. Unfortunately this did not solve the performance issues. We are now live with a UI flow that runs through 10 administrations of an ERP and performs the same actions with similar workload in each. The first process after a server restart takes about 15 minutes and the processing time increases to almost 60 minutes per administration near the end. This process should take about 2,5 hours, but now takes nearly 6 hours to complete. Positive node; the bot is stable even running for extensive periods.
Also, while developing, the PAD software keeps freezing when moving through subflows and gets really slow when adding new actions.
We have monitored CPU and memory but do not see really strange performance. We are now thinking of software or connection to the online environment/pipeline (and firewall?) to be the bottleneck. Do you have any guidance on how to test and improve? If you are interested I can provide logging or other information?