working on the latest data challenge I am processing the data using the Parallel Chunk nodes. Nothing special as I am using them almost on a daily basis.
Though, I noticed an extreme UI latency despite having enough system resources available. Please see my screen recording (download).
I can work with the situation and will share the workflow once I have completed the challenge.
Updaate
I believe I figured out what is happening or at least have a good suspicion:
Some CPUs are highly utilized but not all
The UI lag is triggered upon parallel execution indicating that UI rendering is happening using CPU instead of GPU
The rendering process is done on one of the heavy utilized cores causing congestion
The problem I see here ⌠I utilized all cores before in even more demanding processes like web crawling with +10 sessions all while crunching data in another workflow but without any perceived UI lag.
This issue also appeared only with the latest update of Knime or any of itâs extensions. I am doing nothing fancy or totally new either.
Great to see your detailed documentation of the issue. I opened up a ticket previously in relation to performance, and I have went ahead and added a link to this thread under it also and added a +1 to it as they both relate to hardware not being fully utilized properly.
Iâve reproduced this kind of today. killing âequo chromium helperâ process solved it. Windows11 with KNIME 5.3.3.
scenario: while working an a workflow, with several other workflows also open in other windows, the CPU usage went over the roof suddenly. It made the UI lag too.
I had something possibly similar with KNIME 5.4 today. The MUI has been very sluggish with a couple of workflows open, and switching between workflows or switching in and out of metanodes has been a continual series of waiting a good few seconds for it KNIME to switch.
I have also seen a high incidence of the KNIME 5.4 âunresponsive UI detectedâ message that offers me the opportunity to âsave and restartâ or wait. I havenât had to restart, as waiting is sufficient, but the message keeps bugging me, and to be honest was getting quite annoying when it kept popping up as I waited for a workflow to run, and as far as I could see it wasnât actually unresponsive. It was simply executing a loop. Iâd clear the message and then maybe 30 seconds later the message popped up again.
Having decided to finally shut down KNIME to give the CPU some time off (the fan has been constantly running), I noticed that it made no difference to the fan noise, and opening Task manager I could see that a number of Equo Chromium helpers were still running. I thought Iâd give it a chance to finish doing stuff, if thatâs all it was but some 20 minutes later (as I write this), they were still there even after KNIME had long since closed.
Not only that, but they continue to consume a high percentage of my CPU (and the laptop fan, from the constant noise itâs making!)
I didnât spot it at first but in fact I had also been running KNIME 5.3 at some point earlier today, and looking at the above screenshot you can see there are Helper processes for 5.3 still running too, with one of those also consuming high CPU. Iâm guessing that this is why KNIMEâs UI has been particularly sluggish all afternoon?
Another 5 minutes further on, and still they executeâŚ
@takbb can you check whenever the utilization is on one core only? I noticed that this happened to me pretty much all the time. With the recent conversation amongst the start alliance about performance and our possible findings about MUI, I believe the culprit of any perceived performance regression is due to the poor multi-threading capabilities (under certain scenarios).
Check out what @mwiegand suggests, I think including that on the ticket will be helpful if you notice it only using one core. I added another +1 to the ticket.
Do I understand correctly that the slowdowns, particularly those with orphaned equo chromium helpers, have nothing to do with actually generating large images (as was the original context)?
Since version 5.8, I have had the same problem with âequochro_helper.exeâ. As soon as I open a workflow, the CPU usage exceeds 80%. However, I work on a remote computer, and the problem only occurs here.
It is essentially the UI process that is responsible to render the workflow. We have recently changed the rendering from an SVG based approach (that works mainly on CPU) to a Canvas based approach (this can make use of the GPU of a computer). It turns out that the Canvas variant is really strong for large workflows if there is a GPU available, but it performs very poorly in scenarios where there is no GPU. Most of these remote scenarios either have no GPU available, which is why the performance feels not as good. For these cases I would recommend to go into the settings and change from the WebGL based rendering to the old SVG based rendering. We are currently trying to identify the use cases for such remote setups and trying to explore solutions for it.
What is your use case why you are using this remote setup?
Hi Daniel and thanks for clarifying thisâŚ
I do not have a GPU available, so I will change the setting.
Donât know if would be technically possible to auto detect this (availability of GPU) and change the setting accordingly?
let me clarify this. I am not taking about an external GPU that is now required. Most laptops come with an integrated GPU that should already be sufficient. Do you have no GPU available at all or just no dedicated?
As for your suggestion, something like that might be possible but maintaining both rendering mechanisms takes a lot of effort. In a beste case we find a solution to support both use cases without having to maintain two rendering techniques.
I have no dedicated âexternalâ GPU, not sure how to check if I have an integrated one.
I understand your point about having to maintain 2 different rendering techniques and very likely in the near future almost everyone will have some form of GPU available.
For the moment, I will play around with the setting and see which one works best - it is not a big issue for me (and maybe ask Father Christmas for a GPU boardâŚ)
(the fan has been constantly running), I noticed that it made no difference to the fan noise, and opening Task manager I could see that a number of Equo Chromium helpers were still running. I thought Iâd give it a chance to finish doing stuff, if thatâs all it was but some 20 minutes later (as I write this), they were still there even after KNIME had long since closed.
