Reduce CPU for Microsoft.PowerBI.EnterpriseGateway.exe

[MiLANche]

Hi,  we have on-prem data gateway  as virtual machine  with 6 (six!) CPU on 2.5GHz  AMD EPYC 7502

For 18 minutes these CPU go to 100%  and the main process responsible are

   Microsoft.PowerBI.EnterpriseGateway.exe

   Microsoft.Mashup.Container.NetFX45.exe

VMware is complaining and showing alert to Red.  Other 23 hours and 40 minuts this server is idle?

Please help to reduce this CPU impact or

How to configure the Low Impact  Setting in  the  On-Premises data gateway   v3000.79.6  (April 2021)  

Many Thanks!

[d_gosbell]

That is not likely to happen. The gateway is designed to process the data as fast as possible so that your data is refreshed as fast as possible. If you ran at 50% CPU then the process would take twice as long.

Better options would be to look at what data refresh is running during that time frame and see if you can tune it at all to reduce the CPU load or change the alert settings on your hypervisor for this VM

[MiLANche]

Thanks mate,  On the other hand, I've asked PBI admins to let me know (as System Admin)  what types are 

Cache Data workload   (using high CPU)

Live Connection and Direct Query  workload  (much less demand for CPU)

[d_gosbell · Accepted solution]

---

@MiLANche wrote:

Thanks mate,  On the other hand, I've asked PBI admins to let me know (as System Admin)  what types are 

Cache Data workload   (using high CPU)

Live Connection and Direct Query  workload  (much less demand for CPU)

---

The different workloads are outlined here:
On-premises data gateway sizing - Power BI | Microsoft Learn

The cache data workload is where Power BI is refresh an import mode model so it is processing data to be loaded into an import mode data model.

Typically the Power Query code will attempt to push as much of the transformation logic upstream to the data source as it can, but if the transformations are complex or are not well tuned or if the data source cannot handle complex logic (such as a folder of csv files) the the bulk of the transformation logic will be done on the gateway CPU. 

As I mentioned earlier, I still think it would be worth getting someone to review the report in question or look at adjusting your VMWare alerts rather than slowing the performance of all gateway operations.

[MiLANche]

the high CPU Reports are runing for 20 minutes - no problem if they run for 40 min  (from 7am until 7:40am)

But, we have biiig problem with alerts and VMware lossing conection with 

this Power BI instance as it runs for 20 minutes on 100% CPU.

VMware vCenter v7.0.3 shows status of machine as OffLine  (powered off?)  but  VM is stall runing (for past 59 days ..)   .. the VMware cluster is hosting 71 VM's and this one is kind of crazy 🙂

[v-xiaoyan-msft]

Hi @MiLANche ,

Gateway sizing recommendations depend on many variables. In this section, we provide you with general recommendations that you can take into consideration.

On-premises data gateway sizing - Power BI | Microsoft Learn

Adjust gateway performance based on server CPU | Microsoft Learn

Best regards.
Community Support Team_ Caitlyn

[MiLANche]

Thanks fo rthe links!   as general reading. 

Our Curent setting for Maximum container count for Power BI Direct Query.

…

<setting name="MashupDQPoolContainerMaxCount" serializeAs="String">

        <value>20</value>

There is no guideline  To reduce this to 10 ?

[MiLANche]

Maybe someone from Microsoft can tell Developers to include

one option / setting  in the  On-Premises data gateway   (running on our dedicated Power BI servers)
   CPU impact - Low  (CPU will run under 50% )

   CPU imapct - Medium  (CPU will run under 80%)

   CPU impact - Max  (MS.PowerBI.EnterpriseGateway.exe   will use all virtual CPU on 100% - and generate alerts on underlaying hyopervisor 🙂