Forum Discussion
Power BI Capacity Metrics Throttling
Trying to get my head around throttling as we are seeing regular spikes in our interactive operations. More I read more I get confused. Hoping to get some expert advice here. Attaching the image of one of the spikes:
In the first pic, the CU% reaches over 200% at 10:17. In the second pic, the interactive delay occurs at the same time (10:17).
Based on the documentation, my understanding is that the interactive delay occurs when the CUs available during a 10 min window are used up and the susequent usage continues to exceed the threshold. Which means that delays should occur at the end of the 10 min window. In this case, the throttling occurs at the same time when the CU% reaches 200%. How does that happen? I would have thought the delays would occur at 10:27 unless we are saying that usage between 10:07 to 10:17 was already above the threshold. Someone please assist here.
Hi sthatte_admin ,
I think you're mostly correct, but throttling is not triggered at the end of a rolling window, it’s enforced immediately when usage exceeds your allowed limit, based on both instantaneous usage and a short rolling average.
It’s very likely the workload was ramping up before 10:17, so the rolling 10-minute window was already near the limit.
The spike to 200% pushed it well over, so the service throttled interactive requests immediately to stay within the provisioned capacity. This is why the spike and the delay align exactly at 10:17.
3 Replies
- BITomS
Solution Supplier
Hi sthatte_admin ,
I think you're mostly correct, but throttling is not triggered at the end of a rolling window, it’s enforced immediately when usage exceeds your allowed limit, based on both instantaneous usage and a short rolling average.
It’s very likely the workload was ramping up before 10:17, so the rolling 10-minute window was already near the limit.
The spike to 200% pushed it well over, so the service throttled interactive requests immediately to stay within the provisioned capacity. This is why the spike and the delay align exactly at 10:17.
That makes sense. Looking at the graph its not obvious as it doesnt capture the smoothed out CU utlilization over the preceeding 10 mins window to suggest that it was consistenly above 100%.
Thanks for your reply.
- Tutu_in_YYC
Super User
The 10 mins is not referring to the rolling window nor time. It is actually referring to consumption of CU ie 10min worth of CU consumption.
You can use "future" capacity up to 10mins, without throttling. Once you used more than 10mins worth of CU, that when throttling starts.
So to keep things simple, once the light blue spike in pic 2 hit the 100% ( ie 10min of future CU has been used), it throttles.
