Throttling in Fabric Is Less of a Concern Now—Here’s Why

Managing capacity has become challenge on its own. You can either performance tune the data products or throw money at the problem. However, Microsoft has decided to help us out with this problem by Bursting and Smoothing the capacity to avoid throttling as much as possible. Bursting through capacity refers to borrowing the capabilities of higher levels of SKUs in times of throttling without any additional cost to the customers.  Sounds Unbelievable, right? Why would Microsoft giveaway free money?Because of smoothing.Smoothing is what can be called compensation for the bursting.   For every extra Capacity Unit that is consumed during peak they are compensated from capacity in the next 24 hours.Microsoft is not giving away free computing power,  It’s simply letting you borrow from the future using the very computing power you bought!

Now look at the specifics of this offering: 

Let’s break it down with an example: 

• If you purchase an F2 capacity, you get 2 Capacity Units (CUs) per second for the entire 24-hour period. 

• That means your total allocated compute power is:

  
  

  2 CUs × 60 sec × 60 min × 24 hrs = 172,800 CUs per day 

Now, let’s say your workload runs at an average of 1.5 CUs per second, but you have a peak usage of 10 CUs for 60 minutes when performing a full data load. 

• CUs consumed in 24 hours at an average load:

  
  

  1.5 CUs × 60 sec × 60 min × 24 hrs = 129,600 CUs 

• Extra CUs consumed during peak load (since F2 provides only 2 CUs/sec, but you used 10 CUs/sec for 60 minutes):

  
  

  (10 - 2) CUs × 60 sec × 60 min = 28,800 CUs 

• Total CUs used:

  
  

  129,600 + 28,800 = 158,400 CUs 

• Remaining CUs:

  
  

  172,800 - 158,400 = 14,400 CUs left 

Since you didn’t use all your allocated CUs, your Fabric capacity smoothly handled the peak usage without throttling. 

Looking at Per-Second Consumption 

• You still have 0.17 CUs per second left on average throughout the day. 

• Your peak burst of 10 CUs for 60 minutes consumed 0.33 CUs per second when averaged over 24 hours—well within your allocated budget. 

We were able to use 10 CUs for an hour without ever needing to buy F8 or F16.

Key Takeaway:Fabric allows you to burst beyond your per-second allocation by borrowing from your unused capacity, as long as your total CU usage stays within your daily allowance. However, sustained overuse can lead to throttling if you exceed your total available CUs.

However, there is more to smoothing than just evening out the CUs.It lies in the difference in how the capacity handles ad-hoc interactive user queries vs scheduled jobs: 

• Ad-hoc interactive queries are typically smoothed in the next 5 minutes, to avoid many short-term spikes 

• Scheduled workloads are smoothed in the next 24 hours 

If the workload exceeds the limit, there are three possible outcomes, depending on the “future” availability of the capacity: 

• Next 10 minutes full – ad-hoc interactive queries will be delayed 

• Next 60 minutes full – ad-hoc interactive queries will be rejected 

• Next 24 hours full – scheduled workloads will be rejected 

This whole ordeal of bursting and smoothing helps us plan the capacity usage much better instead of just focusing on peaks.