MS Fabric Pricing/Bursting/Smoothing

[umb2022]

Hello Community

I am investigating MS Fabric in details and have a few questions about pricing and capacity choices. I will list my questions in a list, and it would be good, if they are answered the same way.

I will use F2 as an example in all questions.

1. Currently the price of a F2 (2 CUs) capacity is 0,38 dollars per hour and 277,4 dollars per month in North Europe:

• Does this mean, if i run the capacity 24 hours a day for a month my monthly cost would be 277,4 dollars maximum in the pay as you go model (exluding cost for storage, bandwitch and network)?

2. Questions about bursting, smoothing and throttling in the pay as you go model:

1. If i have a heavy workload from 4 am to 6 am, will bursting make the workload faster than 2 hours or how does it work?
2. If i pause right after the heavy workload lets say at 6 am, then resume again at 10 am, would i have to pay for the extra CUs obtained from 4 am to 6 am because smoothing could not happen as the capacity was paused?
3. What happens if CUs that are bursted are not able to be smoothed in a 24 hour period, would it increase the cost above the 277,4 dollars or carry over to next day?

3. Questions about pricing, bursting, smoothing and throttling in the reserved model:

1. The reserved price for an F2 is much cheaper than the monthly cost when running F2 capacity 24 hours every day in a month, but if i only use the capacity 200 hours a month, wouldnt it be better to choose the pay as you go model?
2. How does bursting and smoothing work in the reserved model - is it the same as the pay as you go model?
3. Is it possible to pause a reserved capacity?

[dtjdtj]

Hello all,

Can somone confirm whether smoothing/bursting policies apply to the non-Power BI Fabric workloads too (e.g. notebooks/pipelines, dataflows gen2 consumption)?

Thanks in advance!

[dbrowne · Accepted solution]

---

@umb2022 wrote:

Hello Community

I am investigating MS Fabric in details and have a few questions about pricing and capacity choices. I will list my questions in a list, and it would be good, if they are answered the same way.

I will use F2 as an example in all questions.

1. Currently the price of a F2 (2 CUs) capacity is 0,38 dollars per hour and 277,4 dollars per month in North Europe:

• Does this mean, if i run the capacity 24 hours a day for a month my monthly cost would be 277,4 dollars maximum in the pay as you go model (exluding cost for storage, bandwitch and network)?

Yes.

2. Questions about bursting, smoothing and throttling in the pay as you go model:

1. If i have a heavy workload from 4 am to 6 am, will bursting make the workload faster than 2 hours or how does it work?

Yes.  You may use more than 2 CU seconds per second when you need to.  This is "bursting".  "Smoothing" allocates the CU Seconds over 5min for interactive and 24hours for background.

1. If i pause right after the heavy workload lets say at 6 am, then resume again at 10 am, would i have to pay for the extra CUs obtained from 4 am to 6 am because smoothing could not happen as the capacity was paused?

When you "burst" over the capacity quota the extra CU Seconds (aka "overage") are "carried forward" and "burned down" when there is available capaicty.  If you pause a capacity that has an outstanding overage, the overage is immediately billed.  When you resume the capacity the overage is gone.

1. What happens if CUs that are bursted are not able to be smoothed in a 24 hour period, would it increase the cost above the 277,4 dollars or carry over to next day?

The capacity will reject new requests until capacity is available, or until you scale up or pause/resume the capacity.

3. Questions about pricing, bursting, smoothing and throttling in the reserved model:

1. The reserved price for an F2 is much cheaper than the monthly cost when running F2 capacity 24 hours every day in a month, but if i only use the capacity 200 hours a month, wouldnt it be better to choose the pay as you go model?

Yes.  Depending on your usage pattern, pause/resume/scale up/scale down may be more cost effective than a reserved instance.  This is the case across Azure.  You can also purchase a RI for a smaller size, and scale up over that as needed at the pay-go price.

1. How does bursting and smoothing work in the reserved model - is it the same as the pay as you go model?

Exactly the same.

1. Is it possible to pause a reserved capacity?

Yes.  But if you don't have any running capaicties that match the RI then you loose the reserved capacity you paid for while you're not running.

---

[umb2022]

Thank you for the informed answer!

I have 3 follow up questions:

Quote 1:

"Yes.  Depending on your usage pattern, pause/resume/scale up/scale down may be more cost effective than a reserved instance.  This is the case across Azure.  You can also purchase a RI for a smaller size, and scale up over that as needed at the pay-go price."

Question 1: so this means that I can buy a reserved capacity, F2 and scale up using pay as you go to the F8, if needed?

Quote 2:

"The capacity will reject new requests until capacity is available, or until you scale up or pause/resume the capacity."

Question 2: I understand that new requests will be declined, I saw the table regarding the 4 levels of throttling, but if I pause and resume, would that give me new capacity?

and does this mean that the overused CUs will carry over to the next day, so jobs running in the morning may be affected, because I overused the capacity too much the day before - in other words, is overage based on a billing month or a hard 24 hours limit (background tasks)? Let's say I don't use my capacity that much in the betting and middle of the month, but in the end of a month, I use it very heavily, how can it carry over?

Quote 3:

"Yes.  You may use more than 2 CU seconds per second when you need to.  This is "bursting".  "Smoothing" allocates the CU Seconds over 5min for interactive and 24hours for background."

Question 3:

If I have a F4, that would mean I would be able to user more than 4 CU seconds per second right? F8 = more than 8 CU seconds per second?

[dbrowne]

>Question 1: so this means that I can buy a reserved capacity, F2 and scale up using pay as you go to the F8, if needed? Yes. > I pause and resume, would that give me new capacity? Yes. The outstanding overage is billed and cleared. And 24 hours is the maximum smoothing period. Utilization cannot be smoothed over multiple days or weeks. >If I have a F4, that would mean I would be able to user more than 4 CU seconds per second right? F8 = more than 8 CU seconds per second? Yes. An F4 has 4 CUs and generates 4 CU seconds per second, an F8 has 8 CUs...

[umb2022]

Hello, I have a follow up question, because I dont totally understand the reply:

I asked:What happens if CUs that are bursted are not able to be smoothed in a 24 hour period, would it increase the cost above the 277,4 dollars or carry over to next day?

You answered: The capacity will reject new requests until capacity is available, or until you scale up or pause/resume the capacity.

New question:

It makes sense that new requests are rejected, it also makes sense that scalling up fixes the issue, but how can pause/resume of the capacity suddenly handle the overage, if overutilizing the CUs lead to the capacity not being able to smooth over a 24 hour period - since each capacity contains a set number of CU seconds (F2 = 2*24*60*60 = 172.800 seconds). The only answer that makes sense is it would reject new requests or I would have to scale up.

I am looking forward to your answer, thanks. 

[dbrowne]

Pausing a F-SKU capacity with an overage immediately bills for all the outstanding utilization.  So when you Resume.

[jjzz]

@dbrowne Hi,I have been following this issue for a long time, and the previous information has been very helpful. However, I still have some questions regarding the billing after pausing due to overage.

What is the billing rate? For example, if my capacity is F2 and I have an overage of 10 CUs, is this 10 CUs divided by 2 CUs/second and then billed according to the F2 per-minute price?

I am looking forward to your answer, thanks. 

[Anonymous]

Hello @umb2022 ,

Thanks for using Fabric Community.
At this time, we are reaching out to the internal team to get some help on this.
We will update you once we hear back from them.