Azure SQL DB tiers comparison

Hi All

In the last few month Brent Ozar gae us 2 masterpiece blogs related to Azure SQL DB:

 

How fast can a $21,468/mo Azure SQL DB load data? 

 

In this blog Brent compared the abilities of Azure SQL DBs to load Data - he compared all combinations of vCors tiers. (When I asked him about comparing the Standard\Premium tiers, he told me to do it.... :-) )

 

There’s a bottleneck in Azure SQL DB storage throughput.

 

In this blog Brent showed us that in the vCors world the storage throughput has limit and there is not need to pay so much money when you need to upload lots of data.
 

So I took have taken up his challenge and done a comparison in Azure SQL DB in Standard\Premium tiers.

I have created a new DB with 1 Table.

I have generated 7 GB of DATA, and created the file in my local on premise drive (Yes, do not kill me, I did not had the time to put it on azure), and uploaded it via BCP command.

 

bcp "TableName" in "T:\MyTable.bcp"  -S"myservername.database.windows.net" -UUserName@myservername -P"Password" -n -d"DBName"

 

After each test I have truncate the table and changed the tier.

 

Here are the main results:

  

	Rows/second	Time(MS)	Cost per month
S2	4,071	18,184,812	$                    73.0
S4	4,083	18,129,531	$                  294.0
S6	9,007	8,219,047	$                  588.0
S9	18,733	3,952,062	$              2,354.0
S12	15,054	4,917,703	$              4,415.0
P1	3,670	20,168,375	$                  456.0
P4	22,851	3,239,875	$              1,825.0
P11	22,188	3,336,688	$              6,868.0
P15	24,487	3,023,344	$            15,698.0

 

It is clearest to see if we view the results in a chart of cost vs Rows\Second.

 

 

What does the Charts shows us?

Most cost effective is S9 tier, but as we would expect P15 is the fastest  

 

Below are various graphs of resource utilization for each tier.

 

S2 tier Log IO and Data IO took time

 

 

S4 Had only LOG IO issues

 

S6 - the same chart.

 

S12 Log IO is 100% and I have also added the DTU.

P4 looks the Same





 

 

P11 start to change Log IO was 100% only for a short time

 

 

P15 \ P11 \ P4 on the same Chart

 

 

 

In conclusion - Just as Brent found, Log IO is the bottleneck, the storage write operations. 

Log IO is the bottleneck but you get what you pay for, the more you pay the more I/O. Therefore the effects of the Log IO bottleneck are reduced the more resources/money you throw at the problem.

 

Next time I will test reads operations and CPU's

 

Have a good day.