Percentage of Data and Index Page Use In Buffer Pool


Below is a query you can use to find out how much of the pages in the memory buffer pool actually contain data.  SQL Server does not read parts of  8 KB pages into memory, so if the page fill is poor based on poor table and index design, then you may be holding up a bunch of memory with nothing in it.

SELECT DB_NAME(DATABASE_ID) Database_Name,
        PAGE_TYPE,
                COUNT(DATABASE_ID) / 128 as MB_USED,
SUM(cast(FREE_space_in_bytes as bigint))/(1024.0)/(1024.0) as MB_EMPTY,
(SUM(cast(FREE_space_in_bytes as bigint))/(1024.0)/(1024.0)/ (COUNT(DATABASE_ID) / 128)) * 100.0 as Percentage_Empty
FROM Sys.dm_os_buffer_descriptors
WHERE database_ID = --<YOUR DATABASE ID>
and page_type in ('data_page', 'index_page')
GROUP BY DB_NAME(database_id), page_type;


Example Results:


The source for the buffer pool data is the DMV sys.dm_os_buffer_descriptors.

The function DB_NAME() with take the DATABASE_ID and return the name of the database.  

There are several types of pages in SQL Server but for this we are only looking for data_page and index_page as you can see in the predicate page_type in ('data_page', 'index_page'). 

Each record is for an 8KB page, so we get the MB_USED by dividing the count of 8KB pages by 128.  

In order to get the MB_EMPTY we take the sum of free_space_in_bytes which as the name implies returns the free space in bytes. Casting the result to bigint ensures there is no data type overflow. Dividing by 1024 twice will convert the bytes to megabytes.

Finally, using the same logic for the MB_EMPTY and dividing by the logic used for MB_USED give the percentage of free space by page.

One of the faults with this script is the possibility of dividing by zero.  In my example I ensured that would not happen by limiting the results to a specific database I knew would always return results greater than zero.

As you can see from my example there is plenty of room for improvement in the data pages.  If you have control over the design of the database you could look for really wide tables and further normalize the tables.  Otherwise, if you don't have control, you can look into the use of data compression with the hopes of getting more data onto a single page.    

Popular posts from this blog

Power BI Report Server: "An Error Has Occurred" or Power BI Report Server: "My Power BI Report is Coming up Blank" or Power BI Report Server: "401 and 403” Errors From Nowhere

Image
The information below is an extremely brief summary of the information collected to solve the elusive problem.  There are points I left out because I could not find supporting documentation and I did not want to release suppositions into the universe (why does an NTLM only issue impact a system using kerberos?).  I also want to thank my friend Josh Corrick for staying the course and seeing this through to the end.  See Josh's blog here:  https://corrick.io/blog/ Without warning, our Power BI Report Server began having issues presenting Power BI reports.   The reports would appear to render but would eventually return a blank report canvas.   There was not an error presented and this was impacting all Power BI reports.   An important note is the Power BI reports had worked as expected previously and no changes had been made to the report server such as upgrades or configuration changes. Subsequently, and seemingly unrelated, an error notification was presented when clicking
Read more

SQL SERVER 2017: STRING_AGG

STRING_AGG is a function released in SQL Server 2017.   The Oracle equivalent LIST_AGG has been available since 11gR2.  In the meantime, SQL Server users have resorted to black magic in order to get string values from separate rows. STRING_AGG is very welcome but there are some gotchas to be aware of.  Below is a simple example that works just like you think it would. SELECT STRING_AGG(FIrst_Name, ', ') WITHIN GROUP (ORDER BY First_Name) LIST FROM             (SELECT DISTINCT   First_name FROM [dbo].[Employee]     WHERE First_Name like 'A%' OR First_Name like 'B%') A ORDER BY LIST; You can run the query on any database you have string values and you will get something like this: A David, Aaren, Aaron, Abdul, Bob, Brian... If you want to split by the first letter of the name, do not make the mistake of using an over clause.  You will get an error stating STRING_AGG is not a window function. The function 'STRING_AGG' is not a valid win
Read more

Dynamic Label Positions With SSRS

Image
On a recent tabular "dashboard" with several bar charts in a small area our team found vertical positioning of the labels to be troublesome.  When the labels were set to "bottom" and the bar was too short the label would truncate the significant digits.  On the other hand, when the labels were set to "top" the numbers would be truncated if the column was too tall. Example of bar chart with label set to bottom. Same chart with the middle column label set to top. First we thought about using Smart Labels  but we quickly found that smart labels are not so smart when the labels are turned 90 degrees to make them vertical.  As you can see in the examples above the size of the numbers would make horizontal labels messy. So the obvious choice was to use a function to place the label on top when the bar was small and on the bottom once the bar was tall enough. In Report Builder, go to your chart and select a specific column - not just the chart.
Read more