In-Memory Optimization in SQL Server: A Simple Experiment

In this article, we will be discussing about In-Memory Optimization in SQL Server, and we will see a simple example, of optimizing a workload using In-Memory Optimization and thus improve performance.

Introduction to SQL Server In-Memory Optimization

In one of my previous articles I briefly introduced In-Memory Optimization in SQL Server. In this article we will conduct a simple experiment utilizing this new technology and discussing the benefits. The experiment will be organized in the following steps:

1. Create the In-Memory processing enabled database
2. Create the tables (disk-based, memory-enabled)
3. Populate the tables with sample data
4. Prepare the processing code for three scenarios (execution modes):
   • Disk-based processing
   • In-Memory optimization
   • In-Memory optimization with natively compiled stored procedure
5. Run all three execution modes and measure execution times
6. Discuss findings

OK, enough words, let’s write some code!

Step 1: Create the In-Memory processing enabled database

--Creates a memory-enabled database by adding a memory-optimized filegroup with a specified directory.
--Note: This assumes that directory c:\tmp exists (the subdirectory is created by SQL Server).
CREATE DATABASE InMemDB;
GO
ALTER DATABASE InMemDB
ADD FILEGROUP [inMemOLTP_fg] CONTAINS MEMORY_OPTIMIZED_DATA
ALTER DATABASE InMemDB
ADD FILE ( name = [inMemOLTP_fg_dir],
   filename= 'c:\tmp\InMemOLTPDir')
   TO FILEGROUP [inMemOLTP_fg];
GO

--Create sample tables representing a scenario involving 200000 Products
USE InMemDB;
GO

--Create traditional table (disk-based)
CREATE TABLE tblDiskBased(
id INT PRIMARY KEY,
code VARCHAR(50),
descr VARCHAR(250),
price float
);
GO

--Create a durable (data will be persisted) memory-optimized table
--Alternative choice: SCHEMA_ONLY = Data will be lost if the server turns off unexpectedly
CREATE TABLE tblMemOptimized(
id INT NOT NULL PRIMARY KEY NONCLUSTERED HASH WITH (BUCKET_COUNT=400000),
code VARCHAR(50),
descr VARCHAR(250),
price float
)WITH (MEMORY_OPTIMIZED=ON, DURABILITY = SCHEMA_AND_DATA);
GO


--Create a durable (data will be persisted) memory-optimized table
--Alternative choice: SCHEMA_ONLY = Data will be lost if the server turns off unexpectedly
--Note: This table will be used by the natively-compiled stored procedure
CREATE TABLE tblMemOptimized2(
id INT NOT NULL PRIMARY KEY NONCLUSTERED HASH WITH (BUCKET_COUNT=400000),
code VARCHAR(50),
descr VARCHAR(250),
price float
)WITH (MEMORY_OPTIMIZED=ON, DURABILITY = SCHEMA_AND_DATA);
GO

So far: We have created a memory enabled database and data structures. Now, it’s time to populate the tables with sample data.

Learn more about SQL Server In-Memory OLTP – Enroll to the Course!

We recommend enrolling to our online course titled  “Boost SQL Server Database Performance with In-Memory OLTP” and learn how to harness the power of SQL Server’s In-Memory Optimization and boost the performance of your data processes!

In this course, you will learn all about SQL Server’s In-Memory Database Processing Engine also known as In-Memory OLTP. To this end, you will learn how to enable In-Memory OLTP in SQL Server, what memory-optimized tables and natively-compiled stored procedures are, as well as, how to boost the performance of your data processes using this powerful SQL Server feature.

Learn More

Step 3: Populate the tables with sample data

--Populate the tables with sample data (200K records)
--tblMemOptimized
DECLARE @i INT
SET @i=0
WHILE (@i<200000)
BEGIN
INSERT INTO tblMemOptimized
SELECT CAST(@i as varchar(6)) as id, ('Code '+ (CAST (@i AS varchar(6)))) as code, ('Description '+ (CAST (@i AS varchar(6)))) as descr, (10 + (500-10)*RAND()) as price
SET @i=@i+1
END
GO

--Populate the tables with sample data (200K records)
--tblDiskBased
--tblMemOptimized2
INSERT INTO tblDiskBased
SELECT * FROM tblMemOptimized;
GO

INSERT INTO tblMemOptimized2
SELECT * FROM tblMemOptimized;
GO

All three tables now have same data (200K rows each). You can easily check it out!

--Disk-based processing
CREATE PROCEDURE updateDiskBased as
BEGIN
UPDATE dbo.tblDiskBased
SET price=price-(price*0.05)
END
GO

--Memory-Optimized
CREATE PROCEDURE updateMemoryOptimized as
BEGIN
UPDATE dbo.tblMemOptimized
SET price=price-(price*0.05)
END
GO

--Natively-Compiled Stored Procedure
CREATE PROCEDURE updateMemoryOptimizedNatively
with native_compilation, schemabinding, execute as owner
as 
begin atomic with (transaction isolation level = snapshot,language = N'English')
UPDATE dbo.tblMemOptimized2
SET price=price-(price*0.05)
END
GO

Discussion

First of all, from the above figure we can see that the output of the three update operations is exactly the same which is something that validates the correctness of the three different execution modes.

As you can see from the screenshot with execution times (Figure 2), the slowest execution was the one that involved the disk-based table (elapsed time: 1924 ms) having also the highest CPU time (328 ms).

The fastest execution was the one of the memory-optimized table with the natively compiled stored procedure where the elapsed time was 329 ms and the CPU 93 ms!

The memory optimized execution was faster than the disk-based with 404 ms elapsed time and 250 ms CPU time but it was not faster than the memory-optimized scenario combined with the natively compiled stored procedure.

In-Memory optimization in SQL Server is a great new technology which can significantly improve performance in many cases. Especially in cases of certain workloads such as staging processes in Data Warehouse systems, high data insert rates (i.e. smart metering) and low latency execution scenarios (i.e. investment trading), In-Memory optimization can boost performance. Of course, before applying it on Production systems, a proper analysis is required along with testing.

Read Also:

• Boost SQL Server Database Performance with In-Memory OLTP (Course Preview)
• Frequently Asked Questions About SQL Server In-Memory OLTP
• In-Memory Optimization in SQL Server: A Simple Experiment
• Introducing In-Memory Optimization in SQL Server
• Getting Started with SQL Server In-Memory OLTP – Part 1
• Getting Started with SQL Server In-Memory OLTP – Part 2
• Where are temporary tables stored in SQL Server?
• SQL Server 2016: TempDB Enhancements
• tempdb growth
• …more