Microsoft SQL Server 2016 comes with a variety of new additions and I am happy for exploring the new features of SQL Server 2016.

Microsoft SQL Server 2014 introduced Hekaton engine which used for In-Memory Database objects.

SQL Server 2014: Create Memory Optimized File Group and Table

SQL Server 2016 introduced Columnstore index for Memory-Optimized Table which was not possible previously.
If you are planning to store large amount of data In Memory, you can apply Columnstore Index on that In Memory OLTP Table.

It comes with one condition: you can not apply Columnstore Index on the particular column. It must apply to all the columns of In-Memory Table.

Small demonstration with all the possible errors:

Create a sample memory optimized table with column store index:

CREATE TABLE tbl_Employees
(
	EmpID INT
	,EmpName VARCHAR(50)
	,EmpSalary INT
	,EmpDOJ DATETIME
	,INDEX idx_tbl_Employees_columnstore CLUSTERED COLUMNSTORE 
)
WITH (MEMORY_OPTIMIZED = ON)
GO

You will get a below error:

Msg 41321, Level 16, State 7, Line 1
The memory optimized table 'tbl_Employees' with DURABILITY=SCHEMA_AND_DATA must have a primary key.

Add primary key on script and execute it:

CREATE TABLE tbl_Employees
(
	EmpID INT PRIMARY KEY
	,EmpName VARCHAR(50)
	,EmpSalary INT
	,EmpDOJ DATETIME
	,INDEX idx_tbl_Employees_columnstore CLUSTERED COLUMNSTORE 
)
WITH (MEMORY_OPTIMIZED = ON)
GO

You will get a below error:

Msg 12317, Level 16, State 78, Line 3
Clustered indexes, which are the default for primary keys, are not supported with memory optimized tables. Specify a NONCLUSTERED index instead.

Apply Non Clustered Primary key and execute it:

CREATE TABLE tbl_Employees
(
	EmpID INT PRIMARY KEY NONCLUSTERED
	,EmpName VARCHAR(50)
	,EmpSalary INT
	,EmpDOJ DATETIME
	,INDEX idx_tbl_Employees_columnstore CLUSTERED COLUMNSTORE 
)
WITH (MEMORY_OPTIMIZED = ON)
GO

If you have not created Memory Optimized File group and related files, you may get a below error:

Msg 41337, Level 16, State 100, Line 1
Cannot create memory optimized tables. To create memory optimized tables, the database must have a MEMORY_OPTIMIZED_FILEGROUP that is online and has at least one container.

Create memory optimized file group and data files:

USE [master]
GO
ALTER DATABASE [Database_Name] 
ADD FILEGROUP [Memory_Optimized] CONTAINS MEMORY_OPTIMIZED_DATA 
GO
ALTER DATABASE [Database_Name] 
ADD FILE ( NAME = N'HekatonMemory', FILENAME = N'C:\Hekaton\hek.mdf' ) 
TO FILEGROUP [Memory_Optimized]
GO

SQL Server 2014 introduced the concept of In-Memory Database which provides the best combination of RDBMS + In-Memory Database.

SQL DBA is monitoring the size of tables, indexes and other fragmentation related information.

When we say that SQL Server tables stored in memory, we should also keep our eyes open on the utilization of memory.

SQL Server 2014 also introduced one DMV (sys.dm_db_xtp_table_memory_stats) to check the memory utilization of Memory Optimized Tables.
Using this view, you can find memory information like allocated memory and used memory for tables and indexes.

Executed below query on AdventureWorks2016ctp3:

SELECT  
	OBJECT_ID 
	,OBJECT_SCHEMA_NAME(object_id) + '.' + OBJECT_NAME(object_id) AS TableName
	,memory_allocated_for_table_kb 
	,memory_used_by_table_kb 
	,memory_allocated_for_indexes_kb 
	,memory_used_by_indexes_kb
FROM sys.dm_db_xtp_table_memory_stats
GO

Please visit other related articles:

SQL Server 2014: What is Hekaton?

SQL Server 2014: Create Memory Optimized File Group and Table

SQL Server: Find the total row count and size of the Tables

The Hekaton is a new database engine, also known as SQL Server In-Memory OLTP.
Hekaton collaborated with Microsoft Research and released with Microsoft SQL Server 2014.

The Hekaton is a Greek word and meaning is 100 because that newly launched in-memory database which is 100 times faster than the existing SQL Server engine.

If you don’t know about In-memory database, you can visit a below article.

Database Theory: What is In-memory Database?

If we look at traditional RDBMS system, memory resources are expensive, and mostly optimize for disk storage.

The Hekaton allows us to work with memory-optimized tables and indexes which entirely stores in the main memory, and still, we can access via T-SQL like any other regular tables.

We can also create a stored procedure by adding the “WITH NATIVE_COMPILATION” clause that only refers to the memory-optimized tables.

The Hekaton eliminates logical locks using automatic SNAPSHOT ISOLATION with multi-versioning, also having a lock free and latch free structures for accessing the data.

How Hekaton is different than SQL Server standard relational engine?

Very simple answer,

SQL Server standard relational engine developed basically for those data which resides on disk, and it reads and writes 8 KB per data pages which are available in the extents. This process requires lots of CPU I/O and memory internal operations.

With the memory-optimized tables, there are no any data pages and extents. It stores only the data rows into memory with index pointer.

The engine also supports MVCC (Multiversion Concurrency Control) in which same data row has may be the different versions at a time so last updated version accessible by last most requests.

Because of MVCC nature, reader and writer process never block to each other in the Hekaton engine.

What is Multi Version Concurrency Control (MVCC)

The In-memory database, also known as a Main-memory database. (IMDB or MMDB).
As a Database Architecture, I like this topic and implementation in which data reside in the main memory area.

What is In-memory database?

In-memory database is one kind of database whose data stored in main memory for faster response and quick retrieval. The data loads into system memory in a compressed form.
The In-memory databases are faster than extensive disk-based databases because it doesn’t require any Disk I/O operations.
The internal algorithms are also simpler and require few CPU instructions.

Why not RDBMS Database Caching?

Most of RDBMS products provide a feature of database caching, but with the use of In-memory database, we should not have only one goal to avoid disk I/O.

The database caching we can use only for reading data and for writing data it requires again disk I/O, even managing database cache is also quite challenging and it also requires strong memory and CPU.

If a system or main memory crash, what will happen?

Once power switch off for device, all stored information is also removed so we can say IMDB not support “durability” of the ACID property.

Most of the IMDB systems offer features like to take snapshots periodically and write to non-volatile media. If system crash or fail, database Rollback to last completed transaction.
We can also increase the durability of IMDB by adding one or more copy of the database.
We can also configure replication and failover procedure for IMDB systems.

What kind of applications typically uses In-memory database?

The IMDB system is one type of analytic database and utilised for the read-only system. It uses by Data warehouse and Data mart for accessing historical data.
The real-time embedded system is also using an IMDB system for fast manipulation.
Other applications like IP-Network routing, telecom switching and DTH system are also using IMDB systems.

You can visit this reference to find list of IMDB Systems.