Buffer Pool Extension
Introduced in SQL Server 2014, the buffer pool extension provides the seamless integration of a nonvolatile random access memory (that is, solid-state drive) extension to the Database Engine buffer pool to significantly improve I/O throughput. The buffer pool extension is not available in every SQL Server edition. For more information, see Features Supported by the Editions of SQL Server 2016.
The primary purpose of a SQL Server database is to store and retrieve data, so intensive disk I/O is a core characteristic of the Database Engine. Because disk I/O operations can consume many resources and take a relatively long time to finish, SQL Server focuses on making I/O highly efficient. The buffer pool serves as a primary memory allocation source of SQL Server. Buffer management is a key component in achieving this efficiency. The buffer management component consists of two mechanisms: the buffer manager to access and update database pages, and the buffer pool), to reduce database file I/O.
Data and index pages are read from disk into the buffer pool and modified pages (also known as dirty pages) are written back to disk. Memory pressure on the server and database checkpoints cause hot (active) dirty pages in the buffer cache to be evicted from the cache and written to mechanical disks and then read back into the cache. These I/O operations are typically small random reads and writes on the order of 4 to 16 KB of data. Small random I/O patterns incur frequent seeks, competing for the mechanical disk arm, increasing I/O latency, and reducing aggregate I/O throughput of the system.
The typical approach to resolving these I/O bottlenecks is to add more DRAM, or alternatively, added high-performance SAS spindles. While these options are helpful, they have significant drawbacks: DRAM is more expensive than data storage drives and adding spindles increases capital expenditure in hardware acquisition and increases operational costs by increased power consumption and increased probability of component failure.
The buffer pool extension feature extends the buffer pool cache with nonvolatile storage (usually SSD). Because of this extension, the buffer pool can accommodate a larger database working set, which forces the paging of I/Os between RAM and the SSDs. This effectively offloads small random I/Os from mechanical disks to SSDs. Because of the lower latency and better random I/O performance of SSDs, the buffer pool extension significantly improves I/O throughput.
The following list describes the benefits of the buffer pool extension feature.
Increased random I/O throughput
Reduced I/O latency
Increased transaction throughput
Improved read performance with a larger hybrid buffer pool
A caching architecture that can take advantage of present and future low-cost memory drives
The following terms are applicable to the buffer pool extension feature.
- Solid-state drive (SSD)
Solid-state drives store data in memory (RAM) in a persistent manner. For more information, see this definition.
In SQL Server, A buffer is an 8-KB page in memory, the same size as a data or index page. Thus, the buffer cache is divided into 8-KB pages. A page remains in the buffer cache until the buffer manager needs the buffer area to read in more data. Data is written back to disk only if it is modified. These in-memory modified pages are known as dirty pages. A page is clean when it is equivalent to its database image on disk. Data in the buffer cache can be modified multiple times before being written back to disk.
- Buffer pool
Also called buffer cache. The buffer pool is a global resource shared by all databases for their cached data pages. The maximum and minimum size of the buffer pool cache is determined during startup or when the instance of SQL server is dynamically reconfigured by using sp_configure. This size determines the maximum number of pages that can be cached in the buffer pool at any time in the running instance.
A checkpoint creates a known good point from which the Database Engine can start applying changes contained in the transaction log during recovery after an unexpected shutdown or crash. A checkpoint writes the dirty pages and transaction log information from memory to disk and, also, records information about the transaction log. For more information, see Database Checkpoints (SQL Server).
SSD storage is used as an extension to the memory subsystem rather than the disk storage subsystem. That is, the buffer pool extension file allows the buffer pool manager to use both DRAM and NAND-Flash memory to maintain a much larger buffer pool of lukewarm pages in nonvolatile random access memory backed by SSDs. This creates a multilevel caching hierarchy with level 1 (L1) as the DRAM and level 2 (L2) as the buffer pool extension file on the SSD. Only clean pages are written to the L2 cache, which helps maintain data safety. The buffer manager handles the movement of clean pages between the L1 and L2 caches.
The following illustration provides a high-level architectural overview of the buffer pool relative to other SQL Server components.
When enabled, the buffer pool extension specifies the size and file path of the buffer pool caching file on the SSD. This file is a contiguous extent of storage on the SSD and is statically configured during startup of the instance of SQL Server. Alterations to the file configuration parameters can only be done when the buffer pool extension feature is disabled. When the buffer pool extension is disabled, all related configuration settings are removed from the registry. The buffer pool extension file is deleted upon shutdown of the instance of SQL Server.
We recommend that you follow these best practices.
The buffer pool extension size can be up to 32 times the value of max_server_memory. We recommend a ratio between the size of the physical memory (max_server_memory) and the size of the buffer pool extension of 1:16 or less. A lower ratio in the range of 1:4 to 1:8 may be optimal. For information about setting the max_server_memory option, see Server Memory Server Configuration Options.
Test the buffer pool extension thoroughly before implementing in a production environment. Once in production, avoid making configuration changes to the file or turning the feature off. These activities may have a negative impact on server performance because the buffer pool is significantly reduced in size when the feature is disabled. When disabled, the memory used to support the feature is not reclaimed until the instance of SQL Server is restarted. However, if the feature is re-enabled, the memory will be reused without restarting the instance.
You can use the following dynamic management views to display the configuration of the buffer pool extension and return information about the data pages in the extension.
Performance counters are available in the SQL Server, Buffer Manager Object to track the data pages in the buffer pool extension file. For more information, see buffer pool extension performance counters.
The following Xevents are available.
Fires when a page or group of pages are evicted from the buffer pool and written to the buffer pool extension file.
Fires when a page is read from the buffer pool extension file to the buffer pool.
Fires when a page is evicted from the buffer pool extension file.
Fires when the eviction threshold is calculated.
Enable and configure the buffer pool extension.
Modify the buffer pool extension configuration
View the buffer pool extension configuration
Monitor the buffer pool extension