Microsoft Azure

For in-memory databases, Microsoft has several offerings. Azure SQL Database In-Memory OLTP is aimed at transactional workloads, while Azure SQL In-Memory Columnstore is aimed at analytic workloads. Additionally, there is SQL Server In-Memory OLTP and SQL Server Hybrid Buffer Pool. Hence Microsoft has chosen to put out use case-specific offerings in this area rather than have a single in-memory database. The Microsoft in-memory products compete against alternatives such as Oracle, SAP HANA, Redis and specialist products such as Dragonfly, Memcached, Singlestore and Volt DB. 

Azure SQL Database In-Memory OLTP stores data entirely in memory, with queries compiled into native machine code for quicker execution. There is an additional feature to reduce contention between transactions, known as latch-free data access. Azure SQL In Memory Columnstore is suitable for analytic processing workloads, since column-based storage is faster than row-based for queries, at the expense of load and update speed. Of course, for analytic queries on a data warehouse there would not typically be any updates other than incremental data loads, which makes the approach well suited for this use case. By additionally storing the columns of data in memory, performance is significantly improved, and as a side benefit, the database indexes usually used to speed up queries are not needed, as everything is already in memory.

SQL Server Hybrid buffer pool uses persistent memory devices, solid-state byte-addressable memory for even faster data access. This reduces IO operations, and is suitable for bulk transaction operations or use cases where there is heavy use of read operations and where performance is critical.

The Microsoft Azure SQL Database In-Memory OLTP is suitable for applications like financial trading, gaming and data ingestion from sensors. Similarly, SQL Server In-Memory OLTP also stores data in memory and translates queries to machine code, with optimistic concurrency to reduce lock contention. This approach allows transactions to proceed until commit, and only then checks for conflicts, which speeds things up in use cases where lock conflicts are rare. To speed up a database you can either begin with new memory-optimized tables or convert existing, disk-based ones; the product provides performance reports to help you identify tables that would benefit, and there is a memory optimization wizard to simplify migration. Stored procedures will typically need to be converted into natively compiled code.

For Azure SQL In Memory Columnstore, the base tables use columnar compression, reducing storage volumes and in itself yielding performance improvements for analytic workloads.

In-memory databases are designed for performance-critical applications where low latency and real-time performance are critical, and where the extra cost of memory compared to disk storage is justified (memory is an order of magnitude pricier than disk). Examples of such use cases are multi-player on-line gaming, real-time, fraud detection, financial trading systems, caching and managing user sessions in web applications, or real-time analytics. While the improvements in speed will vary by use case, it is not uncommon to see ten times improvements in performance compared to regular disk-based database access. Clearly this is a major boost in performance-critical applications.

The bottom line

Microsoft offers a range of in-memory options that are tailored to specific uses cases such as transaction processing or analytic workloads, and these can be used to significantly speed up performance in such critical applications. There are in-memory solutions from this vendor for both on-premises and cloud deployments, such as Azure. The Microsoft database applications have a large customer base and are likely to be on most shortlists for those considering an in-memory database, especially for customers who already have a substantial commitment to the Microsoft ecosystem, such as using the Azure cloud.