Databases store data in secondary storage devices. When applications are built on databases to analyze the data, disk I/O operation is the main problem in data throughput. The CPU waits for the data to be loaded from the disk to a CPU cache, which leads to very high latency. There were many changes made to existing systems to minimize disk access, which in turn have minimized the number of pages loaded to the main memory when processing a query. The following diagram shows the evolution of memory bandwidth and CPU clock speed over the years:

With the advent of multicore CPUs and the declining cost of memory, computer architecture has changed in the recent past by hosting an entire database in the RAM.

In the current scenario, multicore CPUs (multiple CPUs on one chip or in one package) have become standard, which enables fast communication between processor cores. With the advent of these changes in the technology, main memory is not a limited resource now. There are servers that can have a system memory of up to 2 TB. This helps us to store an entire database in the RAM itself. The processors used in these servers have up to 64 cores and, in the near future, they are expected to get 128 cores. When the number of cores increases, CPUs can process huge amounts of data simultaneously. When this happens, a performance bottleneck is considered to have occurred between the CPU cache and main memory, and not for disk I/O operations.

In mainframes, transactional data and applications are stored on the same system. This is due to the fact that applications, operating systems, and the underlying databases share the same hardware resources. This means that we can't process transactions and reports concurrently. The problem here is cost. If we want to scale, we need another mainframe, which includes high cost. The cost of memory has come down drastically. This has brought on a revolution in increasing the size of memory. The following graph shows the fall of memory price over the years:

Source: http://www.jcmit.com/mem2013.htm

From the preceding graph, it is very clear that the cost of memory has come down tremendously and is predicted to go down further in the near future.

Present day applications running on traditional databases follow a 3-tier architecture. This is because databases are not capable of doing any calculations that involve complex logic or huge amounts of data; they are only capable of storing the data. There is a need for one more layer between the database and presentation layers—the application layer—to take care of all the calculations, based on which business logics from the base fields are implemented, as shown in the following diagram:

Let us look at each layer in detail: