A financial trading company processes significantly more trades per second, 40% more, after upgrading the RAM on their servers. Importantly, the same code ran on the same hardware- only the memory modules were upgraded.
The power to upgrade your computer: That additional RAM really does matter. While the world is distracted by the ever-rising clock speed and increasing number of processor cores, we’ve spent the last few pages exploring just what real limits there are to the power of the computers of the future. The answer isn’t as unexpected as you might hope: the real performance limiter is the memory – whether RAM or hard drive storage.
The Foundation of System Performance
RAM (Random Access Memory) is often considered the middle child of computers, serving as a bridge between storage drives and the computer’s processor. RAM is a intermediary step when the computer retrieves data from a hard drive or solid state drive to run applications. Just as the computer is faster at retrieving information from RAM than a hard drive or storage drive, it is also faster when preparing information in the RAM for use by the processor. Thus, maximizing the amount of RAM in your computer will allow it to run most applications more efficiently.
Server memory optimized for high performance is different than regular RAM in several ways. High-performance RAM supports faster data transfer rates than regular RAM which can double or triple the memory bandwidth of a server resulting in increased computer performance. RAM optimized for servers also features reduced latency, the amount of time it takes for the RAM to access data for the processor, which is critical during periods of peak server usage. High-performance RAM also often features advanced error correction to help guarantee the accuracy of data written to and read from the RAM.
Over time, these improvements can really add up, especially as you perform more system actions and run more memory intensive applications. To keep your system responsive and running, it’s very important not to waste memory unnecessarily.
Server Environments Demand Maximum Throughput
Systems such as enterprise servers are designed to handle large numbers of concurrent connections for web serving and other Internet-facing applications, as well as to support the load/unload of data for database systems built to handle real-time transaction streams. In these types of Server-grade systems, memory bandwidth becomes a bottleneck especially in a virtualized multi-tenant environment where one physical server can support twenty fully configured virtual machines, each consuming memory out of a shared resource pool.
To run multiple virtualized instances without a degradation in service, high performance capacity server RAM is required. To achieve several months of around the clock service without shutdown, increased reliability of the server RAM is also a requirement.
It only takes a single bit to go missing or get flipped incorrectly to cause an application to fail or sensitive information to be compromised. However, high-performance DDR3, DDR4 and DDR5 memory helps to mitigate this risk with included ECC bits and advanced thermal (Thermal Management) features that regular, entry-level memory does not possess.
Database Performance Dependencies
Of all system components, database systems are most affected by memory performance because the data that most queries access is typically cached in RAM for fastest processing. If a database system requires data residing on the storage drive to satisfy frequently accessed queries, query execution time can be significantly impacted.
Most modern database engines cache results of some queries, indexes and recently accessed data in RAM. Increasing the RAM on your system can then increase these caches and, as RAM is faster than disk, also improve overall system performance. For demanding applications, enterprise-grade modules like SK Hynix H5CG46AGBDX017N RAM provide the reliability and performance characteristics required for mission-critical operations.
Workstation Applications Push Memory to Limits
Unlike servers designed to handle vast numbers of connections, professional workstations are intended to complete computation-intensive tasks, which differently load on the memory subsystem. Video rendering loads a large texture file into memory at the same time, professional CAD software renders millions of polygons in 3D space, and scientific simulations produce large numbers of results that require increased storage capacity.
High performance memory drastically improves the computing power of content creation computers by duplicating and processing large files in video editing, simultaneously opening numerous video layers for color grading, and rapidly generating very large temporary files for effects work.
The number of memory access operations increases when moving from datacenter servers to workstations, due to the shift from handling many small requests to handle fewer large requests. While current servers for datacenters are optimized for handling small requests to serve web applications, servers for workstations require different memory access techniques. While Software applications already use specialized access patterns, such as X87 floating point instructions that load five operands, and indirect access to memory, other applications like database algorithms, and CAD/CAM systems require additional specialized memory access techniques to efficiently handle large amounts of data.
Data Center Infrastructure Requirements
The data center is the most demanding memory performance environment. The rapid adoption of cloud computing and big data platforms that provision thousands of virtual instances or process petabytes of data drives the need for massive memory capacity and performance.
Memory performance at scale in the data center is critical to data center operational cost by enabling fewer servers to be required to process the same amount of work, higher server utilization, and lower latency.
While sought after in the data center is higher memory performance, there are still other critical parameters at play. Power, for starters. Measured in watts per module, then multiplied by thousands of actual modules in the rack, it plays a critical role. Also important are the thermal characteristics of said memory as it directly affects the space between servers in inches.
In order to address increased demands on computer memory, memory manufacturers are working hard to increase memory density by ramping up memory capacity in existing architectures as well as by developing new memory technologies with increased bandwidth and lower access latency. Simultaneously, the industry is moving to increasingly advanced manufacturing technologies in order to create higher-density devices at lower cost per byte.
When planning Server clusters / data centre infrastructure, designing professional workstations or upgrading an existing platform, it is helpful to have an appreciation of the memory requirements for your specific application. At MemTech we can determine the exact specifications required to support your organisation’s objectives.
