When discussing server performance in what seems like an endless amount of applications today, the subject of server memory comes to the forefront for many IT professionals. With a multitude of server memory options available on the market, selecting the best solutions may seem like a daunting task. How much memory is enough? How do I plan for expansion/upgrade? Where do I begin sizing the correct solution? This article provides some insight as to understanding these options as well as what questions should be asked when trying to decide what technology and hardware are best suited for your needs.
What is server memory?
Memory vs Storage Server memory can be confused with the amount of storage space available on a device, which is a common misconception. Increasing the storage capacity of a server does not improve overall performance, but simply allows for more data to be saved. Instead, server memory allows for a server to write and read information being accessed from storage by the device. The more memory available, the more items/actions can be read, written, and performed by a server. Think of server memory as a person’s short-term memory and storage as long-term memory. Short-term allows focus on a current task at hand but is limited to how much work can be performed at that time. RAM (random access memory) is used by a server to store data that can be quickly accessed, read, and written by the CPU. It is a temporary workspace where instructions are performed along with data processing. When a user is performing tasks on a server such as accessing a program or browsing a web page, the RAM is executing these actions. This is the term most commonly used when referring to computer or server memory. Types of RAM Memory can be classified as volatile vs non-volatile where the former requires power to sustain stored data, but when power is lost the data is lost quickly or immediately. Non-volatile memory (NVM) is the opposite and can still access stored data after being powered off and on. Volatile memory has most practicality when used as main memory for a computing device and also consists of two types; dynamic and static. While dynamic RAM (DRAM) is the more popular of the two due to price point, static RAM (SRAM) operates at a higher speed since it does not require electrical refreshes like DRAM, but is a greater monetary cost. NVM is commonly used for secondary storage applications or long-term storage. DDR4, or double data rate 4th generation, is a form of DRAM that is widely used in cutting-edge server memory technology today due to its low power requirements and high data rate transfer speeds. Within the industry, RAM size is commonly measured in gigabytes (GB) available in terabyte capacity as well, while speed or access time is measured in megahertz (MHz). Today’s server memory is popularly categorized by its data transfer rate, or bandwidth, typically measured in millions of transfers per second (MT/s). A common form of RAM used today is dual inline memory module (DIMM) which generally provide 1066 MT/s, 1333 MT/s, 1600 MT/s, or 1866 MT/s. High-performing DIMMs can reach rates of 2133 and even 2400 MT/s.
How much server memory do you really need?
This is probably the first question to ask when determining the best server option for an application and unfortunately does not have a straightforward answer. Because of the versatile use of servers in modern day business, there are an overwhelming amount of solutions available. The relationship between how many users a server will support and the amount of memory required for sustainability is high. In situations where a server has users constantly surpassing the amount of memory capacity will encounter performance complications where said server will access virtual memory from drive storage and operates much slower. As always, research is key to determining what options work best for your application. Defining a number of users accessing the server and the types of programs and files that will be utilized and accessed is a significant starting point in order to best decide the base requirements for server memory. Plan ahead for future expansions that will occur as well in order to maintain a high-level of consistent operation as added users will reduce performance over time.
What are other companies buying?
As stated before, a baseline for server memory usage differs greatly from one application to the next. Along with the ever expanding technology developing day to day, an 8 GB DIMM was widely used in servers under 2 years ago. However, with the prices of larger capacity DIMMs decreasing over time, 1TB DIMMs are available allowing more flexibility to users along with productivity improvements and room for growth. EchoStreams' customizable server product solutions offer great performance and value for various user types and application sizes ranging from high-density servers for Cloud Computing solutions within the Scalestream server lineup. The OmniStream product line provides general purpose servers for adjustability and performance and DuraStream models are geared toward mission-critical applications. The Flachestream line provides the fastest and highest density rackmount servers available on the market boasting 20GB/s throughput.
Should you fill every slot in the server?
Depending on the usage requirements for a selected solution, this will vary. With the lower cost of high capacity memory options today, one would think not. However, there are many solutions where larger memory size provides competitive business advantages. Balancing performance and capacity is a continued struggle for many data center administrators as adding memory banks to a server will reduce the speed at which it operates. The takeaway here is, accounting for the memory you already have installed and understanding the demands of your specific use case will be important to consider before adding memory banks to your servers. Combining DIMM capacity sizes in a server is also possible, but it is best recommended to maintain similar sizes such as 8GB with 16GB. Using a DIMM larger than 16GB may not be supported and may not be recommended.