Analysis of Performance and Energy Efficiency server evaluation elements (1)

We cannot use concise words to summarize the performance of a computer. In fact, we must use other means for its embodiment. This is just like looking at whether a person's character is correct and having to observe his or her usual words and deeds. He wants to evaluate the performance of a computer, we must also evaluate the knowledge of running applications or computing tasks. There are usually two types of research: one is the execution time and the other is the traffic.

The execution time here refers to the time when a computer completes a computing task and can complete the specified task as quickly as possible. We say its performance is the highest. This method measures the performance based on execution time, which is more suitable for the environment where a single task is executed. Instead, we need to test a computer system that executes multiple tasks in parallel, such as a server, to test its traffic, that is to say, it is necessary to see how many computing tasks the system can accomplish per unit time. In fact, many managers with large-scale data centers are very concerned about improving the server's Work traffic.

How to reflect server performance

When we understand the embodiment of server performance, we also clarify its concept-it doesn't actually focus on functionality and quality as other products do, but is a reflection of technical capabilities, in addition, the level of such capabilities must be reflected and measured by applications.

It is no longer difficult to understand the performance of a computer. It is the ratio of system performance to power consumption, that is, the performance for power consumption per watt. However, this power consumption is not an instantaneous value, but the average power consumption when the system executes one or more tasks.

The concept of energy efficiency has been sought after by the industry and users in recent years, mainly because of the increasing environmental crisis caused by energy shortage and waste, this prompted people to begin to use the principle of "how much grass to eat, how much milk to squeeze", or even "Eat less grass, how much milk" to guide the development of computer technology. Intel's smart server processor, while significantly improving performance using the new microarchitecture, does not forget to import smart energy-saving technologies to precisely adjust its own power consumption based on application load, the goal of achieving the output performance as much energy as possible is actually influenced by this principle.

Here, we will repeat the performance and energy efficiency concepts of servers or computer systems, not only to help you review them, but to show that they are always in a dynamic state. At present, regardless of the minor performance differences between the same system and the same application, I believe that when many media friends evaluate the server, it is often encountered in some theoretical performance tests, but the performance is not satisfactory in close-to-practical tests, or does a system show different performance in different types of application tests? Such changes are indeed disturbing, but they are objectively and thoroughly analyzed to find out the factors that cause the changes and the solutions that can promote the healthy development of such changes, this is something that makes sense for end users to purchase and apply servers.

When talking about factors that affect the ability or competitiveness of a person or an organizational unit, people always love to reference the "Barrel Principle ", that is to say, the amount of water in a bucket is the physical phenomenon determined by the shortest piece of wood. It visually shows the shortcomings of an individual or institution in a certain link or aspect, this may drag your strengths down and bring down your overall level. In fact, this principle applies to all computer systems, including servers. In line with the principle of small opinion, we want to start with a processor to illustrate the impact of this principle on computer systems.

As we all know, before the advent of the multi-core era, the most important specification for processor selection is its clock frequency. After the popularization of multi-core processors, some vendors and users tend to determine their performance based on the number of inner cores. In fact, these two methods are biased. You still remember what we mentioned in the previous article-the performance of a computer system is reflected and measured by the time it takes to complete a specific program or computing task? Based on this, experts in the industry proposed a processor performance formula as early as the 1980s S, pointing out three factors that affect it, that is, the clock frequency, the number of executable commands per clock cycle (abbreviated as IPC), and application optimization.

The effect of clock frequency on processor performance does not need to be repeated. However, only when the IPC is the same, its speed and slowness can reflect the performance of the processor, however, when the processor microarchitecture is different and the IPC is significantly different, the comparison of clock frequency will inevitably go astray.