AWS Amazon account purchase: Amazon EC2 general-purpose cloud server configuration recommendations

2026-05-27 阅读 17

In the cloud computing world of AWS, buying a server is actually very similar to buying a mobile phone.

Some people are developers of African sea games. They are struggling with graphics rendering and high-performance algorithms every day. They have to buy professional flagship machines with "partial performance" (such as computing optimized C series and GPU accelerated P series). However, the vast majority of enterprises and developers have to run nothing more

The company's official website, a standard API backend, a small and medium-sized ERP system, or a daily development and testing environment.

These businesses are characterized:

The CPU needs to be used a little, the memory also needs to be used a little, the network is not too jammed, everything needs to be, but nothing needs to be too extreme.

If you are in this scenario, then what you need most is AWS's famous "general-purpose server (General Purpose)"-that is, the T family and the M family. They are commonly known as the "bucket machine" in the cloud. The development of various indicators is balanced and not partial. They are the star products with the highest turnover rate in the whole network.

However, general-purpose servers are also divided into "lightweight cars" and "heavy trucks". Today, we will use the vernacular to talk thoroughly about how EC2 can be matched to save money, be silky and never step on the pit as an enterprise or individual developer in the face of daily common scenes.

1. core selection: T family vs M family, what is the essential difference?

T series (such as t3, t3a)

and

M series (such as m5, m6g)

. Selection, the core of the tangle is often between the two brothers.

In fact, to distinguish them, you only need to look at one key trait--

Is your traffic "smooth" or "sudden"?

1. T family (T3/T3a/T4g): "little money-saving experts" who can save some money"

Core Principle (Scalable Performance Example):T-series machines are very smart. When no one visits and the CPU is idle, it secretly "saves points" (CPU points). Once a wave of people suddenly comes to your website and the CPU usage soars instantly, it will consume these points, allowing the CPU to break through the original baseline and achieve a short-term "overclocking".

Big vernacular analogy: it is like a sprinter who likes to keep his energy and energy at ordinary times. He sleeps most of the time (saving electricity/saving money) and can sprint at critical moments.

Who is suitable for: development test environment, new official website, code warehouse (GitLab), and those internal office systems that only have a fixed number of times a day when someone logs in.

Avoidance warning: Don't put the heavy business of running data crazily for 24 hours on the T series. If the points are used up (and you don't turn on the Unlimited mode), the machine will be forced to "reduce the frequency and reduce the intelligence" and the website will be stuck instantly.

2. M family (M5/M6g/M7g): "evergreen tree" without emotion (full-function universal)

Core

Principle: The M series does not play points with you. How much calculation power do you buy, it will give you 100 percent of the supply in 24 hours. Its vCPU and memory ratio is always the classic 1:4 (such as 2-core 8G,4-core 16G).

Big vernacular analogy: It is like a calm pure fuel commercial vehicle, regardless of the road conditions, it can drive on the high speed at a constant speed, stable and lasting for a day and night.

Suitable for: Web backends in production environments, medium-sized e-commerce websites, distributed cache nodes, and core databases for small and medium-sized enterprises.

2. hand-in-hand with your configuration: three classic business scenario configuration recommendations

Below, we directly put the abstract concept to the ground. According to the actual combat experience of the old driver, for you to sort out the three most common general-purpose EC2 configuration "formula".

Scene 1: Individual Developer/Geek Toy/Start-up MVP (Minimalist Start)

Scene features: the budget is extremely limited, the number of users has not yet risen, mainly to deploy the code to run up to see the effect, or to make a personal blog.

Recommended models: t3.micro or t4g.micro(2 cores 1G)

Operating System: Amazon Linux 2023 or Ubuntu LTS

Old driver advice: If you want to be white, AWS provides a full year of free quota for newly registered users, including 750 hours of t2.micro or t3.micro per month. If the free period has passed, it is strongly recommended to choose t4g.micro with g suffix (using AWS self-developed Graviton chip, based on ARM architecture). Its price is nearly 20% cheaper than T3 of Intel chip, and its performance is better. It is currently the most cost-effective small mosquito meat in the cloud.

Scenario 2: Enterprise standard Web application/applet/API interface backend (main production environment)

Scene characteristics: there are tens of thousands to hundreds of thousands of visits every day, and the traffic peak is during the day. The system can not hang, requiring a stable response time.

Recommended models: m6g.xlarge(4-core 16G) or m5.xlarge

Storage configuration: select gp3 for EBS cloud disks, and the capacity is determined by the business. IOPS is 3000 by default.

Network configuration: Configure at least two application load balancers (ALB) to deploy across Availability Zones.

Old driver advice: don't blindly choose M5! As long as your code (such as Java, Python, Go, Node.js) can run under ARM architecture (now modern languages are basically perfectly supported), please close your eyes and choose M6g or M7g. The Graviton chip is Amazon's own son, and it's ridiculously cost-effective, allowing you to get a direct 20% discount on your server bill. Don't choose the old gp2 for the cloud tray, you must manually change it to gp3. GP3 not only

The unit price of GB is cheaper than gp2, and it allows you to get the basic performance of 3000 IOPS for free, and it is the conscience of the industry to upgrade the speed independently of the capacity.

Scenario 3: Small and medium-sized enterprise integrated database/internal ERP system (high reliability, large memory)

Scenario features: Although it is a general scenario, it requires high memory and storage throughput because it involves database read and write or multiple modules are concurrent.

Recommended model: m6gd.2xlarge(8-core 32G, with local solid state drive)

The old driver suggested: pay attention to the lowercase letter d in the code name. In addition to its general powerful performance, m6gd also gives you an additional NVMe local solid state drive (Instance Store) plugged into the host computer. You can throw the temporary exchange area (TempDB) and hotspot cache of the database directly into this local disk, which is more than an order of magnitude faster than the EBS cloud disk connected through the network. This is equivalent to the price of an ordinary server, Bai Piao has the performance of a "high-frequency small steel gun.

3. how to buy the most money? Three "wool" tips hidden in the console

The formula is selected, and if you click "Start" directly when you finally pay the bill, you will go with the most expensive "On-Demand". For general-purpose servers that need to be turned on for a long time, the following three tips can save you up to 50% or more:

Make good use of "Reserved Instance (RI)" or "Savings Plans". If your website plans to be open for at least one year, please purchase Savings Plans in the console. This is equivalent to signing a "long-term lease contract" with AWS, promising how much to spend per hour. In return, AWS will directly give you a 50% to 70% discount on your EC2 bill. The more machines are used, the more they are saved.

To develop the test machine, please configure the "automatic shutdown" programmer's development test machine, which is usually only used for 8 hours during the day and is spent in vain at night and on weekends. Use AWS Systems Manager or a simple Lambda script to configure a scheduled task: automatic shutdown (Stop) at 8 pm and automatic startup at 8:30 am every day. After such a month, the saved traffic and calculation costs can make you proud in front of your boss.

Don't configure the machine with too much physical memory. Novice configuration servers always like to pursue the "sense of security" displayed by the system. When they see that 70% of the memory is used, they are scared to upgrade the server. In fact, in the Linux system, memory is fully utilized is a healthy performance. Learn to look at the actual memory index (available memory) of the CloudWatch. As long as OOM (memory overflow error reporting) does not occur, keeping it at 60%-80% is the golden range with the highest cost performance.

Conclusion

Amazon EC2's general-purpose servers are like the most silent servers in the team.

Unheard but most indispensable "good old man". It does not have extremely exaggerated bandwidth, nor does it destroy the GPU computing power, but it uses the most solid ratio to silently support 80% of the daily operation on the Internet.

The essence of the selected configuration is not the pursuit of "the largest and most expensive", but "just right". According to the traffic curve of your business, make a choice between the flexibility of the T family and the stability of the M family, and then complete the final cost-effective closed loop through the Graviton chip (with G model) and gp3 cloud disk, and your cloud architecture has surpassed the vast majority of superficial people.