Aliyun Recharge on behalf of: Enterprise Station Construction Selection: Which ECS model should be equipped for websites with daily average PV 100000?
In the process of operation, enterprises often encounter such a selection problem:
"The average daily traffic (PV) of our website is expected to reach 100000. We are deployed on Alibaba Cloud. Which ECS instance model should we choose?"
As a "veteran" who has been engaged in SEO optimization and website architecture optimization for many years, I often encounter such consultation from CTO or head of operation and maintenance of enterprises. Many novices either blindly pile up hardware to cause the cost to soar when building the station and selecting the type, or they buy "sudden performance" instances with limited performance for a low price. As a result, the website crashes on the slightest traffic peak, which not only extremely damages the user experience, but also directly reduces the power by the search engine.
This article will be from
Traffic data disassembly, architecture evolution path, specific model configuration recommendation and cost optimization
Four dimensions, for everyone to sort out a set of clear and easy to understand selection scheme.
1. Disassembly Data: What Does 100000 PV Mean?
Before purchasing a cloud server, we can't just look at the total amount of "100000 PV", we must convert it into the core index of the bottom layer of the server--
QPS (concurrent requests per second)
and
bandwidth consumption
.
1. Calculate peak QPS
According to the "eighty-two rule" of Internet traffic distribution, 80% of traffic is usually concentrated in 20% of the day (about 4.8 hours, or 17,280 seconds).
Average QPS:$100,000 \div 86,400 \approx 1.16 \text {times/sec} $
Peak Average QPS:$(100,000 \times 0.8) \div (86,400 \times 0.2) \approx 4.63 \text {times/sec} $
Considering the sudden traffic (such as activity promotion, search engine crawler centralized crawling), we need to reserve 3 to 5 times the redundancy factor.
Conclusion: The bottom layer of the server needs to support the stable dynamic processing capacity of 20~50 QPS.
2. Calculate network bandwidth requirements
Suppose a typical enterprise official website or blog system, after image compression, Gzip/Brotli text compression and static resource optimization, the average page transmission size is
1.5MB
.
If all requests are handled by a single ECS instance, the peak throughput needs to reach $50 \times 1.5 \text{ MB} = 75 \text{ MB/s }$, which translates to 600 Mbps of network bandwidth.
Note: The cost of directly purchasing 600M public network exclusive bandwidth on Alibaba Cloud is extremely high. Therefore, in the actual enterprise station construction, you must cooperate with CDN or OSS buckets for dynamic and static separation. After static resources go to CDN, ECS itself only needs to undertake about 5% ~ 10% of dynamic data requests, and the actual bandwidth demand is only 5M ~ 10M at this time.
2. Alibaba Cloud E
CS Selection Core Principles and Pit Avoidance Guidelines
Before selecting specific ECS model specifications, specify three key rules:
Never buy "burst performance type (t5/t6/u1 and other partial quota models)" to run the core business: such models have CPU credit limit, once the CPU utilization rate exceeds the baseline, it will be forced to limit the frequency, which will easily lead to website response timeout during concurrent access.
For exclusive/enterprise (Enterprise Instance) instances, such as g7/g8i/g9a series (general-purpose) or c7/c8i/c9a series (computing). These models provide exclusive 100% vCPU computing power, strong network packet sending and receiving (PPS) capability, and extremely stable performance.
Recommended memory-to-CPU ratio: general-purpose (G series, CPU: memory = 1:4): suitable for database and application deployment on the same machine, or use Java/Python and other back-end languages with large memory consumption. Compute type (C series, CPU: memory = 1:2): Suitable for dynamic computing-intensive (such as Go, Node.js, PHP + Redis) scenarios.
Three practical deployment schemes 3. an average of 100000 PV per day
For different enterprise budgets and technical architectures, we can divide the configuration scheme into three stages:
Scheme A: Single-machine Agile Edition (suitable for saving money and going online quickly at the initial stage)
For phases with a limited budget and a simple site technology stack, such as WordPress, Typecho, Laravel, or lightweight Node.js.
Recommended models: Alibaba Cloud ECS general-purpose g8i or g9a(4-core 16G), or computing c8i/c9a(4-core 8G).
System disk: ESSD AutoPL cloud disk 100GB(IOPS elastic scaling to prevent database disk bottlenecks).
Bandwidth quota: The public network uses a fixed bandwidth of 5M to 10M (or uses traffic-based billing).
Core architecture: Nginx, PHP-FPM, Node.js, MySQL, and Redis are all deployed on a single ECS.
SEO optimization point: Alibaba Cloud DCDN must be configured on the front end of ECS to completely separate static resources such as images, CSS, JS, etc. ECS only responds to HTML and dynamic interface requests to ensure that the Core Web Vitals metric meets the standard.
Scheme B: dynamic and static separation and master-slave architecture (suitable for most mature enterprises, highly recommended)
When the average daily PV is firmly in the 100000 and includes frequent interactions such as user login, comment, and search, the stand-alone architecture is prone to cause the whole station to jam due to the high usage of MySQL. A split architecture is recommended at this point.
Node type
Recommended Configurations/Products
role and advantage
Web Application Server (ECS)
calculation type
c8i.xlarge(4-core 8G) or general-purpose g8i.xlarge(4-core 16G)
Pure running Web business logic and Nginx, can be expanded horizontally at any time when traffic pressure is high.
Cloud Database (RDS)
RDS MySQL Exclusive Kit (2-core 4G or 4-core 8G)
The database is separated from the computing power, with automatic backup and high availability of master and slave, greatly reducing the risk of locking tables.
Object Storage and CDN
OSS + DCDN combination
All media files are uploaded directly to OSS. After CDN global acceleration, ECS bandwidth pressure is saved by more than 90%.
Architectural advantages: The CPU utilization rate of Web servers can usually be stably controlled below 30%, even in the face of sudden traffic, and the average page loading time can be controlled within 500ms.
Scheme C: High Availability Cluster Edition (suitable for e-commerce and high concurrent business)
If your 100000 PV is concentrated in certain specific time periods (such as second kill, rush buying on the hour, regular push), or the business continuity requirements are extremely high:
Load balancing: Alibaba Cloud ALB / CLB (application-oriented load balancing) is used for traffic distribution.
Web computing nodes: 2 computing c7/c8i(2 cores and 4G) are deployed in different availability zones (to realize disaster recovery in the same city).
Database layer: RDS cloud database MySQL high availability edition + Redis community edition (cache hot data).
4. Operation and Maintenance Cost Reduction and Enterprise Account Procurement Guide
For enterprises, choosing the right model is only the first step. How to control the long-term procurement cost of cloud resources is also crucial:
Make good use of the year-to-month package and subscription model: for long-term core models such as production environment, try to choose one-year or three-year package, which can usually save 40% ~ 60% of the cost compared with paying by volume.
Pay attention to "Savings plans": for enterprises with multiple instances, you can enjoy discounts across models and regions by purchasing a general-purpose saving plan.
Compliance Channel Account Opening and Enterprise Preferences: When enterprises deploy or purchase multiple servers on a large scale on the cloud, it is recommended to purchase and authenticate Aliyun accounts through officially authorized high-quality service providers or exclusive enterprise channels. In this way, they can not only apply for exclusive large vouchers and new customer ladder discounts, but also obtain free technical selection guidance and operation and maintenance support at the architect level, further improving the overall ROI (return on investment).
5. Summary
In summary,
Average daily 100000 PV
Enterprise website:
The most worry-free stand-alone solution: directly select 4-core 16G (such as g8i/g9a series) to cooperate with DCDN acceleration;
The most robust industrial scheme: 4-core 8G ECS (application) +2-core 4G RDS (database) + OSS/CDN (static resources)
The separation architecture.
As long as you avoid the "burst" pit area with limited performance and do a good job of dynamic and static separation and database tuning, Aliyun's enterprise-level ECS instance can definitely provide your website with extremely smooth and stable access experience!