On cloud cost optimization practice

01 Necessity of cost control on cloud

As shown in the figure above, the data of Flexera Cloud Status Report in 2022 shows that the surveyed enterprises believe that 32% of their cloud expenditures are wasted, up from 30% last year.

According to the survey report on cloud MSP service development of China Academy of Information and Communication Research, cost optimization has become the primary demand of enterprise cloud management. On cloud cost control is a difficult problem and pain point faced by many enterprises.

The cost management systems on and off the cloud are very different. Under the cloud, enterprises purchase IT assets at one time and pay for them at one time. On the cloud, enterprises' IT assets become on-demand and pay as you go.

At present, enterprises face great challenges in cost control on the cloud. First of all, cost management on the cloud requires multi department collaboration. For example, the financial/procurement, technical/operation and maintenance, product/business teams cooperate. In the process of collaboration, all departments form a real-time decision-making system. Because cloud resource billing methods are diverse, enterprises need to have a deep understanding of cloud resource billing.

Secondly, enterprises need a timely cost reporting and monitoring system.

Finally, enterprises need to complete the adaptation of multi cloud scenarios. In the face of multi cloud scenarios, each cloud has different billing methods. Enterprises need to make targeted adaptation to control their own costs.

The enterprise's cloud cost control method is mainly divided into four modules. First, choose an appropriate payment method; Second, choose appropriate resource specifications; Third, improve resource utilization; Fourth, cost analysis and monitoring.

02 Payment method and resource specification selection

Take Alibaba Cloud ECS as an example. ECS has three main product forms: pay as you go, pay as you go, and preemptive instances.

These three product forms correspond to different product functions, and the essence is the balance of economy, flexibility and certainty.

Pay as you go flexibility is very high. You can create, release, upgrade and downgrade configurations at any time without any restrictions. But its disadvantage is that it is expensive and less economical.

The yearly and monthly package is a prepayment mechanism, which is economical and cheap. However, its flexibility is poor, and resources are tied to finance. The economy of preemptive instances is very good, the price is very cheap, but the certainty is poor.

Preemptive instances have two characteristics.

First, it is cheaper than pay as you go prices, which can be as low as 10%.

Second, the certainty is poor. It may be released by the system at any time after stable operation for an hour. The use of preemptive instances is suitable for stateless task-based scenarios and can greatly reduce costs.

Next, let's introduce the reserved sample coupons. The yearly and monthly resources are bound with the finance, and there are many restrictions on refunds and discounts. Pay as you go ECS plus reserved instance coupons, mainly to solve the flexibility of monthly package.

When the ECS instance type matches the reserved instance coupon, the ECS that pays as you go will not be billed, but will only charge the fee for the reserved instance coupon.

Because the reserved instance coupon is a concept of prepaid or locked duration, it is cheaper than pay as you go and greatly reduces costs. Reserved coupons include zero prepayment, partial prepayment and full prepayment.

The pay as you go ECS can be billed hourly and released at any time, which is also a zero prepayment method. However, the zero prepayment of reserved instance coupons is different from this, which means that users cannot refund or unsubscribe at any time after purchasing a certain amount of time. When the user promises to use it for one year and pay every hour, it must be used for one year.

Partial prepayment means that the user pays a part in advance, and the system will deduct the remaining amount every hour. Full prepayment is the same as monthly package. All the money is paid off in a lump sum.

In order to solve the problem that the reserved instance coupons are not flexible enough, Alibaba Cloud has launched a savings plan. Compared with the reserved instance coupon, it can meet the resource purchase bill deduction requirements in DevOps, containerization, multi specification families, multi region deployment and other scenarios.

There are two types of savings plans, namely universal and ECS. The universal model has no restrictions and can directly deduct the ECS pay as you go bill. ECS models have a small number of restrictions, that is, the restrictions of the regional specification family. At the same time, the savings plan supports multiple products, such as ECS, ECI, RDS, and so on.

An e-commerce user has been using yearly and monthly packages. The demand for resources is unstable and the demand for computing power changes rapidly. There are hidden costs in lifting and refunding. The total price is stable and the total cost is reduced by 9% after the switch.

A certain online education user has different demands for resources in different periods of time. The user uses a batch of resources in one department during the day and another in the evening. Nearly half of the time spent on annual and monthly subscription resources is wasted. The savings plan is matched with pay as you go, cross specification family discount benefits are shared, and the total cost is reduced by 42% after the switch.

A game user has a high demand for flexibility, and the resource pool must be built on a pay as you go basis, resulting in a very high pay as you go cost. The user directly purchased the savings plan, without any transformation, and the total cost was directly reduced by 56%. The resource guarantee side is configured with capacity reservation, no cost increase, and the success rate of volume creation is 100%.

To sum up, it is recommended that users combine multiple payment types. Because of different payment methods, there are different use scenarios.

Preemptive instances support stateless, task-based business loads. Pay as you go instances support stateful, dynamic business loads. The yearly and monthly subscription instances, volume based instances and deduction products support stable business loads.

As shown in the figure above, the payment methods of computing resources are compared. Among them, the flexibility of computing resources mainly refers to whether resources can be created, released, and allocated at will, as well as the coupling degree between resources and finance.

As shown in the figure above, compare the payment methods of storage resources. Among them, pay as you go uninstallation and release are unlimited, and are applicable to the business of indefinite elasticity.

As shown in the figure above, the payment methods of network resources are compared. Among them, billing based on fixed bandwidth is applicable to more stable services, and billing based on traffic usage is applicable to sudden business scenarios, that is, occasionally the traffic is particularly large, while most scenarios are relatively small.

The shared traffic package is applicable to scenarios that have a certain ability to predict traffic. Otherwise, the traffic package will be wasted if it is purchased too much or not deducted in time.

The above figure shows the scenario classification of resource specification selection. Users can select according to their own scenarios. Only by selecting a specification suitable for your own business scenario can you obtain the optimal cost performance ratio.

For example, the burst performance instance t5/t6 of the shared class has low cost and is very suitable for lightweight web applications, development/test environments and other low performance load business scenarios. The price can reach 30% to 60% of the corresponding exclusive specifications.

For example, an e-commerce website chooses a computing instance (4vCPU) based on its own business characteristics, which reduces the cost by more than 20% compared with the general-purpose one.

In terms of resource specification selection, we recommend that you choose the latest generation. Because the latest generation means that the software or hardware of the cloud manufacturer has undergone technical upgrading and can enjoy the technical dividend of cloud computing, Alibaba Cloud will release an official announcement on July 6, 2022: the prices of C6/C7, G6/G7 and R6/R7 will be reduced by 9% to 19% in some regions.

03 Improve resource utilization

Improve resource utilization, mainly for deduction products. Because the deduction products do not match, and the utilization and coverage are insufficient. Therefore, it is difficult to reach 100% of the estimated capacity. Users need to pay attention to the utilization rate and coverage rate of deduction products.

If the capacity reaches 100%, some bills may not be deducted. Users need to purchase additional or re purchase reserved instance coupons or savings plan products.

In addition, elastic scaling can effectively improve resource utilization. Elastic expansion and contraction are divided into vertical elasticity and horizontal elasticity. Vertical elasticity refers to the increase or decrease of the number of instances. For example, 100 ECSs become 200 or 50. Horizontal elasticity refers to the increase or decrease of CPU memory, upgrade or downgrade of an ECS.

Both elastic scaling modes support timed mode, dynamic mode, dynamic prediction mode, health mode, manual mode, and multiple modes.

The figure above shows other cost reduction and efficiency enhancement methods. Preemptive instances can realize the stable delivery of computing power cluster with the help of elastic supply and elastic scaling products.

If the pay as you go instance is used for a long time, it can be converted into a monthly package, and you can choose to pay by week. You can also purchase discount products such as reserved instance coupons and savings plans to reduce costs. Turn on the shutdown saving mode, and realize periodic timing switch on and off with the help of the automatic operation and maintenance tool OOS product.

The subscription instance can be automatically renewed, and the renewal efficiency can be improved by using the unified expiration date. If it is no longer used, you can unsubscribe or pay as you go.

The bandwidth can be raised and lowered periodically with the aid of the automatic operation and maintenance tool OOS. Purchase shared bandwidth package and shared traffic package, and the bandwidth between multiple products can be reused and managed uniformly.

In addition, users can authorize the intelligent advisor product to regularly scan their own resources and give cost optimization suggestions. Cost analysis and optimization are carried out with the help of user centered cost analysis function.

04 Cost Management

From the perspective of financial personnel, cost management has four requirements.

First, clearly know how much each department within the enterprise spends each month.

Second, set a budget to manage the expenses of each department.

Third, we need tools to analyze from various dimensions&to perceive our own costs, to judge whether our own costs are reasonable and whether there is room for optimization.

Fourth, cost anomaly detection capability.

When creating resources, the system will automatically allocate resources to departments or teams using labels according to the allocation policy.

Thus we can clearly see how much money each team and department has consumed.

Budget management refers to setting budgets according to certain conditions. If the budget is exceeded, users can set alerts and send emails or SMS messages.

Cost analysis and optimization is to analyze whether the cost is reasonable from all dimensions. Users can filter and analyze according to label, product, region, available region, instance type and other conditions.

Cost anomaly detection uses artificial intelligence algorithms to identify abnormal cost fluctuations. Anomaly detection supports evaluation feedback. The user's feedback results will participate in the training algorithm. The more feedback, the more accurate, the more conducive to improving the detection accuracy.

Payment methods such as monthly subscription, pay as you go, preemptive instance, and reserved instance coupon are a balance of economy, certainty, and flexibility. Enterprises need to pay attention to the scenarios applicable to each product. Only by choosing what is suitable for you can you achieve cost savings.

Enterprises need to select appropriate instance types according to their own business scenarios. For example, the instance types of computing type, general type, and burst type.

Vertical and horizontal elastic scaling, downtime saving mode, automatic operation and maintenance tool OOS products, etc. can effectively improve the resource utilization of enterprises. Cost analysis and optimization mainly analyze whether the cost of the enterprise is reasonable from various dimensions.

05 Q&A link, user's question and answer

Q1 In order to prevent cost waste and effectively manage cloud services, what practices should enterprises follow to optimize cloud costs?

Answer: For example, regular audit and centralized billing. In addition, Alibaba Cloud has introduced some cost analysis and cost optimization functions to help enterprises reduce costs and increase efficiency.

Q2 How to estimate the capacity?

Answer: It is recommended to optimize while using. Users can use the cloud manufacturer's monitoring indicators or conduct real-time optimization based on their own monitoring.

Q3 How to allocate the cost of resources with the same tag?

Answer: The financial unit supports cost allocation. In the user center of Alibaba Cloud, the financial unit will introduce cost allocation.