How to plan and implement resource capacity after an enterprise migrates to the cloud

1、 Why capacity planning is necessary

Enterprise Digital transformation, enterprise IT service cloud is making great strides in the development of the original, cloud or cloud enterprises, the conventional budget expenditure includes digital information or IT software service expenditure. The budget expenditure of this part includes the budget investment of cloud resources, and one of its accounting basis is cloud capacity planning and implementation.

In daily life, scenarios that require "capacity" planning are very common. For example, reservoir water storage is a typical dynamic "capacity" planning process that requires regulation of storage capacity based on the upstream and downstream water environment conditions. For example, during the epidemic period, scenic spots implemented the measure of tourists booking in advance and then purchasing tickets to enter the park. It is necessary to adjust the total number of daily tourists according to prevention and control requirements.

Similarly, business on the cloud will also develop and change dynamically, and the computing resources that cloud products and services rely on also need to be adjusted accordingly. We abstract the usage planning of computing resources into capacity planning.

The necessity of capacity planning for enterprises after going to the cloud is that their business is dynamically developing, and the cloud computing resources that the business relies on also need to be dynamically adjusted accordingly. Excessive computing resources lead to idle resources and cost waste, while insufficient computing resources affect the responsiveness of industry services and hinder the rapid development of businesses. So, what problems will arise if an enterprise goes to the cloud without capacity planning?

Firstly, there may be a mismatch between cost investment and business development. For example, when the business presents a rapid development trend and the demand for computing resources on which the business relies also shows an upward trend, if there is no capacity planning, it is likely that when the business outbreak period comes, the backend service capabilities will not be able to keep up in a timely manner, which will affect the continuous and stable development of the business, and even miss the golden opportunity for business development.

In addition, the application of internet technology has greatly narrowed the distance between service consumers and service providers, and the high availability and stability of service providers' performance across regions has become a normalized goal. One of the most direct implementation solutions for this goal is to achieve capacity redundancy between regions, so as to switch traffic and achieve disaster recovery in the event of software and hardware failures or other emergency scenarios.

In summary, after an enterprise goes to the cloud, capacity planning for its business is essential and requires continuous planning. Accurate capacity planning can help the rapid development of business, avoid computing power support becoming a bottleneck or obstacle to business development, and at the same time, ensure the high availability and stability of cross regional services for enterprise business.

2、 Transforming business requirements into capacity planning

Capacity planning serves the business, and capacity planning that is detached from the actual situation of the business is meaningless. Based on business characteristics and development stage goals, developing a capacity plan that matches business development is the only reasonable plan.

For example, for a company A, the business of department B requires one office computer per person. Currently, we are purchasing cloud desktop products from Alibaba Cloud. This year, it is expected that the number of employees in Department B will increase by 10%, so the capacity planning for the number of cloud desktops this year also needs to be expanded by 10%. This example is quite intuitive and easy to understand. In fact, there are many factors that need to be considered in cloud capacity planning for different industries and business characteristics. Following a general understanding, we will conduct a disassembly analysis, as shown in Figure 1, gradually breaking down from bottom to top.

Factor 1: Overall Development Assessment of Business Needs

The overall development trend and evaluation of enterprise business are the foundation of all demand sources. Without a sufficient evaluation of the overall development of the business, it is impossible to output reasonable and effective capacity planning and evaluation. For enterprises, they do not plan for capacity planning, as capacity planning serves business development. The overall development evaluation of the business is naturally at the bottom of the 'pyramid'.

Factor 2: Assessment of the development of the cloud native part of business requirements

The development evaluation of cloud native services is directly related to the proportion of cloud capacity planning budget, which corresponds to the development evaluation of the cloud native part at the bottom of the pyramid. For the internet industry, the main business entities may be cloud native; For traditional industries, if only the enterprise management informatization part is uploaded to the cloud, then the development evaluation of the cloud native part is a small proportion.

Factor 3: Demand assessment for cloud based priority support under limited budget

For enterprises, the budget for each item is always limited, and limited resource services should prioritize the development of key businesses to achieve maximum input-output ratio. For all cloud services, storage, database, and computing services are fundamental dependencies, and the planning and investment in these three areas are generally high priority guarantees.

Factor 4: Assessment of continuity requirements for the native part of the business cloud

For enterprises, business continuity is crucial at all stages of business development, especially the continuity of critical business services. Therefore, the capacity planning process needs to focus on and evaluate the reflection of business continuity in the budget. For example, the computing resources that core businesses rely on can be planned to achieve deterministic delivery of resources through instances with annual and monthly packages, flexible resource guarantee services, resource reservation services, etc., thus ensuring the continuity of services.

Factor 5: Assessment of regional disaster recovery needs for the native part of the business cloud

For enterprises, at different stages of development, the priority of business services in different regions may vary, so capacity planning needs to be geographical aware. Meanwhile, the high availability of services often relies on the construction of disaster tolerance capabilities between regions. So, the budget needs to balance the needs of regional development.

Factor 6: Independent Planning vs. Comprehensive Planning for the Native Part of Business Cloud Requirements

Based on the first five factors, capacity assessment is becoming increasingly specific. Starting from factor 6, planning needs to consider the impact of specific operational plans. Independent planning and comprehensive planning rely on different inputs and output schemes. For example, in the scenario of employee oriented work mentioned earlier, there is a demand for cloud desktops, as they are relatively independent of each other and can be planned and delivered independently.

For example, in the case of large-scale web services, because of the dependence on cloud database, cloud storage, traffic bandwidth and other services, the capacity assessment needs to be packaged and delivered as a whole to avoid the short board effect. And when evaluating the specific capacity, the evaluation tools and schemes relied on are also different. For independent planning, general evaluations are relatively easy to provide; For comprehensive planning, Alibaba Cloud's capacity planning service provides a complete set of solutions.

Factor 7: Evaluation of current discount information from different cloud service providers

After the subdivision of business capacity planning is in place and the products and tools relied on for capacity planning implementation are clearly defined, then the next step is to perceive discount information.

Different cloud service providers have related activities and discounts in different regions and computing products. Evaluating this part of the content can lead to purchasing more cost-effective computing resources with the same budget. For example, the SavingPlan+CapacityReservation service launched by Alibaba Cloud has achieved cost savings and deterministic delivery of resources.

Factor 8: Evaluation of planned capacity delivery schedule

The step of evaluating the capacity delivery schedule is to output specific planning information such as when, where, which computing resources will be delivered, and the corresponding budget. Early or too late delivery may not match business development, and even capacity planning may not be implemented in the end.

3、 Capacity planning is mapped to resource purchases

In the previous section, we provided a bottom-up description of the factors that need to be considered in capacity planning in a hierarchical manner. The essence of planning and evaluation is to meet the development needs of the business at the appropriate time and place, and to plan the corresponding computing power requirements at the corresponding time and place.

As shown in Figure 2, there are many methods for mapping specific requirements to computational power. The following assumption: The cloud service capabilities required for the future development of enterprise business are predictable, and based on predictable values, they are converted into specific resource instance purchase volume requirements, and then specific purchase plans are formed. The following is a common technical solution for mapping planning capacity to resource purchase volume.

Method 1: Linear mapping - Horizontal scaling

From a resource perspective, the classic evaluation method is: total resource instance quantity=total business request quantity QPS/QPS supported by a single resource instance. When business development requires more computing power, the total QPS will change. At this time, the number of resource instances that need to be added and expanded=newly added QPS/single machine QPS. This approach corresponds to what is referred to as "horizontal expansion" in the field of resource scheduling. Alibaba Cloud provides services such as Auto Scaling that support automatic horizontal scaling.

Method 2: Linear mapping - vertical scaling

From a resource perspective, vertical expansion is relative to horizontal expansion. By adjusting the computing power of a single machine resource, which means adjusting the size of QPS supported by a single machine (indirectly reducing the number of QPS supported by a single resource instance through resource instance reduction), the total number of resource instances is adjusted, thereby adjusting the total service request QPS. Generally, in scenarios of refined resource scheduling and mixed deployment of business loads, vertical scaling of single instances of resources is carried out.

There are two forms of vertical scaling: one is fixed (without changing the specifications after adjustment), for example, from the original 4VCPU to 2VCPU for vertical scaling. Then, horizontally expand and shrink the instance according to 2VCPU; Another type is non fixed (elastic scaling of a single computing resource in a short period of time), such as resource instances' limiting 'a certain dimension of resources during operation, thereby achieving adjustment of computing power for single instance resources in specific scenarios.

For the business side, the instance specifications seen have not changed. In a typical K8s resource model, such as CPU resource request, there are two parameters: request and limit, which can achieve elastic burst of CPU resources. For example, Alibaba Cloud's burst performance instance, which uses CPU points to ensure computational performance, is suitable for scenarios where CPU usage is low but occasionally high.

Method 3: Nonlinear Mapping - Full Link Evaluation

Large internet services, typically e-commerce trading systems, have multiple business scenarios, dependencies between businesses, and a large scale of business services. It is already difficult to evaluate the system capacity separately based on the application, and it is necessary to conduct an overall capacity evaluation under the pressure of the full link scenario.

Alibaba Cloud's capacity planning services provide a full range of services, including:

Service planning, providing business traffic analysis, data capacity analysis, message capacity analysis, database capacity analysis, and cluster capacity analysis;

Implement service planning and provide full link pressure testing solutions, scenario traffic matching and scheduling plans, current limiting and degradation plans, and exercise plans.

The core value of full link assessment: helping customers detect the optimal pressure, extreme pressure, and failure pressure points of cloud systems, and performing degradation and current limiting protection. Adopting full link assessment is particularly suitable for large-scale and complex scenario applications.

Method 4: Capacity prediction - automatic provisioning

Compared to methods 1, 2, 3, and 4, accurate evaluation of future capacity changes is not required before implementation. Based on system load balancing and QPS water level monitoring, automated resource delivery is carried out, including automatic horizontal expansion, scaling, and cross specification instance delivery. For example, Alibaba Cloud Elastic Container instance ECI supports multi specification instance delivery. Alibaba Cloud Operations Orchestration Service OOS provides cloud based automated operations services that can automate management and execution of tasks. Customers can define execution tasks, execution order, execution inputs and outputs through templates, and then complete the automated operation of tasks through execution templates. OOS supports cross product usage, and you can use OOS to manage cloud products such as ECS, RDS, SLB, VPC, etc.

In summary, the process from business requirements to resource capacity planning and then to resource capacity execution can be summarized as shown in Figure 3.

4、 Resource purchase volume landing and purchasing plan

After determining the amount of resource purchases, the specific purchasing plan for implementation is shown in Figure 4, which involves delivering computing resources with certainty along the business development timeline.

As introduced in the previous article "Guidelines for Selecting Virtual IDC (Private Pool) on Cloud in Three Typical Scenarios", business resource delivery includes daily stable resource requirements, daily elastic resource requirements, and sudden resource requirements. Enterprises need to choose appropriate resource selection plans based on their own business development characteristics and specific resource requirements, in order to achieve cost savings and resource certainty delivery. For example, for periodic resource demands, occasional resource demands, and resource demands during specific periods, flexible resource guarantee related products and services can be purchased. You can directly refer to this article for detailed capacity planning and the final landing and purchasing plan.

After enterprise business is launched into the cloud, resource capacity planning is required. Alibaba Cloud provides rich product capabilities to support precise evaluation of business capacity and flexible selection, especially based on resource guarantee services such as elastic guarantee and immediate effective capacity reservation. It supports deterministic delivery of resources and effectively ensures the continuity of business development.