사용량 관리 및 최적화

Predicting and Managing Existing Workload Usage

Predicting and Managing Periodic Usage Increases

Generally, as cloud usage increases, costs increase as well.

Except in special economic situations, most businesses grow over time and data accumulates continuously, leading to increased demand for cloud resources.

If these resource requirements are not properly managed, costs are highly likely to increase uncontrollably.

Therefore, it is necessary to periodically forecast cloud usage and take appropriate actions accordingly.

Review Cycle Definition

First, define a review cycle for cloud resources, and allocate time and resources according to that cycle to perform cost optimization.

When setting the review cycle, you can refer to the discount commitment period of cloud resources.

For example, Samsung Cloud Platform offers 1-year or 3-year commitment discounts, Cost Savings, and Planned Compute.

Setting a resource adjustment cycle on a quarterly or semi-annual basis based on the expiration dates of these commitments is beneficial for establishing resource operational plans.

Additionally, aligning the review cycle with the company’s fiscal period enables effective use in departmental performance evaluations.

When setting the review cycle, you must also consider software license agreements, operation and management contract periods, and so on.

For example, you can define a review schedule where all resources are reviewed on a 12-month cycle and data stores on a 6-month cycle.

Usage Increase Prediction

The biggest factor affecting cloud costs is the increase in usage.

You must adjust resources in the current cloud environment to reflect the impact of increased usage.

You can perform predictions using the following procedure.

1. Evaluate whether all expense items are properly classified according to the organization’s departments and processes, and correct them if necessary.

   2. Calculate the average expenditure for the last 3 months by group.

2. Calculates average expenditure compared to the same period last year.

   4. Compare the two average values to identify trends and reflect the growth rate of each group.

3. Adjust resource quantity and pricing plans to reflect future plans.

   6. Share with the FinOps stakeholder and coordinate feedback.

During the fifth step of plan implementation, adjust the pricing tier and modify the quantity to reflect any newly increased or decreased resource usage.

The FinOps cost modeling related to this is covered in detail in III. FinOps Strategy Development and Execution 1.3 Cost Modeling.

Review of New Services, Features, and Configuration

Most cloud providers, including Samsung Cloud Platform, are continuously adding new technologies and services.

Some of these enable new business experiments, while others contribute to improving the performance of existing resources.

To keep workloads cost-effective, you must regularly review the possibility of introducing new services, features, and components.

Usage Forecasting and Management Based on External Influences

It is also important to analyze external factors affecting cloud usage.

To do this, understand the patterns and characteristics of computing workloads and define response time as a key performance metric to assess demand fluctuations.

Additionally, an analysis of the predictability, repeatability, rate of change, and scale of external influences is also required.

Set the analysis period to a sufficiently long duration (at least one year) to account for seasonality.

Through this analysis, you can adjust resources based on the predicted impact and evaluate the cost-effectiveness.

• Workload Type Identify the business type of the system under analysis. For example, e-commerce, internal business systems, and machine learning services have different performance requirements and resource characteristics, so identify the required resources through workload analysis.

• Utilization and Performance Metrics
  Analyze changes in resource usage and derive resource adjustment plans for maximum and minimum usage. At this point, we utilize performance metrics such as response time and latency.

• Request load type
  Analyze the traffic patterns of the request load to determine the direction of resource adjustment based on whether the workload is database transaction-centric or content delivery-centric.

You can perform the following tasks to make decisions regarding usage forecasting and resource adjustment.

• Gain insights into workloads using log files and monitoring data extracted from monitoring tools, including Cloud Monitoring. Obtain data on periodic changes and analyze trends in demand volatility and growth/decline patterns.

• Collaborate with departments that can affect demand to verify if an event has occurred.

Resource Optimization

Resource Type, Size, and Quantity Adjustment

The key to resource optimization is balancing the two goals of cost reduction and service stability.

Resizing operations to optimize the size and quantity of resources is not a simple technical adjustment, but requires data-driven strategic decision-making, and the items to review for this are as follows.

• Resource attributes and cost considerations By optimally selecting the type, size, and quantity of resources, you can meet technical requirements at minimum cost. When performing resizing for cost optimization, you must comprehensively consider not only all resources included in the workload and the attributes of each resource, but also the labor costs incurred for the adjustment tasks. If the labor costs incurred for resizing exceed the potential cost savings, it is advisable to perform this task as a one-time operation during service changes or termination rather than repeating it regularly. To resize resources, you must secure visibility into which resources are currently being used and how much. This visibility includes CPU usage, memory usage, network throughput, and disk usage. Based on this data, you can define the server type and disk capacity.

• Resizing for Resource Optimization Resizing should not be performed solely for cost reduction; care must be taken to ensure it does not negatively impact service operations. The primary goal of the operations team is to maintain stable operational capacity required for the service. In particular, resizing resources that support commercial applications involves high complexity due to licensing issues and requires a careful approach. Resources vary in cost, and focusing optimization efforts on high-cost resources can be more efficient. As mentioned previously, if the labor cost involved in resizing exceeds the savings in resource costs, the operation may be inefficient. It is advisable to decide whether to resize by considering the limit of potential savings, and focusing on high-spec Databases rather than low-spec Virtual Servers can yield greater cost savings.

• Data-driven Cost Optimization Strategy
  Data-based resizing can be used for capacity adjustment in a Scale-up approach, and Scale-out can also be considered. This approach is based on the concept of scaling computing Nodes and supports both manual and automatic scaling strategies. Examples of manual scaling include increasing the number of active Virtual Servers in a Load Balancer or adjusting the number of Node pools or pod replica sets in a Kubernetes Cluster. You can also automate these tasks by leveraging metrics such as average CPU usage. By specifying the minimum and maximum number of worker nodes, you can set the capacity range of computing resources required to process workloads, and achieve cost optimization by dynamically adjusting capacity based on metrics.

Disposal of idle resources

Among the tasks required for resource optimization in the cloud, managing idle resources is the most important and effective.

You may have idle resources that were created as needed but are no longer used over time, yet remain without being deleted or scaled down.

Addressing these idle resources is essential for reducing cloud costs and is the most effective way to achieve significant results.

• Idle Resource Management Procedure In a cloud environment, unused idle resources can occur over time. These resources should be deleted, but if not managed, they may be left unattended due to administrator negligence or a lack of resource management visibility. For resources not created directly by the administrator, it may be difficult to perform deletion operations, and they may contain important data or be preserved for a specific purpose. Therefore, to dispose of idle resources, a method to check resource usage history and disposal procedures are required.

• Idle resource lifecycle management
  First, you must manage tags related to the resource lifecycle. From a lifecycle perspective, the resource must identify whether it is for testing and when testing is completed. If a resource uses a license with an expiration period, the administrator must be able to identify that information. This information can be implemented through tags by establishing a tag policy related to the resource lifecycle to define the resource’s purpose, expiration period, and so on. After that, a process for resource disposal must be established, and all resources must include information about the relevant department and responsible personnel. Before discarding resources, notify stakeholders in advance and prevent data loss through a verification process. If resources are managed according to business importance, it is possible to implement automatic disposal or deletion automation for low-priority resources.