Evaluating Suitable Cloud Services

When designing high-performance architectures, the first priority is to focus on the selection of fundamental cloud services such as computing, storage, and networking. It is crucial to have a thorough understanding of the unique characteristics and metrics of your business and become familiar with the types of cloud product solutions associated with the main components of your architecture. This enables your business and workloads to benefit from different product offerings and facilitates continuous optimization of the system architecture.

The selection of the database is a highly professional work. There are many factors considered apart from performance. Here, we wouldn't elaborate on this. It is important to note that the reasonable use of database product combinations in high-performance architecture design is needed, as many business scenarios have significantly improved performance by using cache databases like Redis.

Computation

Computation services

Being familiar with the main computing schemes and, for each business system and workload, evaluating the use of appropriate computing schemes are necessary.

Computation indicators

When it comes to computational performance, people usually use specific indicators to describe and quantify it, such as a disk satisfying 100,000 IOPS capabilities, a single server's network bandwidth of 10Gbps, etc. Fully understanding the indicators of each computing scheme and relating the business system indicators identified and collected to these is key to understanding the business bottlenecks and resource usage rates in depth. Using cloud monitoring can easily track and maintain related indicators continuously to drive continuous performance and architectural optimization.

Computation specifications

For non-Serverless computational services, the selection of types and specifications is particularly important. Making the wrong choice can lead to frequent performance bottlenecks and unnecessary costs as well as increase the complexity of subsequent optimization and transformation efforts. By combining tracking and analysis of metrics, it is important to consider the correlation between workload and the utilization of CPU, memory, disk, and network resources. This helps in selecting the appropriate type and the ratio of memory to CPU cores, ensuring optimal performance while optimizing cost-effectiveness.

For example, for I/O-intensive workloads such as medium-to-large-scale OLTP databases, medium-to-large-scale NoSQL databases, real-time log analysis, and large-scale enterprise-level commercial software, it is advisable to choose storage-optimized instances to achieve better storage performance and capacity. For network-intensive workloads such as NFV/SD-WAN, audio/video processing, telecommunications service forwarding, caching, big data, and machine learning, it is recommended to select network-optimized instances that can meet the higher requirements for bandwidth, packets per second (PPS), concurrent connections, and multiple network interfaces. fFor workloads that can benefit from GPU/FPGA acceleration, it is important to choose the appropriate heterogeneous computing specifications that can leverage these accelerators.

Storage

Storage Services

Compared to traditional on-premises data centers, the cloud offers a wider range of storage business scenarios and corresponding storage services. To fully leverage the advantages of the cloud, it is important to adapt storage solutions flexibly to match specific business needs. Relying solely on a single storage type often fails to meet the optimal performance and efficiency requirements. Therefore, during the architecture design process, it is necessary to analyze the business scenarios and corresponding workload storage requirements, and clearly define the core storage metrics. This enables the evaluation of suitable solutions that meet the specific needs and requirements.

Generally, storage schemes will focus on block storage, object storage and file storage. Usually, the access mode and access link are determined first according to the business scenario. For example, if an internet application that needs to support public network access based on the object API is needed, OSS would be the best choice, while if it is an application deployed in a container, in order to achieve shared read and write access in professional network environments and fully flexible persistency of storage, file storage is more advantageous.

Indeed, Alibaba Cloud provides a wide range of storage services that offer additional functionality, compatibility, and high availability architecture, Alibaba Cloud provides many storage service choices. For example, Tablestore fot the huge amount of structured data, the CSG storage gateway compatible with multiple storage services, the data migration service supporting online and offline migration, HDR services supporting hybrid cloud disaster recovery capabilities, HBR services for hybrid cloud backup, etc.

Storage Indicators

On the basis of business scenarios, functions, and architecture, the business system data must also be converted into storage performance indicators to further evaluate the selection of storage products and types. The main storage indicators include throughput, IOPS, I/O latency, access frequency, data scale, data growth rate, data reliability, etc. Necessary benchmark testing and continuous performance data collection will help obtain and analyze these indicators.

In terms of products, Alibaba Cloud also provides different types of services to meet different performance requirements, such as block storage includes general cloud disks, efficient cloud disks, SSD cloud disks, and various levels of ESSD cloud disks, object storage includes different levels of archive storage, different levels of low-frequency access and regular storage, etc., flexibly addressing storage demands of various business scenarios.

Networking

Networking Services

Networking is critical in a cloud environment including connecting cloud services, local data centers, application program components, etc. It significantly impacts the performance of business workloads. Selecting the appropriate combination of networking products often efficiently achieves performance optimization without changing the business logic.

The above lists some typical networking products. Using them, rapid construction of high-performance hybrid cloud networking, worldwide nearby access acceleration business, cross-availability zone high availability, and public and private address conversion etc., can be quickly implemented. Additionally, they can also be coupled with products such as CDN to amplify the concurrency performance in front of the computational resources.

Networking Indicators

Compared to computing and storage, networking performance is restricted by more factors, and it typically needs to focus on indicators such as latency, packet loss rate, bandwidth, throughput, requests per second, number of concurrent connections, number of new connections, etc. during architectural design. Analysis of networking performance means the analysis of the entire request link. Stress testing and continuous monitoring will help in clarifying the core indicators of workload and locating link bottlenecks and further realize dynamic specifications optimization and scheme optimization.