Dr. An’s Lecture | Tencent Arm Cloud Instance Performance Evaluation – Web Applications

Web services are web applications based on open standards (XML, SOAP, HTTP, etc.), which can be used independently of the hardware or software platform that implements them, as well as independently of the programming language in which they are written. This independence determines that applications based on web services are loosely coupled, component-oriented, and implemented using containerized microservices.

With the development of cloud computing and the elastic advantages of cloud servers, web applications have evolved into a more advanced form of cloud-based applications that have become mainstream. As “evolved” web applications, cloud-based applications typically have advanced features, access to a wider range of services, and provide dynamic content based on business logic to their customers. Typical examples of such web applications today include social applications, interactive media platforms, e-commerce websites, online digital services, and IoT systems.

In cloud applications, the “server” side is the cloud, which is based on a microservices architecture cloud-native model, where a single service is composed of many loosely coupled and independently deployable smaller components or services. The microservices architecture decomposes traditional monolithic applications into their component functions, and application functions are no longer deployed in a single code source. Each function is isolated as a small, self-running application (i.e., microservice), which typically runs in a containerized environment for lighter deployment and execution.

The entire stack can logically be divided into three layers: the frontend routes user requests to the backend via load balancer servers, followed by a loosely coupled modular business logic layer, and finally the data caching and storage layer.

To comprehensively understand the overall performance of the new architecture instance SR1 under web applications, this article will conduct tests on two different size instances, with the specific instance configurations as follows:

NGINX is an open-source high-performance HTTP server and reverse proxy that bundles many other web service-related features. NGINX is commonly used as a load balancer, reverse proxy, and HTTP server in many popular high-traffic websites. It uses a complex event-driven architecture that allows it to support hundreds of thousands of concurrent connections. The performance of NGINX can reflect the business processing capacity and user concurrency capacity of web applications. The following figure shows the test configuration under the testing instance.

Redis is an open-source in-memory key-value data structure store, used as a database, cache, message broker, and stream engine. Redis is a mainstream option in many web service infrastructures. According to the monthly ranking of DB-Engines.com, Redis is also the most popular key-value store system.

MySQL is a relational database management system and one of the most popular relational database management systems. In terms of web applications, MySQL is the best choice for RDBMS (Relational Database Management System) application software.

To make the data more intuitive, this article will use the S6 cloud instance based on the Intel IceLake CPU as a comparison benchmark, which is 100%, with SR1 instance data taken as a relative proportion.

First, we look at the data of NGINX on the 8 vCPU instance. In this test, the number of HTTPS requests per second under a 5ms P99 latency SLA is used as the performance indicator. In terms of performance, the 8-core SR1 instance based on the Ampere Altra CPU is 35% higher than the S6; considering the price factor, the one-year billing price shows that the cost-performance of SR1 is 77% higher than that of the S6 instance.

Next, we look at the data of Redis . Similar to the comparison method with NGINX, the performance indicator in the test is the number of requests processed per second under a 1ms P99 latency SLA. In the 8 vCPU instance, SR1 is 14% higher than the baseline S6 instance. Similarly, considering the price factor, the cost-performance of SR1 in Redis is over 50% higher than that of the baseline S6.

In the 16 vCPU instance, the performance of SR1 is 8% higher than S6, and the cost-performance improvement is 43%.

Finally, let’s look at the data for MySQL . In this test, the performance is measured by the mysql oltp_read_write load, which has higher requirements for computing power and storage, with the number of queries per second under a 100ms P95 latency SLA as the performance indicator. Considering the price factor, the cost-performance of SR1 maintains an advantage of 46% (for the 8 vCPU instance) and 51% (for the 16 vCPU instance) over the S6 instance.

Based on the testing data of the three key workloads in web applications, it can be concluded that the cost-performance of web applications based on the SR1 instance is, on average, over 50% higher than that of the S6 instance. This means that software service providers for web applications can provide over 1.5 times the business processing capacity at the same cost, or use fewer cloud servers while maintaining the current business volume, saving over 33% in absolute costs.

Finally, deploying web applications on ARM architecture is nearly identical to the x86 platform. The three key workloads tested performed very smoothly on the SR1. NGINX can be run directly with Yum installation, and Redis and MySQL can also be installed via Yum, but for performance considerations in this evaluation, the source code was recompiled without any source code modifications. The source code was downloaded via wget, and after configuring the correct compilation parameters, it was compiled.

Ampere Computing’s Free Trial Program has opened an application channel for developers to apply for SR1 instances. Interested individuals can try SR1 instances for free through this program and personally experience the new architecture computing platform.

Additionally, developers can also obtain a large number of directly usable, compatible image information compiled on the ARM architecture, including development languages, applications, and tool software packages, from the Ampere Solutions Website, covering most of the resources required for cloud-native development.