More and more enterprises are beginning to realize the importance of distributed pressure measurement.
With the continuous development of the Internet industry, the system architecture becomes more complex and the business scenarios become more diversified, and the requirements for performance testing become higher and higher. The traditional pressure measurement method has been unable to meet the needs of business and technology development, distributed pressure measurement, is emerged in this context.
As early as around 2006, IT system stability became a challenge for then-centralized architectures. With the rapid rise of the Internet, the “Unix+ Minicomputer” architecture of the time was hit by a data explosion. In particular, key business systems such as online transactions, business analysis and databases entered THE TB or even PB level around 2010, resulting in a heavy burden on the traditional IT architecture and putting forward new requirements on the stability and scalability of IT systems. Since then, Alibaba has started to de-” IOE “transformation, using X86 servers and standard storage and network equipment, etc., to re-erect highly stable and scalable distributed IT system.
In the decade after 2010, Chinese Internet companies have entered the transformation and construction of distributed systems. In 2020, with the rise of new infrastructure, IT systems in telecommunications, finance, power, retail, health care, education, government agencies and other industries will be promoted based on cloud computing distributed evolution. From a centralized architecture to a distributed architecture, the stability of an IT system involves not only equipment room wiring, network communication, hardware deployment, application architecture, and data disaster recovery, but also the detailed control and guarantee of the platform itself, including capacity pressure measurement and evaluation, and full-link pressure measurement.
Entering 2021, with the great prosperity of enterprise Internet and industrial Internet, a new track has been opened for the stability of IT system based on distributed system, and distributed pressure measurement has also been put on the agenda.
How to achieve peak traffic with a smaller budget that specifies the current size of the business is a constant theme of technology.
On the basis of the last performance test, we will talk about the purpose of distributed pressure testing, the problems to be solved and how to organize distributed pressure testing.
What is distributed manometry?
To answer this question, we must first understand what distributed manometry is.
According to Baidu Baike, stress testing refers to actively generating traffic, thus causing computational pressure on the service, testing the performance and robustness of the service, etc.
According to the attention, it can be divided into distributed pressure measurement (client) and full link pressure measurement (server).
Distributed pressure measurement refers to the use of multiple machines to generate pressure to the target machine, simulating tens of thousands of users to access concurrently, extending on the basis of pressure measurement, focusing on the distribution and dispersion of the pressure producing end.
Starting from the pressure measurement itself, the purpose of pressure measurement can be divided into the following four kinds:
1. Optimization: find the shortcomings in the system and distributed system and optimize them;
2. Standard resource requirements: what is the critical value that the existing logic can provide normal service under the specified resource, and synchronously provide data support for subsequent resource expansion;
3. Traffic playback: The phenotypic form of existing services and resources for real traffic;
4. Business rehearsal: rehearse specific business to find and avoid problems in advance.
The full-link pressure test is generally applied to services with long service links by introducing related systems and simulating the real online hardware environment. It is mainly based on requests, simulates real request traffic, and conducts pressure test through scenarios such as traffic diversion. Problems such as data flow funnel model ratio, bottleneck services, high-frequency services, and high availability nodes of system services are found through full-link pressure measurement, which provides real data for reference for online service deployment.
The purpose is to investigate the actual carrying capacity of the core page and key business in the whole chain from the beginning of users’ access to the system to the completion of all business. Simulate the real world to know ahead of time. The best way to verify this is to allow the event to occur early, and a full-link pressure survey can detect the problem early.
What problem does distributed manometry solve?
With the basic concepts in mind, let’s look at some of the problems that distributed manometry can solve.
In simple terms, distributed manometry solves the following four problems:
1, the single pressure generation capacity is limited;
2, the flow pressure has the demand of geographical distribution;
3. Rich data indexes in the process of pressure measurement;
4. Summary and display of pressure measurement results.
However, there are some challenges in the exploration and application of distributed manometry.
For example, the dispatching problem of the compressor, on the one hand, the compressor may break down in the process, on the other hand, because of the different resource configuration of the compressor, the distribution of pressure is different, the real operation of the compressor needs to be monitored.
Another example is the scheduling problem of basic data. It is necessary to deal with the allocation and scheduling of basic data, scheduling between multiple data sources, scheduling between conflicting basic data and the preparation and warehousing of other relevant data. Any error in any link may affect the whole pressure measurement process.
How to organize distributed manometry?
So, what does a complete distributed manometry process look like?
In general, distributed manometry is divided into six steps:
1. Preparation: Prepare the pressure test environment, either a separate test environment or a formal environment, and determine the pressure test time;
2, determine the pressure curve: can be ladder type, linear rise type;
3. Determine the distribution of the compressor: make clear the demand of the flow source;
4, clear purpose: according to the purpose to determine transactions and interfaces;
5. Prepare basic data: prepare relevant data and plan data scheduling;
6. Summary of process monitoring results: monitoring and alarm during the process, data analysis after pressure measurement, combined with full-link monitoring, such as Bonree Net, Bonree Server and other basic monitoring products, accurately locate the performance bottleneck.
It should be noted that in the process of organizing distributed pressure measurement, it is necessary to check whether the traffic set at the sending end is sent to the target server. If the service architecture is complex, traffic may be lost due to other factors. In addition, the resource usage of the source voltage may be insufficient, so you need to monitor the resource usage of the source voltage. In addition, you need to monitor all links to accurately locate performance bottlenecks.