Capacity testing for cloud applications

The larger the project, the more it should be tested for capacity. In 2008, the Heathrow Airport in the United Kingdom opened the newly built terminal, that is, terminal 5th. The terminal project cost 4.3 billion pounds, but there is no bigger project than the project. The opening day of Terminal 5th, however, was not ideal, with a variety of conditions, including human and technical factors, that led to the cancellation of 34 flights that day, a hot topic in the next few days of British media speculation. The lack of capacity of baggage handling systems, pain waiting for passengers and other issues are the focus of media attention.

Can cloud computing's vast resources help us achieve a more accurate model of capacity requirements in the face of big projects like the 5th terminal at West Carolina airport? There are many reasons for the 34 flight cancellations and thousands of of passengers suffering anger, but the overall impact on the Heathrow airport is not large, which is its success. And can companies demand better capacity management tools in cloud computing?

Check Capacity

I want to know the size of the program before the application is actually publicly running. It's like testing before a plane is flying to avoid flying on the sky and suddenly breaking wings. So I'm going to figure out the following three questions in the capacity test:

System Properties: Before my application serves a customer, it should list all available resources and find the maximum system load. Generate a comprehensive load: simulate customer usage through an emulator to simulate the daily customer load on the system. Monitoring system changes:. Monitor system load changes when the customer volume of the application service increases.

In addition to testing system capacity, such tools can test other tasks such as stress testing, performance monitoring, and capacity management. Some companies use open-source tools such as Multi-mechanize, selenium, and jmeter to develop their own capacity-testing tools, while others directly from cloud testing tool vendors such as Soasta, Loadstorm and Cloudsleuth purchase the corresponding products. I used to use a lot of command-line tools, such as LSHW, Top, tcpdump, DF, and AB.

Check the entire technology stack

Any cloud application is built on multiple layers of technology.

At the top of the

are our custom business applications and purchased Office applications. Virtual machines and operating system support programs. The operating system and regulatory procedures of the cloud vendor running on the physical host. The lowest level of hardware and network equipment owned by a cloud vendor.

The levels that cloud vendors can control, such as hardware, networking, and hyper-regulatory programs, are hidden from users and cannot be measured by users. The strength of a chain depends on the weakest link in the chain, and in the cloud architecture, the chain is very long, such as fiber, PDU, router, switch, load Balancing, proxy server, and so on.

The AWS console can help us monitor CPU usage, disk reads and writes, network traffic, and so on, and these features are free. Although this free tool is not available for more advanced monitoring, the tool is sufficient for beginners. For more complex monitoring requirements, AWS provides cloudwatch services that can be implemented.

AWS Monitoring Window

The virtual machine layer of the technology stack includes disk, CPU, and memory. Through the operating system, these devices can be monitored, and the monitoring data can be obtained by the System Monitoring tool and analysis processing. The top layer, the application layer, is the same.

Problems in

cloud service capacity Measurement

I am sure that the cloud service provider will provide sufficient redundant capacity. However, I can not provide the specific data, because no vendor will be their own system architecture data to share the details. But I believe that in the next two years, this data will be gradually transparent, because there will be more and more cloud service providers, enterprises to the cloud platform will reach a high speed, service level will become a key factor. The price of future cloud services will decline, and the proportion of virtual machines to physical hosts will increase.

The method of measuring virtual machines may not be as accurate as the physical host testing of cloud vendors, as there are various regulatory mechanisms behind the virtual machines. These regulatory procedures will adjust the resource allocation of each virtual machine at any time according to demand. It's like the Air pipe Center faces a lot of airplanes and a limited number of runs to the job.

If my application is often idle and another call center service is quite busy, then CPU resources and network bandwidth resources will certainly not be assigned to me. If my storage needs are extremely low, and a user needs to store dozens of g of data, then the physical disk space is definitely more assigned to that user. And since the system clock is working in virtual time and not in real-time, the delay, error reading, or other minor problems caused by the regulatory process will affect my measurement of cloud services.