The reference test environment used in this test is shown in Figure 5.
Load generation
The load generator for reference testing is Microsoft inetmonitor. The generator runs on many PCs. The reason why it is because it can accurately reflect the shopper's workload, easy to use, and simulate a large number of shoppers on each PC. Figure 6 shows the screen of INETMONITOR.
If your browser does not support embedded framework, click here to view in a separate page.
Figure 5 Environment for reference testing
If your browser does not support embedded framework, click here to view in a separate page.
Figure 6 INETMONITOR
Main result
The maximum number of shoppers can support each configuration while providing the appropriate response speed. The standard used is: on each server, the total usage rate of the system CPU is less than 80%, and the average response time of the transaction is less than 1 second.
Figures 7 and 8 show how the number of people supported simultaneously in the two platforms of Windows NT and Windows 2000 increases with the system upgrade and scale.
If your browser does not support embedded framework, click here to view in a separate page.
Figure 7 System Upgrade
If your browser does not support embedded framework, click here to view in a separate page.
Figure 8 Expanded scale
Detailed performance information
The following figure is a detailed information on system performance when performing a benchmark test. This graph is compared to the test data of Windows 2000 and Windows NT 4 on a single 4 web server.
In Figures 9 and 10, the response time and throughput increases with the increase in the number of shopping at the same time.
If your browser does not support embedded framework, if your browser does not support embedded framework, click here to view it in a separate page.
Figure 10 throughput
The CPU usage of the web server and the database server is also measured. All single tables and multiple web servers running on the Windows NT 4 and Windows 2000 platforms are less than one of the equivalent database servers. The CPU usage of the web server is shown in Figure 11.
If your browser does not support embedded framework, click here to view in a separate page.
Figure 11 WEB server CPU usage
Total shopping
Benchmark simulates a lot of shoppers. Every simulated shopper is submitted to the transaction, waiting for a reply, thinking for a while, submitting the next transaction, waiting for a reply, thinking in a while, so continuously until the benchmark test ends. Of course, the real shopper will make a lot of transactions in this way, but when they have completed the transaction, it will not return to the site for a long time, maybe for a few days or weeks. The total number of people who use the site is far greater than the number of shoppers that use the site at the same time.
It is difficult to estimate the total number of shops that a site can support, but a good approximate method is given below and has used it to analyze the event store.
The premise of this method is:
A shopman's average session includes 20 pages requests and lasts for about 20 minutes. Regular shoppers an average monthly access to the site once. The 20-minute reception of the 20-minute peak period is 4 times more than 20 minutes a week.
This means that 1 game is equivalent to approximately 500 regular shoppers.
The arguments about this are as follows:
The 20-minute period included in a month is approximately 2,000.
Each regular shopper visits a site once a month, so each shopper has access to the site within 20 minutes to 1/2000.
The shoppers recently received within 20 minutes of the peak period of 20 minutes, 1/500.
Since each visit is averaged for 20 minutes, it is equivalent to a 1/500 shopper; that is, every gamer is equivalent to 500 regular shoppers.
ICL solution capacity - 5 million assignee
Benchmark upgrade the web server to 3 4-way servers and connects to an 8-way database server. It is 10% smaller than the capacity of the database server, and there is no reason to not believe, at least 6 web servers can be connected to a database server, while the response time does not have significant increase. This allows the number of shopping people who support a shopping center to increase to a very large, see Figure 12. If your browser does not support embedded framework, click here to view in a separate page.
Figure 12 Scalability of the shopping center (Windows 2000)
The capacity of this benchmark test is as follows:
With 3 web servers, 5,400 bits can be supported at the same time, while the average response time is within 1 second. In addition, the test results show that by increasing the number of web servers to 6, the capacity of the shopping center will increase to 10,000 games. Under the same conditions as the load conditions simulated in the benchmark test, 5,400 games can carry about 7.5 million transactions per day, while 10,000 shoppers can perform approximately 1,4 million transactions per day. Suppose each at the same time is equivalent to 500 shoppers, then ICL's Internet shopping solutions can support the number of 27 million purchases, up to 5 million purchases. This performance has greatly exceed the future needs of most stores.
Microsoft Windows 2000
ISSK is very rapid and very easy to upgrade from Microsoft Windows NT 4 to Microsoft Windows 2000. Then, Microsoft uses the Internet shopping reference test in Leadows to verify the performance improvement of Windows 2000. Measurement shows that Windows 2000 Advanced Server Build 2195 has great improvements than Windows NT 4. Each Windows 2000 web server is 50% more than the number of shoppers supportable than Windows NT 4.
Enterprise shopping solution
The benchmark test mainly examines the scalability of the solution from the perspective of supporting more shoppers. It is also important, that is, this solution size management and system management (for example, the backup and update process of site content) must be able to handle a large number of sites and a large number of shoppers.
In order to establish a complete enterprise solution, it is also important to consider all other properties of the company, including:
Easy management availability reliability security
ICL has experienced a long but very successful development process in business server business, which uses knowledge accumulated in all of these issues to ensure that its Internet shopping solution meets all the requirements expected by the critical enterprise solutions.
in conclusion
This white paper has introduced the work made by ICL and Microsoft's scalability and performance of ICL's Internet shopping solutions. The joint test industry has proven to be very successful, and ICL and Microsoft have benefited from the knowledge and experience gained.
ICL's Internet shopping solutions and Microsoft products, which have been based have shown that they respond quickly and have high scalability on ICL's Windows NT servers. It is expected that ICLs can reproduce these results on other compatible Windows NT servers, but have not tested this.
Internet shopping is still in the early days of development, and it may increase in the next few years. ICL solutions can support the number of shoppers far beyond the future needs of most Internet stores.
Thank you
I am very grateful to all those involved in this benchmark: ICL's Stewart Sutcliffe, Gill Newbold, Andy Kirk, Mark Coleman, John Ruscoe, Stephen Hoyle, Dave Williams, Steve Picken and Colin Rutland, and Microsoft's Simon Davies, Robert Barnes, Sally Martin and Carl Carter-Schwendler. Glossary
ASP Active Server Pages COM Component Object Model (Component Object Model) ISSK Interactive Shopping Solution Kit HTML HTML (Hypertext Mark-up Language) IIS Internet Information Server MTS Microsoft Transaction Server NLBS Network Load Balancing Service (Network Load Balancing Service) SQL Structured Query Language SQL Server Microsoft Relational Database (Windows NT Load Balancing Service)
references
Picken, S., "Internet Shopping Services", ICL Systems Journal, Volume 14, Issue 1, Autumn, 1999.
About the Author
Stuart Forbes is a system designer of Manchester. He has been engaged in various development and customer projects, and performance research is his focus. Currently engaged in the design of Windows NT and Windows 2000 enterprise solutions.
STUART joined ICL in 1986, he is a member of IEE, but also an outstanding engineer of ICL.
Email address: stuart.Forbes@icl.com