Stuart Forbes, ICL, British Manchester Summary With the growth of e-commerce, companies need telescopic, high performance, able to handle heavy trading volumes and large users' Internet solutions. Many demonstrations indicate that these solutions have high performance, but they are usually not based on a real-world solution. ICL and Microsoft combine Betmark in Manchester and Redmond to test the scalability of interactive shopping solutions based on Microsoft BackOffice applications. Created a large Internet shopping center and simulated the number of shoppers. The test results are as follows:
On the Windows 2000 web server, the number of supported shoppers is 50% higher than in the Windows NT 4 web server. The mall supports 5,400 shoppers at the same time, and the average response time is maintained within 1 second. In addition, the test results show that the shopping mall also supports the shoppers to increase to 10,000. Under the same conditions as the load conditions simulated in the reference test, 5,400 games at the same time, the number of transactions per day, which is expected to be per day, less than 7.5 million, and 10,000 shoppers can perform more than 4 million transactions per day. According to the hypothesis and the frequency of the presentation and the frequent extent of the shopping center, the performance test results show that ICL's Internet shopping solutions can support at least 2.7 million purchases, up to 5 million purchases. This performance is much higher than the performance required to most stores. Introduction ICL can provide operational management of Internet-based shopping ICL's Interactive Shopping Solution Kit solutions, Interactive Shopping Solution Kit for Microsoft's Windows NT 4 Server and Windows 2000 Advanced BackOffice applications on top of Server, especially the Internet Information Services, Site Server Commerce Edition and SQL Server. This white paper describes the combined benchmark tests between ICL and Microsoft to study the scalability of ICL Internet shopping solutions. The purpose of this test is:
Study the scalability, limitations and strength of ICL shopping solutions. Provide scale information to make the size of large shopping sites meet cost-effectiveness, thereby reducing the cost of the ICL solution while maintaining confidence of shopping loads, to demonstrate their products in Windows NT 4 and Windows. The scalable Windows NT 4 benchmark is conducted in the laboratory of ICL companies located in Manchester, England, and the Windows 2000 benchmark is initially carried out in the Microsoft scalability laboratory in Redmond, USA. Manchester conducted. Betmark Configuration Use ICL Enterprise TrimeTra P2000 Server (4 and 8 Pentium III Xeon 550 MHz Processors). As the reference test results discussed later in this white paper, the ICL's Internet shopping solutions can provide features and scalability than most stores need more than any time. Interactive Shopping Solution Kit Interactive Shopping Solution Kit (ISSK) is used to develop ICL Internet shopping solutions. It provides a set of components, and ICL uses these components to generate actual customer Web sites under reducing costs and reducing the time to market. You can use the Solution Toolkit to establish a standpoint for hosting a single store, one or more shopping centers. The toolkit allows the content of the site to manage the site, and manages the site itself. This is derived from experience accumulated in many customers and has been applied to many important customers (such as BBC SHOP). ISSK is based on IIS, Site Server Commerce Edition and SQL Server. It uses the HTML web template to provide site's appearance and feel, using the ASP page to provide business logic. It uses SQL Server to save your product catalog, site structure, promotional items, shopping baskets, and orders and receipts. You can use the stored database procedure to access SQL Server to improve performance. A small amount of COM objects and MTS are also used, for example, for combining templates and processing orders. Site Server is used for search, personalization, analysis, advertising, and business. The basic structure of this solution is shown in Figure 1, but more complete descriptions [Picken, 1999]. If your browser does not support embedded framework, click here to view in a separate page. Figure 1 ISSK structure ISSK also includes a stand-alone management server management site and an upgrade site content. The benchmark test mainly focuses on providing a large number of workload shopping. Initially, the reference test is based on ISSK 1. However, ISSK 2 is already available at the end of the test, all results in this white paper are based on this new version. ISSK 2 has made many improvements in performance, some of which have been determined by the benchmark team, and some are planned to improve. As a result, the number of shoppers supported by ISSK 2 is more than twice ask 1, and the response speed is faster. Such a solution is a solution: Once a hardware resource is added, the ability to complete the effective work has also increased. In fact, it is almost impossible to achieve this ideal; more practical scalable solutions are: After the hardware resources are added, the ability to complete the effective job is expected, and the economy is reasonably increased. The basic hardware configuration of the ISSK solution is shown in Figure 2. Figure 2 Basic configuration can expand hardware by two ways: Upgrade system: Increase the capacity of each server to expand the scale: Increase the number of servers, in progress, the two methods have been studied. The system is upgraded by adding the CPU of each web server from 1 to 4.
By adding the number of web servers from 1 to 3, and uses Windows NT Load Balancing Service (WLBS) to balance the load between the server, the system scale is expanded, and WLBs can support up to 32 in a cluster. Table server. As shown in Figure 3. The system capacity can be further expanded by dividing multiple database servers and partitioning in multiple servers. Note: In Windows 2000, WLBS is called Network Load Balancing Service, NLBS. In the benchmark test process, each configuration has been tested multiple times, and each time it adds to the number of simulated shoppers accessing the shopping center. Record the number of peaks that can be handled while the system can be processed and the response time can be accepted. However, it is important that the peak request for the Internet shopping solution exceeds its set value. Therefore, the reference test activity also detects the health of the number of people when the number of shoppers is much greater than the number of responding times that the system can provide satisfactory response time to all shoppers, to check if the solution is strong, and the throughput does not decline. Figure 3 Upgrade system and expanded Scale BBC SHOP This reference test is based on BBC SHOP, a sample (http://www.bbcshop.com) that successfully deploys Interactive Shopping Solution Kit (http://www.bbcshop.com). Figure 4 shows its home page. If your browser does not support embedded framework, click here to view in a separate page. Figure 4 BBC SHOP reference test benchmark is a model with a shopping site, which consists of a shopping center with two stores. These two stores are based on the above BBC SHOP site. Benchmark supports a large number of different shopper types, which are obtained by analyzing the actual Internet store as shown in Table 1. Table 1 Shopper Type Better Test Shopper Type The Merger Percentage of Shoppers The number of transactions Note The first viewer 20% 18 Shoppers View the editing page, product summary and product details, but not for any search or No items are added to the shopping basket. The second browser 20% 12 is similar to the first viewer, but the page browsing is less. Adder 20% 24 Shoppers browsing the website and adding items in the shopping basket, but did not buy it. Buyers 20% 32 Shoppers browse this site, adding a commodity in the shopping basket and purchasing it. Search by 20% 14 shoppers have conducted a search and browsed the product details returned. No items are added to the shopping basket. The average 20 simulation environments are described in Table 2 and Table 3, respectively. Table 1 Shopper Type Better Test
Shopper types account for a percentage of shoppers, the number of transactions, indicating that the first viewer 20% 18 Shoppers View the editing page, product summary and product details, but no search or not add anything in the shopping basket. project. The second browser 20% 12 is similar to the first viewer, but the page browsing is less. Adder 20% 24 Shoppers browsing the website and adding items in the shopping basket, but did not buy it. Buyers 20% 32 Shoppers browse this site, adding a commodity in the shopping basket and purchasing it. Search by 20% 14 shoppers have conducted a search and browsed the product details returned. No items are added to the shopping basket. Average 20 table 2 for a benchmark test hardware
Server Number Description Web Server 3 ICl Trimetra P2000 PIII 550 MHz, 1 MB Cache, 1 GB RAM [1 ~ 4 CPU] Database Server 1 ICl Trimetra P2000 PIII 550 MHz, 1 MB Cache, 2 GB RAM, 8 CPUs Load Builder 2 FJ Ergopro PII 266 MHz, 64 MB RAM Monitoring Workstation 1 FJ Ergopro PII 266 MHz, 192 MB RAM Network 1 100 MB Switch Table 3 Performing Better Test Software Software Windows NT 4 Test Windows 2000 Test Web Server ICL Interactive Shopping ICL Interactive Shopping Solution Kit 2.0 Solution Kit 2.0 Microsoft Windows Microsoft Windows 2000 Advanced NT 4 Server SP5 Server Build 2195 Microsoft Site Server Microsoft Site Server Commerce Edition 3.0 SP2 Commerce Edition 3.0 SP3 Microsoft Data Access Components Microsoft Data Access Components 2.1 SP2 2.1 SP2 Microsoft Internet Information Microsoft Internet Information server 4.0 server 5.0 Microsoft Transaction server Microsoft Transaction server database server ICL Interactive Shopping ICL Interactive Shopping Solution Kit 2.0 Solution Kit 2.0 Microsoft Windows Microsoft Windows 2000 Advanced NT 4 server SP5 server Build 2195 Microsoft Data Acce SS Components Microsoft Data Access Components 2.1 SP2 Microsoft SQL Server 7 SP1 Microsoft SQL Server 7 SP1 Load Generator INETMONITOR 3.0 For more information on ICL Trime P2000 servers, you can get from ICL's Web site, especially the following sites: http: / / www.icl.com/itservices/servers/servers.htm. The reference test environment used in this test is shown in Figure 5.