InformationWeek
Defining The Business Value Of Technology
|
September 25, 2000 |
 |
Web-Site Scalability
Business Drives Web Scalability
continued...page 3 of 3
 |
| More on scalability: |
|
|
| TechEncyclopedia: |
 Send Us Your Feedback |
Defining an initial investment is very easy and can prove quite inexpensive. However, once Web-site activity starts growing, the infrastructure must be able to keep up--and it's best to anticipate such evolution.
The amount of money needed to handle increasing traffic can be estimated rather accurately with appropriate testing. Such testing provides information about system capacity, how much money is required to upgrade a system, and how fast it can be done.
Testing must come into play very early in the life cycle of a project. Of course, functional testing can't be avoided without risking improper functioning of the site. However, even if time-to-market is top priority; it would be a big mistake to skip performance testing as well.
There are many tools for load testing on the market today, among them Mercury LoadRunner, Segue SilkPerformer, and RSW e-Tester.
By carrying out performance bench-marks in the early stages of development, when it's easier to adapt source code or configurations, it's possible to identify potential bottlenecks or deadlocks. Putting off performance testing until the very last minute--and this is often the case--is extremely unwise because any problems discovered can bring development back to square one.
It's often difficult to calculate (and even harder to convince business management of) the value of testing, but it's definitely worth the extra effort.
Once a Web site goes live, it's time for monitoring. Companies such as Keynote Systems Inc. provide performance and availability monitoring services. One of the benefits of outsourcing this service is that monitoring companies have agents throughout the world, offering a global view of a Web site's availability. Detailed monitoring enables a better understanding of the life of Web site and makes it easier to anticipate a weak link before it breaks.
A great example of well thought-out distributed computing is Google's search engine. Craig Silverstein, Google's director of technology, says the company was committed to building a fully distributed system from the very beginning. As a result, Google is made up of roughly 6,000 Linux servers that work together to provide the most accurate search results in less than a second.
"We designed Google architecture to be highly distributed from the beginning, using homemade programming, based on open-source software like Linux and Apache," Silverstein says. In light of Google's goals, the choice of open-source software as the main technology was a strategic move toward scalability.
Silverstein says the open-source community was very helpful whenever his teams ran into problems.
Finally, in becoming Yahoo's Web search feature in July 2000, the number of requests Google has to handle has jumped from an average of 15 million to more than 30 million per day. Despite this greatly increased traffic, neither Google nor Yahoo users have noticed a change in Google's quick response time.
Before Yahoo began using Google, both companies conducted workload benchmarks to flush out any potential problems, but the distributed nature of the Google system proved to be very scalable.
Silverstein's best advice: Think about scalability from the very beginning of a project.
The new Internet economy has made it even easier for unsatisfied customers to take their business elsewhere. Moreover, customers are becoming increasingly demanding. This is why, apart from intrinsic service value, performance and availability are the keys to successful Web ventures--and redundancy and distributed computing open the door to this success.
Jean-Christophe Cimetiere is an analyst and the CEO of TechMetrix Research, a technically oriented analyst firm. He can be reached at jcc@techmetrix.com
Illustration by John Howard
Back to This Week's Issue
Send Us Your Feedback
Top of the Page
