Getting a Boost: Zend Helps AuctionWatch Increase Performance

by Andi Gutmans and Zeev Suraski

Since its launch in 1998, AuctionWatch has relied on PHP for rapid solution development on its Linux dual P3 systems and Apache server platform. "PHP makes prototyping very simple and fast," explains Ben Margolin, Director of Seller Services and Tools. "Plus, it's easy for developers to learn. C and Perl coders have no problem with it, and HTML producers can navigate it easily." Approximately 70 front-end servers run the site's PHP applications.

With 4 million unique visitors a month accessing an estimated 100 million page views, AuctionWatch offers aggregation services that make it a one-stop solution for buyers and sellers. Its universal search feature allows buyers to find items across hundreds of auction sites. Its seller services provide the tools needed to conduct business with maximum efficiency and profitability.

While monitoring the performance of applications in production, AuctionWatch saw that one of its largest clusters was running out of memory. It was an urgent signal that demand on the system was beginning to outpace performance. In addition, the peak load on the problem cluster was running between 2 and 4. "Our goal for peak load is 1, and when we hit 1.5, we take action. So seeing 2 to 4, we had to act quickly," recalls Margolin. At that point, AuctionWatch turned to Zend Technologies and became a field test site for the new Zend Cache. "The timing was perfect. We're on such a fast ramp that re-architecting our applications or developing new ones is not always a practical option, because of the time required. Adding more servers is a quick alternative but, long term, not a cost-effective one."

To help navigate its high-growth business, AuctionWatch was also looking for a cost-management strategy to keep system and maintenance expenses under control. "Each dual server with support runs about $4,000. Plus, for each 10 machines we add, we need one more person on staff to operate and maintain them. Over time it just doesn't make financial sense to add a machine whenever more performance is needed," Margolin notes.

During the field test, AuctionWatch tested the Zend Cache on their problem cluster. "The results were immediate. With the Zend Cache, an unacceptable peak load of 2 immediately dropped to an optimal .5." AuctionWatch had even more ambitious goal for the Zend Cache. "We wanted to see if installing the Zend Cache would enable us to reduce the number of machines in a cluster. It did. We reduced our 16-system cluster to eight systems, while maintaining an acceptable load of 1 -- a proven TCO drop of 50 percent." With such favorable results, AuctionWatch is currently testing the product on other clusters.