Google updates enterprise search appliance

The new box is the first upgrade for Google's enterprise search appliance since the appliance was launched two years ago, and it marks a significant boost in performance, according to David Girouard, the general manager of Google's enterprise division.

The new system is bigger, with more processing power and memory, allowing a single box to index up to 1.5 million documents and 300 queries per minute, fivefold increases over the earlier model, he said. The boxes cost between $32,000 and $175,000 each, depending on the configuration.

Girouard also touted software improvements such as secure sign-on, which limits queries to only those documents that a particular employee is authorized to see. In addition, the new system continuously crawls corporate intranets seeking out and updating only new documents and changes in existing documents. The previous system used a less efficient batch-crawling technique, Girouard said, which updated the entire index once every two or three days.

Google dominates Web search, but it has struggled to transplant its success seamlessly to the business of simplifying document searches for corporations. Significantly, some of the techniques that have made the company a household name for serving up fast and relevant Web results don't work in the vastly different arena of corporate intranets.

For example, the company's PageRank system, licensed exclusively from Stanford University, looks at link structures of pages on the Web as a hint to their relevance. Pages that have lots of other pages pointing links at them are likely to be more relevant than others that have fewer links. PageRank is widely considered the key to Google's early success, although it has become less important recently due to unceasing attacks from link "spammers" who hope to boost their rankings by exploiting its quirks.

Since corporate networks do not generally have rich link structures, PageRank is not an efficient way to rate the relevance of documents there, Girouard admitted. Instead, the company relies on some 100 supplementary algorithms, including several that it has optimized for enterprise network structures.

Two years after Google jumped into the market, enterprise search remains an open field for a number of competitors, including Verity, Fast Search & Transfer, Copernic and IBM.

In 2003, Google enterprise search accounted for $48 million, less than 5 percent of the company's sales, according to a securities filing. That's well behind longtime market player Verity, which racked up $121 million in sales last year.

Announced customers include Boeing, Cisco Systems, PBS.org and National Semiconductor.

Google is also said to be developing separate search software for PC desktops, as it seeks to prepare for expected competitive clashes with Microsoft and Yahoo, both of which are busy developing and expanding search products.

Girouard declined to discuss business prospects for the new appliance, citing a quiet period in advance of an initial public offering slated for later this year.

"There's a lot of skepticism in the market," Girouard said. "The solutions out there haven't been that great. People have concluded that Web search is easy and enterprise search is the hard stuff. It doesn't work that well, and people wonder if it can really be done well at all...We believe there are a lot of parallels between the Web and intranets that can help us solve the problem."