Language Selection

English French German Italian Portuguese Spanish

Debate Over the Size of the Web

Filed under
Web

How big is the World Wide Web? Many Internet engineers consider that query one of those imponderable philosophical questions, like how many angels can dance on the head of a pin.

But the question about the size of the Web came under intense debate last week after Yahoo announced at an Internet search engine conference in Santa Clara, Calif., that its search engine index - an accounting of the number of documents that can be located from its databases - had reached 19.2 billion.

Because the number was more than twice as large as the number of documents (8.1 billion) currently reported by Google, Yahoo's fierce competitor and Silicon Valley neighbor, the announcement - actually a brief mention in a Yahoo company Web log - set off a spat. Google questioned the way its rival was counting its numbers.

Sergey Brin, Google's co-founder, suggested that the Yahoo index was inflated with duplicate entries in such a way as to cut its effectiveness despite its large size.

"The comprehensiveness of any search engine should be measured by real Web pages that can be returned in response to real search queries and verified to be unique," he said on Friday. "We report the total index size of Google based on this approach."

But Yahoo executives stood by their earlier statement. "The number of documents in our index is accurate," Jeff Weiner, senior vice president of Yahoo's search and marketplace group, said on Saturday. "We're proud of the accomplishments of our search engineers and scientists and look forward to continuing to satisfy our users by delivering the world's highest-quality search experience."

The scope of Internet search engines, and thus indirectly the size of the Internet, has long been a lively area of computer science research and debate.

Moreover, all camps in the discussion are quick to note that index size is only loosely - and possibly even somewhat inversely - related to the quality of results returned.

The major commercial search engines use software programs known as Web crawlers to scour the Internet systematically for documents and index them.

The indexes themselves are maintained as arcane structures of computer data that permit the search engines to return lists of hundreds of answers in fractions of a second when Web users enter terms like "Britney Spears" or "Iraq and weapons of mass destruction."

On Sunday, researchers at the National Center for Supercomputer Applications attempted to shed light on the debate by performing a large number of random searches on both indices. They ran a random sample of 10,012 queries and concluded that Google, on average, returned 166.9 percent more results than Yahoo. In only three percent of the cases did the Yahoo searches return more queries than Google. The group said the Yahoo index claim was suspicious.

Neither Yahoo nor Google makes public the software algorithms that underlie their collection methods. In fact, those details are closely guarded secrets, which lie near the heart of heated competition now going on between Google, Yahoo and Microsoft over who can provide the most relevant answers to a user's query.

"It's a little bit silly," said Christopher Manning, a Stanford University professor who teaches a course on information retrieval. "It's difficult, and the whole question of how big indexes are has clearly become extremely political and commercial."

Even if the methodology is unclear, there is no shortage of outside speculation about what the different numbers mean, if anything.

Jean Veronis, a linguist in France and director of the Centre Informatique pour les Lettres et Sciences Humaines, posted a discussion on a blog noting that the increase in Yahoo references in French appeared consistent with the larger overall number that Yahoo was now reporting.

He added a caveat, however. "All of this should of course be taken with a large pinch of salt," he wrote. "So far, I haven't quite caught Yahoo red-handed when it comes to fiddling the books, but this could simply be because they are smarter with their figures than their competitors ;-)"

In contrast, a fellow blogger, Akash Jain, did his own random query test and wrote that it appeared that Google's index remained about 50 percent larger.

Other search engine specialists remained skeptical about the ability to estimate Web or index size as long as the search engines were being secretive about their methods. "I don't have any good way of checking," said Raul Valdes-Perez, a computer scientist who is chief executive of Vivisimo, which operates the Clusty search engine. "It feels a little like Harvard and Yale decided to argue over who has the most books in their respective libraries."

By JOHN MARKOFF
The New York Times

More in Tux Machines

More of today's howtos

GNOME News: Black Lab Drops GNOME and Further GNOME Experiments in Meson

  • Ubuntu-Based Black Lab Enterprise Linux 11.0.1 Drops GNOME 3 for MATE Desktop
    Coming about two weeks after the release of Black Lab Enterprise Linux 11, which is based on the Ubuntu 16.04.2 LTS (Xenial Xerus) operating system using the HWE (hardware enablement) kernel from Ubuntu 16.10 (Yakkety Yak), Black Lab Enterprise Linux 11.0.1 appears to be an unexpected maintenance update addressing a few important issues reported by users lately.
  • 3.26 Developments
    My approach to development can often differ from my peers. I prefer to spend the early phase of a cycle doing lots of prototypes of various features we plan to implement. That allows me to have the confidence necessary to know early in the cycle what I can finish and where to ask for help.
  • Further experiments in Meson
    Meson is definitely getting more traction in GNOME (and other projects), with many components adding support for it in parallel to autotools, or outright switching to it. There are still bugs, here and there, and we definitely need to improve build environments — like Continuous — to support Meson out of the box, but all in all I’m really happy about not having to deal with autotools any more, as well as being able to build the G* stack much more quickly when doing continuous integration.

Fedora and Red Hat

Debian and Derivatives

  • Reproducible Builds: week 108 in Stretch cycle
  • Debuerreotype
    The project is named “Debuerreotype” as an homage to the photography roots of the word “snapshot” and the daguerreotype process which was an early method of taking photographs. The essential goal is to create “photographs” of a minimal Debian rootfs, so the name seemed appropriate (even if it’s a bit on the “mouthful” side).
  • The end of Parsix GNU/Linux
    The Debian-based Parsix distribution has announced that it will be shutting down six months after the Debian "Stretch" release.
  • Privacy-focused Debian 9 'Stretch' Linux-based operating system Tails 3.0 reaches RC status
    If you want to keep the government and other people out of your business when surfing the web, Tails is an excellent choice. The Linux-based operating system exists solely for privacy purposes. It is designed to run from read-only media such as a DVD, so that there are limited possibilities of leaving a trail. Of course, even though it isn't ideal, you can run it from a USB flash drive too, as optical drives have largely fallen out of favor with consumers. Today, Tails achieves an important milestone. Version 3.0 reaches RC status -- meaning the first release candidate (RC1). In other words, it may soon be ready for a stable release -- if testing confirms as much. If you want to test it and provide feedback, you can download the ISO now.