Linhagem do fabricante
AltaVista
Built to stress-test a processor, it became the best search engine on the web - and was then turned into a portal.
AltaVista was not conceived as a search engine. DEC's Alpha 64-bit machines were so fast that the standard benchmarks of the day could not stress them, and a researcher on vacation sketched a way to generate a workload chaotic enough to try: point a crawler at the whole World Wide Web.
Paul Flaherty had the idea. Louis Monier wrote the crawler - Scooter - and Michael Burrows wrote the indexer, at DEC's Network Systems Laboratory and Western Research Laboratory in Palo Alto. It launched on 15 December 1995 at altavista.digital.com with an index of sixteen million documents, an enormous number at the time, offering full-text search with Boolean operators and natural-language queries when the alternative was a hand-maintained directory.
It worked immediately. By 1996 it was the exclusive search provider to Yahoo, which had itself begun as a directory. AltaVista also shipped Babel Fish, the web's first machine-translation service, named for the fish in Douglas Adams.
Then ownership changed hands four times in five years. Compaq acquired DEC in 1998 for $9.6B, the largest technology takeover to that point, and reportedly some Compaq executives did not know AltaVista existed. Compaq paid a further $3.3M for the altavista.com domain, which someone else had registered in 1994. Under a new chief executive the search engine was rebuilt as a portal - shopping, free email, news - to compete with Yahoo, which meant competing on everything except the thing it was best at.
In June 1999 Compaq sold 83% to CMGI for around $2.3B in stock. The planned IPO was cancelled when the bubble burst in 2000. AltaVista shed the portal features and refocused on search quality by 2002, which was good work arriving late: Google had taken the market. In February 2003 Overture bought it for $140M - roughly six percent of the CMGI valuation three years earlier - and Yahoo acquired Overture five months later.
By 2004 Scooter and the original ranking engine were switched off and AltaVista became a front end for Yahoo's results. It was formally shut down on 8 July 2013. The technology was not what failed; the decision to stop being a search engine was.
Histórias de fundação
AltaVista
The name is Spanish for high view, chosen for the hills above Palo Alto. Ilene Lang, brought in to grow DEC's software business, became its first chief executive - a detail worth noting because the project began as research and acquired a commercial structure around it rather than the other way round.
A linha do tempo
- Launch
Three hundred thousand hits on the first day.
- Eighty million hits a day
And around $50M of sponsorship revenue - the first demonstration that search itself was a business rather than a feature.
- Twenty machines, and the reason they mattered
The back end ran on twenty multi-processor Alpha systems holding 130 GB of memory and 500 GB of disk between them. In the same year a survey of professional researchers put it first at 45%, more than double the runner-up.
- Mapping the shape of the web
Researchers from AltaVista, IBM and Compaq used the crawl data to publish the first serious analysis of how the web is actually connected. The crawler built to generate a benchmark load ended up producing the reference dataset on the structure of the web itself.
Produtos e soluções emblemáticos
- ScooterThe crawler. Multi-threaded when competitors' were not, which is the whole reason its index was larger: it fetched many pages at once rather than one after another.
- The full-text indexEvery word of every page, held in memory across the cluster so a query could be answered without touching disk - which is the part that made it feel instant.
- Babel FishBuilt on Systran's translation engine rather than in-house, and it outlived AltaVista's own search business - still running under Yahoo's name in 2008, years after the index behind it had been switched off.
- Image, audio and video searchAdded before the rest of the market thought of the web as containing anything other than documents.
Inovações-chave
- Sixty-four bits was the point, not the marketingA 32-bit machine can address four gigabytes. A web index large enough to answer quickly has to sit in memory, and by 1998 this one occupied 130 GB across the cluster - more than thirty times what a 32-bit architecture could reach. DEC's Alpha was one of very few processors that could address it at all. The search engine did not merely demonstrate the hardware; it could not have existed without it.
- Crawling in parallelFetching one page, waiting, then fetching the next is bounded by network latency rather than by anything the machine can do. Running hundreds of fetches at once is obvious in retrospect and was not standard practice, and it is the difference between an index of millions and an index of hundreds of millions.
- Full text, not curationIndexing every word of every page was a bet that the web would grow past any editorial team's ability to catalogue it. The directories that lost were not badly run - they were structurally unable to keep up, and the company that most needed that answer was buying results from this one within a year.
- A research artefact that mapped its own subjectThe crawl produced the data behind the first real study of the web's connectivity structure. That is an unusual afterlife for a benchmark workload, and it is a fair illustration of what corporate research laboratories produced when they were funded to be curious rather than productive.
Principais mercados
General web search, at a moment when the market did not yet know it was a market. It reached roughly thirteen million queries a day at peak and was the eleventh most-visited site on the web in 1998 and again in 2000 - and, revealingly, the clear favourite among professional researchers, who cared about recall rather than about a home page.
Its competitors were the directories, the other crawlers, and eventually a company with a better ranking idea and no interest in being a portal.
Posição nos relatórios de analistas
- There is nothing current to assess. What there is instead is one of the clearest cautionary records in this industry, and its clarity comes from the fact that the technology kept working the whole time.
- The index was the largest, the hardware was the fastest, the professional users preferred it, and the research it produced is still cited. The reversal of 2002 is the evidence rather than the argument: the same team and the same technology, returned to the same job, three years too late.
Esta empresa não opera mais sob este nome.