If Xerox PARC invented the PC, then Google invented the Internet.

Jeff Dean and Sanjay Gemawat

The truth about Jeff Dean surfaced on April 1, 2007. A list of funny facts about Dean, one of the first and most valuable employees of Google, appeared on Google's internal resource.

“Once Jeff Dean failed the Turing test. He calculated the 203rd Fibonacci number in less than a second.
')
Jeff Dean compiles and runs his code before submitting only to check the compiler and the central processor for bugs.

Before, the speed of light in a vacuum was only 35 miles per hour. Jeff Dean optimized the physics of light over the weekend "( more facts )

All these facts were, of course, fiction. But not so far from the truth. They came up with Kenton Ward (another Google employee), by analogy with the facts about Chuck Norris. Ward tried to remain incognito, but Dean figured it out.

Outside of Google, Jeff Dean's name was heard by units. Inside the company, it is perceived with trepidation. With the participation of Dean created the main technologies of Google, thanks to which the search giant today occupies a dominant position on the Internet.

Many have heard about Xerox PARC - the science center, where they invented a computer mouse, laser printers, laptops, Internet protocol, graphical interfaces, WYSIWYG principle and other important technologies. Much less is known about Google’s think tank. Google diligently protects its secrets, and its employees - such as Dean - do not seek publicity themselves. Meanwhile, Google is the Xerox PARC of our time.
“Google gathered the most talented people from around the world. The best minds of Bell Labs and Xerox PARC went there, ”says Mike Miller, a professor of particle physics at the University of Washington and principal researcher at Cloudant .

The key technologies for Google are those that manage a huge fleet of its servers. These are Google File System, MapReduce and BigTable, programs that break the load from online applications into small pieces and distribute among thousands of machines. This new generation of servers, network equipment and data centers, specially designed for Google software. Google data centers are like an anthill: each component of the system does a small job, but the data center itself works as one.

At a time when Silicon Valley was obsessed with social networks and touchscreens, Google was in the process of rebuilding its network infrastructure. Soon other online giants followed his example. Google File System and MapReduce inspired others to create a Hadoop distributed computing system. BigTable triggered the emergence of the NoSQL movement. And Google’s approach to data center equipment was picked up by Facebook, Amazon, Microsoft and others.

The impact of Amazon on the development of key web technologies is also great. But the influence of Google is much broader. Unlike Xerox PARC, whose inventions were mainly used by other companies, Google derives maximum benefit from its unique technology. With GFS and MapReduce, the company is far ahead of the rest of the world. Now they have changed the technology of the new generation. And again the whole world falls behind.

Twins

Kenton Ward chose Jeff Dean as a target for his April Fool jokes not by chance. Dean is the most vivid example of Google's "superman". But there were other alternatives.

For example, Sanjay Gemavat, Dina’s long-term partner. In 2004, they were listed as authors of a document from which the world first learned the details of MapReduce. They also played a big role in the development of BigTable. And they (plus one more person) first told about the Google File System.

Dean and Gemavat work in such close contact that they often even sit at the same computer. Prints while Gemavat. And when working separately send each other a code to check. If the code is OK, the verifier answers something like: “LGTM” (“Looks good to me”).

They met about twenty years ago, in the pre-Internet era, in the research laboratory of the company DEC. Dean was at the Western Reserarch Lab, and Gemavat worked two blocks from there, at the System Research Center. In DEC, they were engaged in the creation of a new compiler for Java and server software.

In the late 90s DEC was not in the best condition. The company made servers on RISC-architecture, while everyone else switched to cheap x86-chips. In 1998, DEC became part of Compaq (and later HP).
In the combined company, research laboratories are no longer needed. The exodus of people has begun. Someone went to Microsoft, which created a new office in Silicon Valley, someone in VMware, someone in Google. Ironically, DEC disappeared in the very year Google appeared. Simultaneously, the collapse of Xerox PARC and Bell Labs.

Dean, Gemavat, and several other DEC employees — Mike Burrows, Soon-Tac Leung, Luis Andre Barroso — decided to settle at Google. At that time, they were just looking for an interesting place to work. But the circumstances were such that Google, willy-nilly, became the heir of DEC. The first search engine AltaVista was created in DEC. She worked on big, slow cars. Google went a different way: they created technologies (GFS, MapReduce, BigTable) to break down the load between small and cheap servers.


Former building of the research center DEC. This is now an Amazon office.

Jeff Dean and his mentor

Jeff Dean first moved to Google. He appeared there after his academic mentor Urs Holtze.

Holtz was one of the top ten Google employees and the first vice-president of engineering. He oversees the infrastructure of Google, numbering about 35 data centers. He came to Google from the University of California, Santa Barbara, and before that he worked on web technologies at Stanford under the direction of David Younger.

Supervisor Dean studied with Jünger. As a result, Holtz became the academic mentor of Dean. In 1999, Jeff Dean left DEC for the MySimon startup. But then he met Holtz, who already worked at Google. Dean asked him about the vacancies. And soon received an invitation to work from Larry Page.
The first time Dean was engaged in the development of an advertising system. But soon it switched to fundamental technologies: the Google infrastructure was barely able to cope with the rapid growth of the Internet.

Following Dean, Google’s DEC acquaintances, Sanjay Gemawat, Krishna Bharat and Monika Heninger, moved to Google. Several more people followed.

Over the next three to four years, Dean and Gemavat created several iterations of Google search technology. How is their union? Very often they sit at the same computer and drink cappuccino in buckets. Dean gushes with ideas, and Gemavat restrains him, corrects the direction of thought and prints. Gemavat himself is not so passionate, he is often worried how to do it right. And then the situation is saved by Dean with his inexhaustible energy.

The most notable breakthrough of Dina and Gemavata was the creation of GFS and MapReduce in the early 2000s. The most remarkable side of these technologies is that they do not fall when one or more servers fail. When you deal with tens of thousands of cheap servers, like Google, the technique falls constantly. With GFS and MapReduce, you can create copies of data on multiple machines. If one breaks, it is replaced by another.

BigTable, in turn, works like a huge database. It does not provide such control as a regular database, but allows you to operate with volumes of information that are inaccessible for platforms running on one machine.

Creme brulee as a clincher

Luis Andre Barroso also moved to Google from DEC. But he could not go.

In 2001, he was invited at the same time by Google and VMware. After several meetings in both companies, Barroso made a huge list of arguments in favor of each company. 122 reasons for working at Google and the same for VMware.

Then Barroso talked to Dean, and he asked if he included Google Chef in the creme brulee list. It turned out not. Creme brulee was Barroso's favorite dessert. The next morning, he accepted the Google offer.

In DEC Barroso participated in the creation of the first multi-core processors. At Google, he initially worked on software, but soon Holtz appointed him responsible for the entire physical infrastructure of the company.
The results were not long in coming. In 2003, Google for the first time ordered its own developed servers at Asian factories. Each server contained a 12-volt battery, which was turned on when the power was turned off. It turned out to be much more economical than supplying data centers with massive UPS.

Having dealt with the servers, Google took up the design of buildings for data centers. Before that, she just rented them. The first data center of the new generation opened in Oregon, in rural areas with low taxes and cheap electricity. But the main idea was to make the whole complex work as a whole. Barroso and Holtz called this "warehouse-scale computing".

During the construction of the data center, a new approach was applied. Servers and related equipment were placed in standard railway containers, and the containers themselves were used as modules in the “assembly” of the data center. The Oregon data center earned in 2005, but Google only revealed details about it to 2009.

Tesla effect

Larry Page has a thing about Nikola Tesla. According to Stephen Levy’s In the Plex book, Page puts Tesla as inventor on par with Edison, but notes his inability to commercialize his inventions.

The history of Nikola Tesla clearly influenced the mindset of Google management. Like Apple, Google keeps its key technologies in deep secrecy. And if something tells, then only a few years. “We try to be as open as possible,” says Holtz. “We share ideas, but not ways of implementation.”

In 2003-2004, Google talked about GFS and MapReduce. Shortly thereafter, the developer Doug Cutting created Nutch, an open source search engine that later transformed into Hadoop. Hadoop adopted Facebook, Twitter, Microsoft, and other companies followed them. A market of hundreds of millions of dollars has formed around Hadoop.

The same story repeated with BigTable. Google talked about this technology in 2006, and in 2007, Amazon talked about its Dynamo system. This gave rise to the emergence of NoSQL - an attempt to create databases that are scalable on thousands of machines.

“If you look at any NoSQL solution, you will immediately notice that it is based on either Google BigTable or Amazon Dynamo,” said Jason Hoffman from Joyent . “And if Google and Amazon wouldn’t tell anything, where would the world be?”

Little is still known about Google's hardware technology. Own servers are now designed by many - Facebook, Amazon, Microsoft. And modular data centers have become the backbone of the Internet.

Mike Manos, a former Microsoft data center specialist, denies that modular data centers were invented by Google. According to him, something similar happened in the 1960s. Even so, the merits of Google does not detract. As in the case of GFS / MapReduce, the company adapted the old idea to solve new problems.

Google's past is our future

Google has already replaced many of its germ technologies. Colossus came to replace GFS, and the search index is formed using the Caffeine system, which includes a MapReduce piece and works in a completely different way. Also noteworthy are Pregel (a base for creating a map of relationships between data) and Dremel (a high-speed data analysis system). Pregel already has several open source versions, Dremel has at least one.

This is just the tip of the iceberg. Surely there are other developments about which we still do not know anything. Last May, Eric Brewer, a professor at the University of California at Berkeley, announced that he was leaving to create a new generation of Google infrastructure. “Cloud technology is still very young,” he said goodbye. - Much more needs to be done. Much to achieve.

Brewer's example - one of the best experts in distributed computing - shows that Google was not just the successor of Xerox PARC. Google has surpassed Xerox PARC.

The DEC (System Research Center, SRC) Research Center was created by Robert Taylor, who had previously founded a computer lab at Xerox PARC. Corporate arrangements were such that the path from invention to finished product took years. By the beginning of the 80s, Taylor got tired of it, and he left for DEC. But the same story repeated there. And that was where Jeff Dean worked.

Google managed to solve this problem. The company was built in such a way as to combine the advantages of scientific work (solving problems that no one has yet solved) and a commercial enterprise (quick implementation).

The only serious obstacle for some Google employees is secrecy. It is psychologically difficult to work at the forefront of technology and at the same time not be able to talk about their achievements outside the company. Xerox PARC has never been as good as Google today.

Cade Metz / Wired