CACHET: Content providers—e.g., CNN, L. L. Bean—need to attract customers to their site…and keep them there. This requires fast, easy, reliable access, personalized content, fancy graphics/multimedia. But ISPs today cannot give content providers the quality of service that they need. Result: The World Wide Wait.
PROBLEM ONE: Flash crowds. Content becomes unexpectedly popular and swamps both the server and network. Existing solution: design system for max load, which is too expensive and unpredictable. CNN’s solution: remove all large content, which degrades the site quality and is an administrative nightmare.
PROBLEM TWO: Distance. Almost all users [are] distant from [a] central site, which costs money and takes time. One solution: Content Providers install servers at hosting farms (e.g. Digex, Exodus, UUNET) that are located near network access points—but these are still less than halfway to [the] user.
CURRENT TECHNOLOGY: Mirroring, which fully duplicates all content at a few locations, but is very expensive, limited in scale, and requires huge administrative overhead.
THE AKAMAI BREAKTHROUGH…Enables ISPs to host content at thousands of locations worldwide with the flip of a switch. This means: content is always near the user that wants it, intelligent replication handles flash crowds, content providers maintain control over their databases, content is always fresh, technology is transparent and implemented in the software, and it allows for enhanced graphics and multimedia and reduces bandwidth costs.
What the Akamai team didn’t know at the time, however, was that asking Crosbie to present would prove a fatal misstep.
Mark Gorenberg of Hummer Winblad still remembers the presentation clearly; it was the only time in his four years of judging that he’d seen someone other than an MIT student on the competition’s stage. “They decided that, in order to have a better chance of winning, they would let the boat’s anchor, in venture parlance, make the presentation,” said Gorenberg. “The judges who didn’t know the company ahead of time didn’t know what they were trying to do; it wasn’t clear from the presentation. He (Crosbie) had a great resume, but he didn’t really understand what they were doing.”
Crosbie was no match for a few of the contestants who followed, particularly MIT graduate Mike Cassidy, who was entering the 50K for the second time as an advisor to Direct Hit, a company that would make software for Internet search engines. Cassidy’s 1993 entry, Stylus Innovations, Inc., had made him a millionaire. This time, he was banking on Direct Hit—and his presentation was a sensation. The Akamai team left feeling uneasy. But if Lewin was anxious, he didn’t show it.
THE MAIN EVENT OF THE COMPETITION was held on May 6, 1998, before a large crowd of students and business folks who crammed into MIT’s Kresge Auditorium. Bill Porter, a Sloan graduate who founded E*Trade, delivered the keynote address. Following this, the teams presented again before the crowd. This time, Seelig and Nijhawan made the pitch, and the reception was much more enthusiastic. “At that point, it was just theater; the die had been cast,” remarked Seelig, referring to the fact that the Akamai team already knew, for a fact, they had not won.
In contrast to their losing performance the day before, Akamai made a lasting impression. “They blew it out of the water,” said Gorenberg, who remembered being chased down the hall by investors asking him why the judges had not selected Akamai as the winner. One of them, Gorenberg recalled, was entrepreneur Alexander Vladimir D’Arbeloff, an MIT graduate who co-founded the Boston-based, high-tech company Teradyne and chaired the MIT Corporation. To this day, Gorenberg says the story of the Akamai 50K serves as a critical lesson for budding entrepreneurs: a startup’s greatest weapon is often the passion of its founder.
At the end of the presentations, the judges spoke. For the first time in history, they said, “Our panel is deadlocked, and two teams will share the top prize.” The judges continued: “And the winners are…Direct Hit, and Volunteer Community Connection!” Then they delivered the list of runners-up. “In third place…CarSoft, which will connect home computers with diagnostic tools in cars. In second place…SiliconTest, which will design cards for the testing of semiconductor chips. In third place…WeddingBell.com, which will register wedding gifts on the Internet…” By the time they announced Akamai, which took fifth place, it seemed no one was even listening.
The winning teams took home $30,000 each, after an anonymous donor fattened the prize money with a $20,000 check. The CarSoft team took home a $10,000 prize. Then, in another unexpected turn, the Direct Hit team donated its $30,000 winnings to the other top two finalists after announcing they had received a commitment for $1.3 million in funding. The winners went on to launch successful businesses: Direct Hit was sold to Ask Jeeves for $507 million, and United Way acquired Volunteer Community Connection. The brains behind CarSoft, Diego Borrego, transformed the plan into Networkfleet, one of the most successful automotive technology companies.{27}
After almost a year of preparation and planning, the 50K loss proved a big blow to Lewin: “It was a slap in the face,” Leighton recalled. “Danny didn’t like losing.” Lewin was also angry with himself for pushing Crosbie out as their front man. And, as Seelig noted, “Danny was at a low after 50K. But he wasn’t as much down as he was pissed off, and I think that made him more selective about the team going forward, and more determined to make this work.”
In retrospect, both Leighton and Seelig conceded that the judges were right: they hadn’t deserved a win. Not because of Crosbie, but because their business plan, which positioned the company as a vendor of software to ISPs, was fundamentally flawed. The software segment was already crowded with potential competitors, and their technology seemed better suited to something larger scale. “I don’t think, at heart, we disagreed with the assessment,” Leighton admitted. “We realized where we were deficient. We didn’t think our business plan was ready for primetime or that it could really work.”
Lewin didn’t have time to linger on the loss, however. He had a master’s thesis to deliver, which included two new algorithms he still believed had the potential to change the delivery of content on the Internet. Leighton, concerned that his supervision of Lewin’s thesis could appear as a conflict of interest considering their collaboration on Akamai, suggested that Lewin secure a co-signer on the paper. He needed someone whose name would carry with it a seal of approval, signifying that his work was academically rigorous. The best person for this was obvious: Professor Karger. Lewin agreed, and put aside his earlier displeasure with Karger in order to deliver the best possible thesis.
On May 22, 1998, Lewin delivered his thesis, titled: “Consistent Hashing and Random Trees: Algorithms for Caching in Distributed Networks.” In it, he developed what he called “the algorithmic foundation of a large-scale distributed caching system for the World Wide Web.” The algorithmic foundation was consistent hashing, which Lewin combined with a data replication technique called “random trees” to come up with a way to allow a large network of caches to cooperate without overloading any one server. The system would remain robust, he wrote, even when overloaded and in a constant state of flux: “Servers can become swamped unexpectedly and without any prior notice. For example, a site mentioned as the ‘cool site of the day’ on the evening news may have to deal with a ten thousand-fold increase in traffic during the next day. This is known as the “flash crowd” phenomenon and the server in such a case is called a ‘hot spot.’”
In the acknowledgments, Lewin thanked Leighton, then added: “Most importantly, I would like to thank my lovely wife and kids for being there.” The paper garnered him the Joseph Morris Levin Award for the best Master Works oral thesis presentation, an academic honor awarded to just a few MIT students a year. Despite his academic home run, though, Lewin couldn’t let go of Akamai and the belief that it could build a better, faster Internet.{28}
A week after the 50K loss, on his twenty-eighth birthday, Lewin sent Greenberg a contemplative e-mail from Toronto, where he was presenting research on “Probabilistically ch
eckable proofs” at an academic conference:
Silicon valley makes money on air…There are many in this business who become horrendously wealthy AND THEY DON’T EVEN HAVE A PRODUCT OR CLIENT!!!!! …You may say—so [what,] that is great—you do nothing and become a zillionaire. Unfortunately this is not the case. The main thing you do have to do is dedicate yourself to telling lies for a number of years and to spreading bullshit for many, many years. Some people are comfortable with this as long as the payoff at the end is high enough. I am not. The plan is to become a successful company in the right way. That is: have a product, have a market, and have customers who are buying your product. In order to do this well, we have to focus on building the technology and not on fundraising.
Toward the close of the e-mail, Lewin also expressed hope that once he got the company off the ground, venture capitalists would take over the bulk of the work: “This will allow me to return to my family life and to see my wonderful kids and wife,” he wrote. “That’s something far more valuable than millions $’s!!!”
Lewin didn’t need millions. But he desperately needed some money in the bank. The financial woes he and Anne faced were only worsening; they were close to broke. Lewin had his long-term hopes pinned on Akamai, but the company still existed only on paper. He needed a summer income, and the best prospect came in an offer from the algorithms research group at Bell Laboratories, owned by Alcatel-Lucent in Berkeley Heights, New Jersey. It was a coveted position for any graduate student: an internship that paid in both prestige and money. It was also one Lewin had been hoping for, despite the distance it would create between him and his family, who would have to remain in Cambridge for the summer.
Lewin first learned about work at the legendary labs earlier that year, when two of his classmates, Anna Lysyanskaya and Yevgeniy Dodis, secured internships of their own. By the time he got around to applying, however, rumor had it that all the paid positions had been filled. When he heard this, Lewin immediately e-mailed Dodis, pleading with him to ask the scientist in charge of the internship program to make an exception:
FROM: Daniel Lewin
TO: [email protected]
SUBJECT: Re: elementary question
Tell him that the life of a man with two children has been devastated by this (decision not to hire more students)—he cannot justify this pain and suffering that has been cast down upon me.... Oy! It is a sad story... So sad... That surely will convince him!
Fortunately, no one needed convincing. Lewin got the job, and in June of that year, he left Anne and the boys in Cambridge and moved into a shared student apartment in New Brunswick, New Jersey, with three MIT classmates for the summer. It could not have been easy on Anne, juggling schoolwork and parenting Eitan and Itamar alone. But they needed the money, and Lewin promised to make the long drive back to Cambridge every weekend to be with them.
Chapter 5
Secret Sauce
“We can’t solve problems by using the same kind of thinking
we used when we created them.”
—ALBERT EINSTEIN
THE SUMMER OF 1998 marked the splintering of Akamai’s 50K team. There was no questioning the fact they had a great idea expressed in some beautiful theorems, but you can’t sell theorems. Their business plan was flawed, their technology unproven and to date, they had not one line of code. A few of the initial advisors to the company lost interest, some lost faith. Others simply couldn’t stay in Cambridge and wait for something to happen. Preetish Nijhawan, who first inspired the idea of forming a company for the competition, parted ways for a plum job at McKinsey & Co. in his hometown of Houston, Texas. Of the core team, however, Lewin, Leighton, and Seelig held fast to the idea that Akamai could be realized. In an act of faith—one aimed at protecting themselves against potential competitors—Lewin and Leighton submitted a patent application for Akamai’s technology. They knew they couldn’t stop anyone else from finding a way to speed up the Internet, but hoped that by making the mind-bending math underlying the Akamai software proprietary, they could prevent competitors from offering a better solution than their own.”{29} Despite going their separate ways for the summer, they decided they would each forge ahead on their own: Lewin in New Jersey, Seelig in California, and Leighton in Massachusetts. Sometime around Labor Day weekend, they agreed, they would regroup in Cambridge and make what they knew would be a difficult decision: abandon Akamai or go full steam ahead.
That summer, Leighton’s day job in Cambridge was to supervise a group of a dozen undergraduate students on a research project sponsored by LCS called Hacker Haven.{‡} Leighton was given free rein with the students, so he decided to enlist them to try building a prototype of the Akamai technology. If it worked, Akamai would be able to prove the technology could function—at least in the confines of LCS. To motivate the young, industrious hackers, Leighton had what he considered to be a powerful force of inspiration: Star Trek. During the course of the 50K competition, the Akamai team connected with some representatives from Paramount Digital Entertainment, which owned the rights to the original Star Trek TV shows. With a plan in the works to eventually distribute all the old shows online, Paramount took great interest in Akamai’s proposed technology. To demonstrate their enthusiasm, company executives sent a gigantic box filled with Star Trek paraphernalia to Leighton’s office at LCS. He wisely used it as bait: “I passed out all these goodies and said, ‘Look, if you guys build a prototype that actually works, we’ll be able to distribute Star Trek TV shows.’” The result? “They went nuts. They worked around the clock.”
Seelig spent the summer in the San Francisco Bay area, working at a consulting firm as a paid intern on a project for a Cambridge-based company called Cascade. With a few months to form West Coast connections that would be critical to Akamai’s success, Seelig began meeting with everyone he knew and everyone in their collective Rolodexes. Friends. Family. Friends of family. And friends of friends. Gorenberg, the 50K judge at Hummer Winblad in San Francisco, was impressed enough by Akamai after the contest that he agreed to orchestrate some key meetings, most importantly with a worldwide data center company, Global Center, which Seelig would approach as a potential host for Akamai’s servers. Through investors at Sequoia Capital, Seelig was able to connect with someone at Yahoo, who offered hope of an eventual meeting with the company’s founder, David Filo. Seelig also sought advice on a business model, which remained undecided. Marco Greenberg also came through with some big name connections, including a few potential investors in the world of media and entertainment.
At Bell Labs, Lewin enjoyed a more relaxed, predictable pace compared with the frenetic energy of his first few semesters at MIT. Built in the 1920s by AT&T, the lab’s campus in Murray Hill, New Jersey, (later razed in 2002) was rich with prodigious intellect. The internationally renowned lab was the birthplace of some of the most innovative scientific advances of the century—radar, satellites, and the wireless network. In his book The Idea Factory, author Jon Gertner describes the atmosphere of the labs as both magical and eccentric. Scientists would crisscross the campus on unicycles and occasionally set off explosions.{30}
Despite the fact that he was tasked with several research projects for the algorithms group, Lewin spent a lot of the summer having fun. He shared a dorm at University of New Brunswick with three friends from LCS, including Yevgeniy Dodis and Anna Lysyanskaya. Together with a select group of students from all over the world, they caught a shuttle bus every morning for the one-hour drive to Berkeley Heights. In the spare time he had—evenings, or weekends when he didn’t travel back to Cambridge—Lewin hung out with his roommates playing tennis, hiking, and canoeing down the Delaware River. “That summer, it was an incredible time,” recalled Dodis. “There were so many of us there from MIT, and it was one of the happiest times of my life.”
Even outside of MIT’s earnest atmosphere, Lewin and his roommates didn’t party. Every once and a while they’d enjoy a few beers, but most of their
entertainment came from late nights challenging each other with math puzzles. In one e-mail, dated July 21, 1998, Lewin sent a joking reply to fellow interns when it was suggested that he was slacking off.
FROM: Lewin Daniel
I object to your insinuation that people do work here. If you are trying to subtly suggest that we are not trying to get work done, then I can save you the perturbation: WE ARE NOT TRYING TO GET WORK DONE!
—Danny
In fact, Lewin was working hard, focusing his research on problems related to Akamai. At the end of the summer, all interns at Bell Labs were required to present their work to the fellows and researchers. According to Dodis, these talks didn’t usually draw a big audience, but the day Lewin presented, people packed into the large room to hear about his work to speed up the Internet. “When we walked in, he was already the center of attention. People were bombarding him with questions, but he was very confident. In that moment I saw him as this super ambitious guy,” recalled Dodis.
By summer’s end, Leighton had a promising development for Akamai from LCS. The prototype constructed by his Hacker Haven students was a success. To simulate a global network of servers, the students used separate floors of the Tech Square building: Paris was the seventh floor, and London sat on the sixth. They spent most of the summer cranking out the code to program the servers. With a batch of machines in place, they set up the mock network and created a functioning prototype. The biggest victory was demonstrating, inside the walls of LCS, the system’s fault-tolerance: if they shut down one machine, the system would rebalance the data load and avert a crash. As the students increased the flow of traffic on the prototype, they were stunned to see it continued to run smoothly and efficiently. The more content they loaded into it, the better it performed, even under peak load.
No Better Time Page 8