Chief Of The Year:Amazon’s Werner VogelsAmazon’s CTO is the face of the new tech leader: a technologist who can articulatethe business value of a technology like cloud computing to customers. This reportexamines how Vogels ended up in this role at the cutting edge of business technologyleadership.

Jan. 9, 2009

IN THIS REPORT

The New Face Of Tech Leadership . . . . . . .2

Vogels On Architecture And His Role . . . . .9

What We Can Learn From This Year’s Chief . . . . . . . . . . . . . . .14

Copyright 2009 United Business Media LLC. Important Note: This PDF is provided solely as a reader service. It is not intended for reproduction orpublic distribution. For article reprints, e-prints and permissions please contact: PARS International Corp., 102 West 38th Street, Sixth Floor, NewYork, NY 10018 ; (212) 221-9595; www.magreprints.com/quickquote.asp

The New Face Of Tech Leadership

FROM AN EIGHTH-FLOOR CONFERENCE ROOM at Amazon.com’s headquarters on Beacon Hill, with an expansive view of down-town Seattle to the north, Werner Vogels contemplates how largecompanies might use cloud computing. He points to system automa-tion and “autoscaling” beyond what IT departments have implement-ed internally. He envisions “partner clouds” of shared IT resources,data, and applications. He even suggests FedEx could one day run itsvital package-tracking application in the cloud.

“For most of these things,it’s really Day 1,” saysAmazon’s bearish CTO—he’s 6 feet, 5 inches tall—dressed, as usual, in grayand black.

If we’re on the cusp of the computer industry’snext major architecture,the one beyond client-server, Vogels has playeda key role in getting us tothis point. A former re-searcher in Cornell Uni-versity’s computer sci-ence department, wherehe specialized in large-scale distributed systems,Vogels joined Amazon in2004 to help the e-retailerdesign and scale its ITinfrastructure to handle

Werner Vogels

Amazon’s customer-facing CTO and cloud

evangelist isredefining the role

of the businesstechnologist

2 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

B y J o h n F o l e y

3 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

workloads many times Amazon’s own.“Amazon was reaching a point wherethey needed to rethink how theywere building their software—what

scale really meant, what reliabilityreally meant,” he says.

The overhaul was required to support theemerging Amazon Web Services business,which over the past four years has come to rep-resent the state of the art in on-demand, pay-as-you-go computing. With a user ID and acredit card, application developers and otherscan provision servers, storage, and other ITinfrastructure with unprece-dented ease.

Vogels’ name and face areoften associated with Ama-zon’s cloud, but AWS isn’t aone-man show. Senior VPAndy Jassy conceived the bus-iness model five years ago andhas had his hand at the wheelever since. VP Charlie Bell isthe lead technical manager ofAWS. VP Adam Selipsky is theliaison to the 440,000 develop-ers who have signed up so far.Vogels, Bell, and Selipsky re-port to Jassy, and Amazon balked at our sugges-tion that one of them could be singled out. Butwe did it anyway, selecting Vogels as Informa-tionWeek’s Chief of the Year, our highest editori-al honor. Here’s why.

Amazon’s 50-year-old CTO has emerged as theright person at the right time and place toguide cloud computing—until now, an emerg-ing technology for early adopters—into themainstream. He not only understands how toarchitect a global computing cloud consistingof tens of thousands of servers, but also how to

engage CTOs, CIOs, and other professionals atcustomer companies in a discussion of howthat architecture could potentially change theway they approach IT.

Amazon aspires to be “the earth’s most cus-tomer-centric company.” Vogels’ job des-cription as “an external-facing technologist”isn’t just consistent with that mission state-ment, it’s cutting edge. Too many CIOs, CTOs,and IT organizations as a whole remain inter-nally focused. They get treated like a cost cen-ter because they are a cost center—they don’tventure outside their organizations. Vogels, in

contrast, is constantly on theroad talking with customersabout what Amazon can do to address their computingneeds in its data centers, thenreporting back to the rest ofthe AWS team with ideas onhow to make that happen. Ifyou think Vogels’ situation isdifferent because he works fora “vendor,” think again. CIOsand CTOs in a variety ofindustries must get better atarticulating the business valueof their technology to cus-tomers. Vogels even talks

about Amazon building a “customer-orientedarchitecture.”

“Everybody has a lot of questions,” says Thors-ten von Eicken, CTO of RightScale, a startupwith a management platform for AWS andother cloud services, and a former colleague ofVogels at Cornell. Vogels, he says, can both “ex-plain the vision” of cloud computing and diveinto the technical minutia.

If all goes as planned, Amazon’s cloud willserve as an extension of corporate data centers

Joined Amazon as director ofsystems research in 2004;promoted to CTO and VP thenext year

Computer science researcherat Cornell University from 1994 to2004, specializing in large-scaledistributed systems

Research engineer with INESC, a computer science researchinstitute, from 1991 to 1994

Doctorate in computer sciencefrom Vrije University in Amsterdam

Writes All Things Distributedblog and uses Facebook, Friend-Feed, LinkedIn, and Twitter

Vogels Vitae

4 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

for new applications and overflowcapacity, so-called cloud bursting.Over time, Amazon will then take onmore and more of the IT workload

from businesses that see value in themodel. Customer-centric? What Amazon’s do-ing goes beyond that. Amazon’s cloud becomestheir cloud; its CTO, their CTO.

EVOLUTION OF AWS

Standing in the lobby of Ama-zon’s headquarters, a 1930s ArtDeco building that the $14.8 bil-lion-a-year company has out-grown, I see Amazon’s customerfocus in action. The receptionist, a burly guywith tattooed arms, is on the phone for fiveminutes, patiently helping an author whocalled fretting over inaccuracies in the des-cription of his book on Amazon’s site. Problemsolved.

My first meeting was with Jassy, a soft-spokenNew York transplant and sports nut with sevenTVs in his basement, who served as technicalassistant to CEO Jeff Bezos in 2002 and 2003before launching Amazon’s Web services busi-ness. As Jassy explains it, the seed for AWS wasplanted in 1999 when the company decided to“decouple” system components to support itsstrategy of letting other e-commerce compa-nies integrate the features of Amazon withtheir own sites. “We found religion around SOA,and it changed our thinking about systems andcomponents,” he says.

As the number of third-party Web sites tappinginto Amazon’s functionality to sell productsapproached a million, the APIs through which itexposed its underlying technology as servicesgrew ever more important. In 2003, Bezos askedJassy to develop a business plan. The next year,Amazon introduced its Simple Queue Service, a

hosted queue for buffering messages betweendistributed applications, into beta testing.

It was the first of many Amazon infrastructurecomponents to be offered as services. In 2006,

1958 Jerry Lee Lewis, Great Balls Of Fire 1959 Ray Charles, What I’d Say 1960 Miles Davis, Sketches Of Spain1961 Robert Johnson, King Of The Delta Blues Singers 1962 Booker T & the MG’s, Green Onions1963 James Brown, Live At The Apollo 1964 John Coltrane, Love Supreme 1965 Bob Dylan, Highway 61 Revisited 1966 Cream, Fresh Cream 1967 The Doors, The Doors 1968 Johnny Cash, At Folsom Prison1969 Rolling Stones, Let It Bleed1970 The Who, Live At Leeds1971 Marvin Gaye, What’s Going On 1972 Deep Purple, Made In Japan1973 Pink Floyd, Dark Side Of The Moon 1974 Genesis, The Lamb Lies Down On Broadway1975 Led Zeppelin, Physical Graffiti 1976 Eagles, Hotel California 1977 The Stranglers, Rattus Norvegicus 1978 Herman Brood & His Wild Romance, Shpritsz 1979 The Clash, London Calling 1980 AC/DC, Black In Black1981 The Police, Ghost In The Machine1982 Steel Pulse, True Democracy 1983 U2, Under A Blood Red Sky 1984 Talking Heads, Stop Making Sense1985 John Cougar Mellencamp, Scarecrow1986 Run DMC, Raising Hell 1987 Guns N’ Roses, Appetite For Destruction 1988 Public Enemy, It Takes A Nation Of Millions To Hold Us Back1989 Eric Clapton, Journeyman1990 Angelo Badalamenti, Twin Peaks Soundtrack1991 Nirvana, Nervermind 1992 Rage Against the Machine, Rage Against The Machine1993 Live, Throwing Copper 1994 Neil Young, Sleeps With Angels1995 Garbage, Garbage1996 James Cotton, Deep In The Blues 1997 Erykah Badu, Baduizm 1998 DMX, Flesh Of My Flesh, Blood Of My Blood1999 Red Hot Chili Peppers, Californication2000 Eminem, The Marshall Mathers LP 2001 The Strokes, Is This It 2002 Richard Locker, Jewish Cello Masterpieces2003 Linkin Park, Meteora2004 Green Day, American Idiot 2005 Fiona Apple, Extraordinary Machine 2006 Matisyahu, Youth2007 Foo Fighters, Echoes, Silence, Patience & Grace

Fine-TuningIn a blog post, Vogels picked a favorite album fromhis collection for each year since he was born.

5 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

Amazon introduced Simple StorageService (S3) and Elastic ComputeCloud (EC2), foundational pieces forAWS users. Over the past 12 months, it

has stepped up the rollout, letting cus-tomers launch Red Hat Linux, OpenSolaris, andWindows Server operating systems and MySQL,Oracle 11g, and SQL Server databases as virtual“machine images” in its cloud. Other recentlyintroduced capabilities include persistent stor-age, remappable IP addresses, user-selected ge-ographic zones for EC2 instances, fulfillmentservices, development tools for integrating withFacebook and Salesforce.com, and a contentdelivery network called CloudFront (see time-line, below).

Jassy, like everyone else at Amazon, is evasive oncertain questions about AWS. He won’t say howmany data centers the company operates, wherethey’re located, how much revenue AWS pulls in,or how fast it’s growing. (“They’re extremelysecretive—paranoid,” says one business partner.)But Jassy will say this much: “Much to our sur-prise, enterprise adoption is happening muchfaster than any of us anticipated.”

Amazon offers a few metrics as proof points.There are 440,000 registered AWS developers,29 billion objects stored in S3, and, earlier thisyear, the amount of network bandwidth con-

sumed by AWS surpassed that required for Am-azon’s retail site.

In terms of revenue, Amazon lumps AWS salesinto its nonretail “other” category, along with itsEnterprise Solutions Web hosting business (cus-tomers include Target and Marks & Spencer) andco-branded credit cards. In the first nine monthsof 2008, revenue from “other” operations totaled$367 million. That’s only 3% of Amazon’s $12.5billion in total sales over that period, but here’sthe cause for optimism: While Amazon’s overallbusiness grew 36% through the first nine monthsof 2008, revenue from those other operationsclimbed 45%. It’s reasonable to surmise that AWSis fueling that growth.

Jassy describes cloud computing as the undiffer-entiated heavy lifting of IT infrastructure—or the“muck”—and says businesses stand to benefit byoff-loading that task of keeping it all running. Hepoints to the 70-30 trap (sometimes characterizedas 80-20), where 70% of IT budgets are spent onmaintenance and only 30% on innovation. “One ofthe mantras of the business is to flip that on itshead,” he says.

CHIEF CLOUD OFFICER

Vogels’ assignment is to serve, in effect, as a chiefcloud officer for the computer industry. A high-mileage schedule has him on the road much

Amazon Web Services

Simple QueueService, ahosted queuefor storing messages isintroduced

MechanicalTurk, an onlinemarketplace foron-demandworkers, debuts

Simple StorageService (S3) provides on-demand access to Amazon’s datastore; ElasticCompute Cloud(EC2) goes intotesting

DevPay, SimpleDB, and Flexible Payments services go intobeta testing; RedHat EnterpriseLinux becomesavailable on EC2

Elastic Block Store, persistent data storage forEC2, and CloudFront CDN introduced

Elastic IP Addresses and Availability Zones giveusers control over IP addresses and geographicdeployment of EC2 instances, respectively

AWS Premium Support launched

Microsoft, Oracle, and Sun server softwarebecome on-demand options on EC2

2004 2005 2006 2007 2008

6 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

of the time. Earlier this month, he flew to San Francisco to attend aCTO roundtable on cloud computinghosted by the Association of Comput-

ing Machinery (ACM), and, from there,to London, Amsterdam, and Paris, where hespoke at LeWeb, an industry shindig sponsoredby Google, Microsoft, and a dozen other techcompanies.

On Dec. 10 at LeWeb,Vogels announced the ex-pansion of Amazon’s servers-by-the-hour EC2service to Europe. A regional cloud minimizeslatency and improves redundancy and, equallyimportant, lets companies comply with Europ-ean regulations requiring that customer data bestored locally. “This addresses the requests ofmany of our European customers and fromcompanies that want to run instances closer toEuropean customers,” Vogels wrote in his AllThings Distributed blog.

Vogels started his blog in 2001 while at Cornell,and he continues to post regularly, offering atech guru’s point of view on AWS services andfeatures and an occasional deep dive into theguts of the Amazon infrastructure, such as histreatise on internally developed technology forhigh-availability storage. The technology, calledDynamo, integrates a bunch of arcane distrib-uted system techniques such as DHTs, consis-tent hashing, versioning, vectorclocks, quorum, and anti-entropy-based recovery. “As far as I know,Dynamo is the first production sys-tem to use the synthesis of all ofthese techniques, and there arequite a few lessons learned fromdoing so,” Vogels wrote. Dynamoisn’t exposed as a Web service,though it underlies S3, he added.

Vogels, a native of the Nether-

lands, developed his technical chops at VrijeUniversity in Amsterdam, where he earned aPh.D. in computer science. At Cornell, hisresearch focused on large-scale distributedsystems, including networking protocols, mid-dleware, and cluster management. “In myeyes, Amazon is probably the world’s largestdistributed system,” he says.

Vogels’ academic and research backgroundgives him added authority when talking upcloud computing. Oh, sure, he wears a market-ing hat when on stage at industry events, butunderneath that hat is the mind of a technolo-gist who hears “cloud” and thinks of looselycoupled software services on a global distrib-uted network operating with near-perfectavailability.

INTERNAL INNOVATION

Within Amazon, a dedicated development teamworks on each of the major Web services—EC2,S3, SQS, SimpleDB, CloudFront. It’s an organi-zational structure intended to give those teamsmaximum autonomy. “We believe the groupsshould own their customers down to the metal,”says Bell.

A 10-year company veteran, Bell knows Ama-zon’s IT infrastructure as well as anyone. I firstinterviewed him eight years ago when Amazon

was on a shopping spree for

‘Werner provides a greattranslation of what we’redoing to the outside world.He’s a great communicator,’says Amazon veteran Bell, who overseesAmazon Web Services development.

7 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

commercial software, licensing prod-ucts from Excelon, Oracle, Manugis-tics, SAS Institute, and others. At thetime, Bell made the point that Ama-

zon’s differentiating features—its per-sonalization and customer-contact capabilities,for example—were internally developed. “Youcould characterize us almost as a software com-pany,” he said.

That’s even truer today than it was then. As thedemands on Amazon’s IT infrastructure haveincreased through the success of its retail, Webhosting, and AWS businesses, Amazon is build-ing more software and buying less. “There’shardly any third-party software left,” says Vo-gels. “If I could buy things, I would. But vendorshaven’t reached a point where they can deliversoftware that can reliably work at Amazonscale.”

(Also in that Nov. 6, 2000, InformationWeekstory, titled “Amazon’s IT Agenda,” was a com-ment by Amazon chief programmer JeremiahWilton that foreshadowed the hiring of Vogels afew years later. “Our greatest challenge, frommy point of view, has been scaling our systems,”Wilton said.)

Bell is responsible for making sure thatAmazon’s Web services work together in acoordinated fashion, as well as for the busi-ness side of AWS. The general managers ofeach Web service—including Peter De Santis,general manager of EC2, and Alyssa Henry,general manager of S3—report to Bell. Theirteams are spinning out new services andfeatures every week or two, and increas-ingly those capabilities—user-designated“availability zones,” block storage, EC2 inEurope, the CloudFront content delivery net-work—are geared for large companies.

That pace will continue into 2009, with EC2monitoring, a Web-based management console,load balancing, and autoscaling due soon.Amazon’s ability to propagate Web servicesgoes back to architectural decisions madebefore the term cloud computing was firstuttered. Bell puts it this way: “If you’re going tobuild a service in the company, you’re going tobuild it as a Web service, and you’re going tobuild it to be exposed.”

In its boilerplate, Amazon says 440,000 devel-opers have signed up for AWS. In fact, they’renot all developers; they range from entrepre-neurs to enterprise IT pros and everyone inbetween. Selipsky’s team supports that bur-geoning population and plays to its diversity.“We want to support any programming lan-guage that our customers want—Java, C#,Python, Ruby, PHP,” he says.

AWS is a popular platform among startups,Webcompanies, and software-as-a-service compa-nies. Increasingly, Amazon’s customers arehousehold names: Nasdaq, The New YorkTimes, Philips, SanDisk. Eli Lilly is using EC2 todeploy SQL Server/Windows Server instancesas needed for research data. The IndianapolisMotor Speedway uses AWS for Web site mirror-ing, video streaming, and digital image archiv-ing. Selipsky says there are a “slew” of otherenterprise customers, including a hedge fundcompany and a mutual fund company, thathaven’t been disclosed.

As the early examples demonstrate, enter-prise adoption tends to be limited to specificapplications; CIOs don’t want to commit toomuch, too soon. Vogels says we’ve onlyscratched the surface. When I ask whetherFedEx, one of the largest and most sophisti-cated IT users, could seriously consider AWSas an alternative to running mission-critical

8 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

apps in its own data centers, hedoesn’t hesitate. “I would love themto do that,” he says. “I think thatwould be a great benefit to them.”

The world of research, Vogels’ old stompingground, is another target market. This month,Amazon introduced AWS Public Data Sets, ano-cost repository for government, research,and other public data. And Amazon andHarvard Medical School co-sponsored a day-long forum in Boston where doctors, biomed-ical researchers, and other experts discussedhow to apply cloud computing. Harvard’sLaboratory for Personalized Medicine alreadyuses Oracle on EC2 for genetic testing modelsand simulations.

SPREADING THE WORD

After spending half a day at Amazon’s head-quarters, I still had questions for Vogels, but,traveling somewhere in Europe, he had gonesilent. Nevertheless, it’s easy enough to followVogels. He’s all over the Web—blogging,Twittering, and talking up cloud computing ininterviews and on video.

With his Dutch accent,Vogels comes across as abit of a Renaissance man—a music buff, pho-tographer, and patron of the arts who speaksfour languages and rides a motorcycle. HisTwitter posts alternate between references toserver optimization and network latency and anappreciation for “marvelous absurdistic physi-cal theater.”

I click on a video replay of Kara Swisher’s on-stage chat with Vogels at LeWeb. Amazon’s CTOtells the audience that the conversation withcustomers has shifted from cloud security toapplication deployment. Customers want toknow how to move existing applications into

the cloud, how much engineering work isinvolved, and how to exploit the functionalitythat Amazon makes available, he says.

Vogels, surprisingly, tells Swisher that securityisn’t the big issue it once was with customers,but he admits that AWS—which has experi-enced a number of hours-long service outagesthis year—has work to do in the area of reliabil-ity. “There’s no excuse for any downtime or fail-ure,” he says. “One hundred percent availabilityis the only goal that you can have.”

It’s a reminder that, even though Amazon isfour years into cloud computing, it’s onlyrecently begun addressing the stringentrequirements of enterprise customers, andthat work is unfinished. Amazon’s flagshipEC2 service, in beta testing for two years,became generally available just two monthsago. Its SimpleDB relational database serviceis in beta testing. EC2 monitoring, manage-ment, load balancing, and autoscaling are stillon the road map.

As my deadline approaches, Vogels resurfaces.I ask him about lingering concerns over datagovernance, security, and reliability in thecloud. He responds via e-mail that Amazonworks with customers to address these con-cerns, but that “no customer has exactly thesame needs as another.”

In other words, with Bell driving AWS develop-ment and Selipsky managing front-line rela-tionships, Vogels will need to keep racking upthe frequent-flier miles. As Amazon reachesout to customers, his ability to articulate thebenefits of cloud computing will have a lot dowith whether they return the embrace. Years ofengineering and cloud development are behindAmazon; the bigger job lies ahead.

© 2009 InformationWeek, Reproduction Prohibited

Q&A Vogels On ArchitectureAnd His Role As CTO

By John Foley

A MAZON CTO WERNER VOGELS spends more than half ofhis time on the road, helping IT departments with the transition tocloud computing. Vogels joined Amazon in 2004 from Cornell Univer-sity, where he worked for 10 years in computer science research, focus-ing on large-scale, distributed enterprise systems. At Amazon, he’s putthat background to practice in helping the e-retailer scale its IT infra-structure and open it to other companies in the form of Amazon WebServices.

In an interview with InformationWeek’s John Foley at Amazon’s Seattleheadquarters,Vogels discussed the architectural design and philosophybehind AWS, his role as a customer-facing technologist, and his visionof what’s next in cloud computing.

InformationWeek: Let’s start with the nuts and bolts. What’s the ITarchitecture behind Amazon Web Services?

Vogels: Around the time I joined, we already had established this large-scale service-oriented architecture. The phase before that, Amazon wasmainly databases and application servers. That had come to sort of anend of life as an architecture around 2000, 2001. We moved to this serv-ice-oriented architecture by taking individual pieces of business logicthat sat in the application servers, looked for the data that they operat-ed on, brought those together, and put an API on them, that’s what wecall a service. It allowed us an evolutionary path to a service-orientedarchitecture. Each of those pieces that make up the e-commerce plat-form are actually separate services. Whether it’s Sales Rank, orListmania, or Recommendations, all of those are separate services. Ifyou hit one of Amazon’s pages, it goes out to between 250 and 300 serv-ices to build that page.

It’s not just an architectural model, it’s also organizational. Each servicehas a team associated with it that takes the reliability of that service andis responsible for the innovation of that service. So if you’re the team

10 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

that’s responsible for that Listmaniawidget, then it’s your task to innovateand make that one better.Around thetime that I joined, we also did a deep

dive on what these teams are actuallyspending their time on, and when we did ananalysis on that, we found that a lot of thoseteams were spending their time on the samekind of things. In essence, they were all spend-ing time on managing infrastructure, and thatwas a byproduct of the organization that wehad chosen, which was very decentralized.

So in a traditional enterprise architecture step,we decided to go to a shared-services platformand that became the infrastructure servicesplatform that we now know in the outside worldas AWS. We first had to develop it for ourselvesin a way that those teams could focus on theinnovation side and not become super appadministrators and super operators, becausethere’s no glory in that, although at Amazon-scale, all engineers need to be aware of scale,reliability, and be able to failover their servicesfrom one data center to another.

So we have an e-commerce platform that con-sists of all of these services. On top of that, werun a number of very large e-commerce opera-tions—Amazon.com, our international sites,and sites like Marks & Spencer and Target. It’sa multitenant, large service-oriented architec-ture. Then we drop one step down and you getto the infrastructure services that power the e-commerce platform; most of those reaching theoutside world we know as AWS, but they weretargeted initially at internal customers. Belowthat sit our hardware services, the teams thatconstruct data centers and build them out anddo networking.

InformationWeek: Amazon is known as anopen source shop. Is that still true?

Vogels: Where in the past we could say this wasa pure Linux shop, now in terms of the largepieces of the e-commerce platform, we’re apure Amazon EC2 shop. There’s an easierchoice of different operating systems. Linux isstill very popular, but, for example, WindowsServer is often a requirement, especially if youneed to transcode video and things that have tobe delivered through Windows DRM [digitalrights management], so there is a variety ofoperating systems available for internal devel-opers. We have a long history of not trying torestrict our developers, allowing them to pickthe tools that they feel work best. If they wantto do prototyping, these days engineers willoften flock to Ruby, there are some experimentsgoing on with highly concurrent services, theymay be using Erlang, other groups are usingJava and C++ and things like that.

That also drove a very important principle overhow we constructed the infrastructure services,namely that if you were using Amazon EC2, youweren’t necessarily required to use S3 orSimpleDB or Elastic Block Store. Of course, youhope that all of these services are seductiveenough for our engineers to use them, but ifthey felt there were better tools available forthat particular task, then they should be free todo so. So it’s required in the way that we con-structed these services that there was no inter-nal lock-in. That’s the same advantage we giveto customers in the outside world; none of thesethings are locked-in.

Most of our software is best characterized asbeing homebuilt or homegrown. There’s hardlyany third-party software left, and it has nothingto do with that we don’t think third-party soft-ware is great. If I could buy things, I would, butin the past we’ve seen that to get to the reliabil-ity that Amazon requires, and to have the con-trol over both cost and performance, we need tohave much better control of the software that

11 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

runs our services, and for that ven-dors just haven’t reached thepoint yet where they can reliably

deliver software that can operateat Amazon scale. We build our soft-

ware, and if we do buy third-party software, weuse it in a form that every other customer isusing.

InformationWeek: At Cornell University, yourresearch centered on large-scale, distributedsystems. What’s that like in the real world ofAmazon?

Vogels: In my eyes, Amazon is probably theworld’s largest distributed system. That mightseem strange if we compared it to a number ofother things like large Internet sites that areout there, but Amazon is unique in that thereare many different software patterns that areactive; it’s not just request/reply, or indexing, ormassive caching; there’s large workflow piecesto it, dissemination pieces, content delivery. So,the many different pieces in the Amazon archi-tecture make it a very large—I won’t say com-plex, I don’t know if complex is the right word—very diverse platform. For us, there are twoprinciples that were important, especially inthe context of reliability. Every second of down-time has financial impact. Reaching availabilitythat—Jeff Bezos recently had a great term for it,“indistinguishable from perfect”—is as close to100% as possible is essential for us.

So there are two main principles that we useinternally. One is isolation, driven by the serv-ice-oriented architecture. There’s no directdatabase access except for the pieces of soft-ware that run on the service, and the servicehas a hardened API. That is the only way thatservices or software pieces can interact witheach other.

The second piece, and it’s a more fundamentalarchitecture piece, is these software compo-nents are loosely coupled, where the interac-tion between them is there’s no tight connec-tion or dependency between different pieces,which means if failure happen or if overloadoccurs, it’s easy for software components toswitch to other components that aren’t faulty orprovide better availability.We do that at a microlevel and at a number of higher abstraction lev-els, even to the point our systems are designedto withstand complete data center failures. Wehave a rule internally in the e-commerce spacethat we should be able to lose a complete datacenter without the SLA to the customer gettingviolated. So isolation and loosely coupled arethe two building principles that we use to con-struct the overall architecture.

InformationWeek: Opening the architecture tothe rest of the world seems to add a level ofcomplexity. You backed away from the termcomplexity, but it seems highly complex to dothis for thousands of customers, all of whomhave their own idea of what they’re going to dothere.

Vogels: We have quite a bit of experience withopening up Amazon to the outside world. Weallowed merchants onto the platform, and weopened up all of the data that lived inside theAmazon e-commerce platform for outsidedevelopers to use. The enterprise services plat-form is another one, dealing with all of theselarge platforms such as Target and Marks &Spencer. Having such a multitenant platformhas taught us a lot about how to guarantee per-formance and reliability while serving multiplemasters. There are many other pieces ofAmazon that we’ve opened up to the outsideworld, even the fulfillment center.You can go toan Amazon fulfillment center and sell your

12 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

goods and use a web service call to say,‘Mail that package to that customer.’

Whether there is a big differencebetween the external customers and

the hundreds of internal services we havethat can be seen as small businesses within theoverall Amazon frame, I don’t think there is thatmuch of a difference. Given the huge diversitywithin Amazon’s software architecture, each ofthese services has different requirements,whether it is storage, compute infrastructure,queuing. We’ve been surprised by how cus-tomers have started to use, for example, S3 foreverything from a backup system—the writeonce, read never approach—to a content deliv-ery network or for software distribution, andevery style of computing in between. Has thatmade it more difficult? I haven’t seen anythingin terms of software architecture or in the waythat we do operations that has demonstrated tome that we were not well prepared.

InformationWeek: Describe your job as CTO.

Vogels: My role has changed a bit. I once wrotea blog post about the different CTO roles, and Idescribed four different patterns. One is that ofan infrastructure manager, and one is that of atech visionary and operations manager. I’m notin that part, those are more CTOs that worktogether with a CIO and manage operationsand a data center and things like that.There aretwo other roles. One is what I call a big thinkerwho thinks about the bigger patterns: What isthe technology that you need to develop? And,at Amazon, what are the kinds of things that weneed to develop for our customers and how dowe drive customer-oriented development andcustomer-oriented architectures into the waythat we do things? When I started at Amazon, Iwas asked to be this big thinker, thinking aboutwhat are the principles, what is the strategy?

But now that we deliver more and more tech-nology services to the external world, my rolehas shifted to becoming a more external-facingtechnologist, someone that talks to our cus-tomers, understands what their needs are, thentakes that back and drives our internal roadmap based on the needs of our customers. Theother side of it is helping our customersbecome successful on our platform. What arethe kinds of things that Amazon can do to makesure that the transition into the cloud is asseamless as possible? I still help teams inter-nally with their architectures, but mainly basedon feedback from our customers.

InformationWeek: What are some of the pos-sibilities for cloud computing beyond some ofthe obvious Web applications?

Vogels: When customers move over to thecloud, they start thinking about how they canautomate their environments even more thanthey do now. Customers that find higherdegrees of automation turn out to be the oneswho gain the most benefit from moving to thecloud. There’s a range of things that customersdo to achieve that. For example, the Indy 500guys moved into Amazon EC2 and the feedbackwas that on people cost alone they save about50% because they could go from a hand-man-aged environment to EC2 where they couldautomate the scale-up and scale-down. That’s astory that I hear frequently, and I like to believethat that’s one of the key advantages.

I also see quite a few of our customers, giventhe times that we’re in, asking how they shouldprepare for when things get better again.Rapidly scaling at the moment may not be theirprimary concern, but they do see that there willbe a time in the future when that will need tohappen again, and they’re taking a look atwhether their architectures are horizontally

13 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

scalable and can exploit the advan-tages that cloud computing can givethem. For the next year to two, thosewill be the focus of a lot of our cus-

tomers: How they can automate theiroperations more so that they can have com-pletely hands-off operation. And how they canmake sure that their systems are horizontallyscalable so that they really benefit from the costeffectiveness that the cloud has to bring.

InformationWeek: Is there a limit in terms ofthe amount of scale that an enterprise cus-tomer should expect from a cloud service?We’re not going to see FedEx throw its package-tracking system into the cloud yet are we?

Vogels: I would love them to do that. I thinkthat would be great benefit to them. And this issort of a side comment, one of the drivers forwhy enterprises are considering clouds is thereis a very good financial driver—less cap ex,more op ex. Shifting investment from up-frontto usage-based investment is something thatCFOs at this moment clearly love. From aninnovation point of view, quite a number ofenterprises are considering moving some oftheir services into the cloud and then openingthem up such that they can become part of thecloud ecosystem, making it easy for Company Xto access their services in the cloud and thirdparties to extend the platform they’re building.Whether that will be FedEx or telcos, there’s awhole range of enterprise services that compa-nies are considering moving into the cloud,opening up their services so that they canbecome part of the ecosystem and so other peo-ple can build applications against that. Thisgoes beyond traditional enterprise services.Salesforce released a toolkit so customers caneasily run Salesforce applications using EC2and S3. Integration between these differentplatforms will become increasingly important.

InformationWeek: Let’s talk about multipleclouds working together, private clouds, andhybrid clouds. Do you see that businesseswould want to create private clouds in theirown data centers?

Vogels: It’s clear that a number of enterprisecustomers would like to have some of theadvantages of easy deployment and a uniformapplication container, where they can seam-lessly migrate between their own data centersand the external cloud. Whether you call that aprivate cloud, fine. It’s mostly important in thatcontext that there are a number of advantagesto deploying things in the cloud, the ease of useand flexibility that they would like to have intheir own data centers.

And there are shades of gray. I’ve heard cus-tomers talk about partner clouds, where theymay have an internal one for ease of develop-ment, maybe some shared services with part-ners and, after that, they move things into thepublic cloud. For most of these things, it’s real-ly day 1. Who will deliver this software, and canthis be done, is still under discussion.Who, how,what, and where that will happen. For us, it’simportant to focus on the kinds of services thatAmazon is delivering now, making sure that weexecute against performance, security, reliabili-ty, scale, and at a cost level that makes sense toour customers.

InformationWeek: As this grows with morecustomers serving more users using the infra-structure in more ways, does it begin to testwhat you and your colleagues know about scalein the sense that you’re doing things that havenever been done before?

Vogels: I think we are doing things that havenever been done before. The kinds of thingsthat we needed to do to develop the services

14 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

as they have been running now for the pasttwo years were already really unique. Wehaven’t only borrowed on principles of dis-tributed systems techniques out of research,we also needed to develop a lot of engineeringtechniques ourselves. Most of those wereproven through time to work really well with-in the Amazon e-commerce platform beforewe brought them into the AWS world.

Will this grow beyond that? I think there are anumber of standard principles that we can applyin terms of hierarchies, of loose coupling, ofprobabilistic techniques that I’m confident willserve us for quite a bit of time. When we devel-oped these services, we were looking ahead interms of what kind of scale we could achieve,and we’re not there yet. Even then, I’m confidentthat the choices we’ve made were the right ones.

TO BE RECOGNIZED as InformationWeek’s chief of the year, Ama-zon.com CTO Werner Vogels has obviously done a few things right.Following are five takeaways for other business technology pros.

First, the IT world needs more leaders who can cut through the hypearound cloud computing and articulate its business value and how itworks.Vogels explains the cloud in a way that people understand, includ-ing the technical nitty-gritty. There are still more cloud skeptics thanbelievers out there, and I’ll bet some of the skeptics work in your compa-ny. More IT pros need to be able to explain what’s different about cloudcomputing and where it might be applied to do new things at lower costs.

Second, customer service has moved beyond CRM and self-service Webapplications to something deeper and more significant, especially inthis tough economy. Vogels describes himself as an external-facingtechnologist and talks of “customer-oriented development” and “cus-tomer-oriented architectures.” Does your company think and act inthose terms? It should. It’s a way of ensuring that customer require-ments are deeply rooted in your company’s business technology strate-gy and infrastructure.

Third, surround yourself with good people. One of the first things Vogelsdid upon learning he had been selected as InformationWeek’s chief of theyear was to acknowledge, via a blog post titled “Teamwork,” the contribu-

Takeaways What We Can Learn From Werner Vogels

By John Foley

15 Jan. 9, 2009 © 2009 InformationWeek, Reproduction Prohibited

tions of his colleagues to the success of AmazonWeb Services. At InformationWeek, we alreadyhad figured that out, which is why our “Chief OfThe Year” magazine cover featured Vogelsalongside Andy Jassy, Charlie Bell, and AdamSelipsky, three of the other key executivesbehind AWS’s success.

Fourth, engage peers outside of your company.Vogels launched his All Things Distributed blogseven years ago while a researcher at CornellUniversity, and that’s made him accessible to aworldwide audience. I asked a software archi-tect with a cloud computing company if heknew Vogels, and he responded, “I’ve never methim, but I’ve read his blog for the past fiveyears.” Nearly 1,900 people follow Vogels on

Twitter. This type of community building is re-warding on a personal level, and it has all kindsof intangible benefits for your IT departmentand company as well.

Finally, big ideas and technical execution mustgo hand in hand. Vogels made the transitionfrom computer science researcher in academiato CTO of the world’s biggest e-retailer, andAmazon continues to push the envelope onscalability and reliability as it opens its datacenters to customers in the form of AmazonWeb Services. The challenge ahead for Vogels-and for other CTOs and CIOs-will be to ensurethat cloud computing works as promised asusage grows.