Placeholder Image

字幕列表 影片播放

  • TONY VOELLM: I head up the Google Cloud Security

  • Performance and Test Team.

  • I've been working on Cloud for over two

  • years now up in Seattle.

  • And don't let my title fool you.

  • It says "Engineering Manager," but managers

  • at Google are different.

  • I really do get my hands down in the code.

  • In fact, I checked something in last week, and about five

  • minutes later, somebody on my team fixed it and checked it

  • in again, so.

  • [LAUGHTER]

  • TONY VOELLM: I've done a bunch for Cloud.

  • Some of the things that are externally visible for Cloud

  • is I created the Google Compute Engine Units--

  • GCEUs.

  • And we'll talk more about what Google

  • Compute Engine is later.

  • Being the engineer I am, I thought, wow, GCEU, it's

  • really cool.

  • We should call that GQs, because who doesn't want GQs--

  • or, which engineer doesn't want to be GQ.

  • And now you know why I'm an engineer.

  • OK, great.

  • So here's what we're going to do.

  • We're going to talk about cloud computing.

  • So I'm going to take a quick step back, talk a little bit

  • about what cloud computing is.

  • I'll take you through some history about how we got to

  • where we are today in 2013.

  • This is a history of cloud computing.

  • Then I will definitely take you through the Google Cloud.

  • I'll give you pivots of what the Google Cloud looks like,

  • how we think about it, how the industry thinks about it.

  • I'll run through a series of demos.

  • And then, to prove I'm an engineer, I'm going to take

  • you through the pitfalls.

  • And this is how you know I'm not a marketing person,

  • because a, my slides are not pretty; and b, I'm going to

  • tell you why things may not work for you

  • from time to time.

  • And in the end, I'll just wrap it up with

  • the Team and questions.

  • So with that, let's dive in.

  • So what is cloud computing?

  • It's a good question.

  • I had this question two years ago.

  • And in fact, I probably still have this question today.

  • But one of things that I did to try to figure out what this

  • industry is, is I started looking different places.

  • I go do web research, Google search.

  • I looked up at How Stuff Works, and they had a really

  • interesting definition.

  • They talk about remote machines owned by another

  • company, and like maybe your email and word processing

  • would be out there.

  • And this seems like a really dated definition, but it's

  • actually still fairly accurate.

  • I did things like go out to conferences, because there are

  • several conferences that happen throughout the year

  • around cloud computing.

  • And in there, you'll see interesting terms, like Hadoop

  • elastic environments, grid software--

  • there's all these terms that start to pop up.

  • I even went out and started to survey my peers at Google,

  • like, what do you think cloud computing is?

  • And you can see down here, they start talking about, oh,

  • it's this computation that you can do in the cloud.

  • You don't have to worry about stuff.

  • So what I did is I said, OK, well, with any definition,

  • let's pull together the properties of cloud computing,

  • because there's certain things we're hearing often in these

  • definitions.

  • You know, one was nothing--

  • nobody cares where it is.

  • Everything's accessed over a network.

  • So cloud computing is something that

  • happens in the network.

  • That's sort of like the beginning of what cloud

  • computing is.

  • The second part is a really important

  • part, which is utility.

  • You can turn it on or off.

  • You pay for what you use.

  • And when it's not on, you turn it off.

  • It's like a light switch.

  • So if I want a database now, I have a database.

  • And if I don't want a database in five minutes from now, I

  • turn it off, and I'm not paying for it.

  • There's an elastic component where resources grow and

  • shrink on demand.

  • And you've seen this many times across like YouTube

  • scalability.

  • We broadcast the Olympics.

  • And whether one user is watching, or 8 million users

  • are watching, it all seems to work.

  • And so there's this elastic component on the cloud that.

  • grows and shrinks on demand.

  • For sure, it's programmable.

  • Programability is an important aspect of Cloud.

  • And then access control models.

  • So here's where there's some pieces we'll talk about later,

  • where like software as a service, versus you developing

  • your own code.

  • And platform and infrastructure as a service

  • layers, where access models are different.

  • There's some models where the end user owns the control, and

  • there's some where the person providing the

  • service owns the control.

  • But there's some method of controlling

  • access to data and resources.

  • And while not required, there's this thing that often

  • comes up, which is multi-tenant.

  • Your workloads run side-by-side

  • with somebody else's.

  • Some companies, this can be concerning--

  • why is Company A and Company B running on the same server?

  • We're competitors.

  • We don't want our data anywhere near each other.

  • And it's really the cloud providers that create this

  • partition or this barrier between those workloads.

  • So there's no flow of data from one to the other.

  • And this is called multi-tenancy.

  • So I tried to be really smart.

  • I'm like, OK, that sounds really good.

  • So let me give a definition of cloud computing that tries to

  • roll in all these sources of information.

  • And my definition is it's a set of programmable resources

  • that's pay-per-use.

  • It happens over a network.

  • It's elastic.

  • And it removes the developer from having to worry about the

  • hardware resources, the operating system, or all these

  • minutias that she doesn't want to worry about.

  • And she can just focus on delivering the application

  • that she wants.

  • And it's there, and it grows and shrinks on demand.

  • So that's my definition.

  • So that's cloud computing in a nutshell.

  • So here's a little mini-quiz.

  • So I'll just take a quick show hands.

  • I'm only going to go through a couple of questions to see if

  • things are already resonating with everybody here.

  • So is Gmail cloud computing?

  • Yes.

  • It could depend on perspective.

  • One perspective is Gmail is a hosted software service that

  • companies can buy.

  • You can buy Gmail.

  • And for some nominal fee per month, you can host your

  • mailbox there.

  • So on the upper end of software as a service--

  • this is where Google started--

  • yes, Gmail, I would say, is cloud computing.

  • Now what about hosting Python applications.

  • Is that--

  • show of hands-- is that cloud computing?

  • Yes.

  • Yeah, it has the programmable aspect we talked about.

  • It's elastic.

  • This is where Google App Engine enters the picture.

  • Here, let's try one more.

  • Are physical servers hosted by a hoster--

  • is that cloud computing?

  • Yeah, it depends.

  • Yeah, this is the one that gets sort of tricky is--

  • if you can on demand request these servers, and have them

  • go away on demand, then you sort of enter into the cloud

  • computing, versus just pure hosting.

  • We tried this whole thing in the '90s called application

  • hosting and server hosting.

  • And so, what's different today is this

  • whole elastic component.

  • And you might be thinking already, like,

  • why is all this important?

  • Why is this definition of cloud computing important?

  • And so I asked that question, too, actually.

  • Like, why is this important?

  • And so, I went out, and I found this "Forbes" article.

  • And I realized that not only can Google help you find

  • dating sites, we can also help you on your dates.

  • Where it says, "some fake it to make it," where there's 17%

  • of us, apparently, that on our first dates lie and tell

  • people we know what cloud computing is.

  • So I'm here to help you with your dating.

  • That's really what this is all about.

  • So cloud computing.

  • OK, so let's go back.

  • So let's go back and let's talk a

  • little bit about history.

  • So computers--

  • they've been there for a while.

  • Almost 60 plus years now.

  • Hard to believe we've lived with technology that long.

  • But in 1961, John McCarthy, also sort of dubbed as The

  • Father of AI, was living in a world where there were very

  • few computers relative to today.

  • There were 6,000 computers.

  • And he had this thought.

  • He said, computation one day will be a utility.

  • For him, this was really important, because there were

  • so few of them, and it was a constrained resource.

  • But it kind of made sense.

  • He was predicting the future, that one day we

  • will get to a utility.

  • Now the way we got there is very different

  • than perhaps he expected.

  • But nonetheless, some credit goes to him in terms of

  • thinking about, yes, computers could be a utility one day.

  • But before we could get there, something had to happen.

  • We had to have the internet happen.

  • We didn't have an internet in 1961.

  • We had some sort of point-to-point stuff.

  • People were trying things out.

  • But it wasn't until 1969 that ARPANET, the predecessor to

  • the internet, happened.

  • And at the time, transfer speed between computers was 50

  • kilobits a second.

  • And to give you an idea of what that actually means,

  • because it's hard to sometimes think about all these bits--

  • your cell phone's about 2000 times faster than computers

  • could talk to each other back in 1969.

  • Some other things had to happen.

  • We had to have the microprocessor revolution.

  • That started roughly in 1971.

  • And so after the microprocessor revolution, the

  • internet's kind of there.

  • We have these acoustic couplers and modems.

  • In the '80s, we have the BBS era.

  • This is where you have laptops.

  • They were called luggables.

  • You'd carry them around like this.

  • They were like 67 pounds, or something like that.

  • This is an example, it's the Kaypro.

  • But something interesting happened where services

  • started to show up, like CompuServe and GENie.

  • And you would dial up, and for $0.10 a minute, you could

  • access this super awesome service.

  • And all you wanted to do was find out where

  • the free BBS was.

  • So you could go connect to this free thing, and start to

  • share files or data.

  • And so you'd go connect up to something, like Koala Country,

  • which was free.

  • And so there was this whole sort of culture around the

  • internet starting to happen--

  • connectivity, sharing.

  • Data should be universally accessible.

  • So roll time forward a little bit.

  • So from the '80s, if that was our BBS era, we really needed

  • the '90s to happen, because these are pretty important to

  • where we are today.

  • So in '93, if you can believe it, this

  • is the Mosaic browser.

  • It was first to enter in the scene, and make the internet a

  • lot more accessible.

  • Prior to that, we all ran LAN line tools, like Archie and

  • all these things he hope to forget one day.

  • And users started showing up.

  • And there were a few users.

  • There were about 25 million users connected up to the

  • internet in '93 roughly.

  • By '95, something interesting happened--

  • we had 6 million hosts serving data on the internet.

  • And so we had 25 million users, and so you have to

  • figure the users have grown a lot.

  • And you'll see this on the next slide.

  • Now I'm sure I'm not going to pronounce his name correctly,

  • so Ramnath K. Chellappa, I apologize if you ever watch

  • this video.

  • But this really smart guy was out there, and gave what's

  • often deemed to be the first definition of cloud computing.

  • Where Ramnath, he says, cloud computing-- he basically says

  • that computing is going to be a paradigm where computing is

  • determined not by technology, but by economic decisions.

  • Yeah, look at that.

  • Isn't that surprising--

  • economic decisions.

  • And when we were all really flush with cash in the late

  • '90s, nobody was too worried about server rooms.

  • And the fact that 10% of the CPUs we all owned were

  • actually being utilized, 90% unused was just fine.

  • We had plenty of money to run all of the air

  • conditioners and power.

  • But computing resources grew.

  • 2000, we had 72 million hosts.

  • And then by 2006, we had almost four million hosts on

  • the internet.

  • And that is astonishing growth, if you think of how

  • fast that happened.

  • And then the number of users that are using these also

  • grew, which obviously grew the servers.

  • And we had like 461 million users in 2001 using internet.

  • Pretty astonishing.

  • So 2006 was kind of a magic year.

  • This was the year where people like Eric Schmidt, our

  • executive chairman of the board for Google--

  • super smart guy--

  • started to piece things together, and started

  • connecting together these ideas of

  • software as a service.

  • In fact, by all my reading, he was deemed one of the first

  • people to ever use the term "cloud computing." Because as

  • I mentioned before, in the '90s, we tried app hosting.

  • Nobody wanted to hand over their data.

  • Nobody wanted us to run their applications.

  • But somehow, magically in 2006, when you called it

  • "cloud computing," everybody seemed a lot more comfortable.

  • At least, beginning to be comfortable with this idea.

  • And then also in 2006, Amazon launched Amazon Web Services.

  • And this was the first provider that was out there.

  • And I love this picture here, because it's kind of like

  • Amazon's the baby--

  • bright-eyed, I'm new to the party.

  • And the older sister is really looking there, like hm, didn't

  • we try this in the '90s called application hosting?

  • And as the way with technology, sometimes if you

  • know nothing, you're going to win, because you're just sort

  • of looking at it from a different perspective.

  • There's this whole notion of innovation cycles that you can

  • read about that happen over and over again in technology.

  • So when did Google enter the picture?

  • So Google entered into this picture in 2008.

  • We launched two things kind of simultaneously, hitting

  • different parts of helping to make the cloud happen.

  • One part is we launched Chrome beta, which is astonishing,

  • because just recently it looks like Chrome has become the

  • most popular browser in the world.

  • And that's in like four years, which is astonishing growth.

  • But Chrome came out in beta.

  • And Google App Engine-- this place where you can host

  • Python and Java applications--

  • came out.

  • So we're trying to tackle the cloud from two parts.

  • One was the usability end, where we're making the

  • internet easier to use with the browser.

  • And we were starting to create a platform where you could

  • actually create applications and host

  • applications on the internet.

  • In 2010, we launched actually Google Cloud Storage.

  • I can't tell you the number of tera x--

  • petabytes--

  • whatever happen to be in cloud storage.

  • I can tell you it's sufficiently big.

  • So Google Cloud Storage is an object store in the cloud.

  • It stores a ton of data.

  • There are many, many places you probably hit this, whether

  • you go to a website and images show up, or things like that.

  • But we had launched that service actually in 2010.

  • Then 2011, we took App Engine out of beta.

  • This is where you could tell Google was very serious in

  • this market.

  • We put down an SLA where we said, hey, we will guarantee

  • certain sets of APIs will be around.

  • We'll have a deprecation policy, and so forth.

  • Very important to all of us as developers and engineers that,

  • if you're going to build it, it will still be there a year

  • from now if you're betting your business on it.

  • And then in 2012, last year, we shipped a couple of things.

  • We shipped something called Google BigQuery, which is this

  • amazing query technology.

  • You can query terabytes of data in a second.

  • I know, that's astonishing.

  • We have hosted MySQL in the cloud, and I'll talk more

  • about this later.

  • And we launched Google Compute Engine, which is our virtual

  • machine offering in the cloud.

  • And I noted here that it's limited preview.

  • Google differentiates here.

  • Limited preview means there are workloads in production.

  • People are paying us.

  • And we are happy to have you come and try Compute Engine.

  • But we're just not going to open up the doors

  • fully open just yet.

  • But we are happy to help you use it.

  • Beta is different.

  • Beta means we're not totally sure.

  • Maybe it's going to change, and so forth.

  • And this is something that has kind of confused the market a

  • little bit, where people think, oh, you have Compute

  • Engine but it's in beta.

  • Actually, it's in limited preview.

  • People are paying us money to use the service.

  • So we go through this whole history, and then I learn

  • something new from Eric.

  • So we learned all this technology stuff, and I

  • thought I was really smart with my

  • definition of cloud computing.

  • But it takes somebody like Eric to really bring it home.

  • He goes, you know, I'm not sure anybody knows what cloud

  • computing is, but I do know one thing--

  • it's a marketing term.

  • So there you go.

  • It's a marketing term.

  • So if anybody asks you what cloud computing is, you can

  • confidently now answer--

  • it's important to my dating, and it's a marketing term.

  • So let's talk about the stack.

  • And oftentimes, cloud computing is

  • thought of as a stack.

  • A set of sort of tiered services, from the lower

  • level, which is called infrastructure as a service--

  • IAAS--

  • which is maximal flexibility and maximal complexity.

  • This means you're your own administrator.

  • You have to have your own backup plans, retention plans.

  • It's a lot of work.

  • In fact, a lot of times, it's sort of perceived as the IAAS

  • layer is like taking the server from underneath your

  • desk, and sticking it up in the cloud.

  • Whatever that means.

  • AUDIENCE: [INAUDIBLE]

  • TONY VOELLM: Hold that thought.

  • Ask me that later.

  • Somebody said without their privacy.

  • Actually, I disagree with that, but I'll tell you why I

  • disagree with that later.

  • And I'll show you some of the things that Google is doing to

  • be very, very, very careful about end user data.

  • And I'll talk about some of the messages around that.

  • But infrastructure as a service is really the thing

  • we're familiar with.

  • It's interesting, whenever I've given talks about this

  • next layer up, called platform as a service, a lot of people

  • ask me where the server is.

  • The platform as a service tier tries to remove the user from

  • this infrastructure layer, but you have a different

  • programming model.

  • And in fact, it's very hard for developers to think in

  • this model.

  • But if you can get into this model, it's very powerful.

  • And then at the top of this tier--

  • I'll go through each of these little bit more-- is software

  • as a service.

  • This is where you have things like Google Drive, Google

  • Docs, Google Spreadsheets, Gmail, YouTube, all these

  • things that you can use.

  • At the service layer, you can even build on top of them.

  • They are programmable.

  • You can put JavaScript into your Google Spreadsheets.

  • And you can pull data from the lower tiers in the

  • infrastructure layer up.

  • And there's all these examples out there that

  • allow you to do this.

  • It's super cool.

  • But that's the software as a service tier at the top.

  • What's interesting about SAAS is it's less flexible, but

  • it's lower complexity.

  • So it's easier to use, but you don't have perhaps all the

  • controls you want.

  • So there's trade-offs at each of these layers.

  • So let's talk about each one just a little bit more.

  • And then I will take you into how the Google pieces plug in

  • to these tiers.

  • So infrastructure as a service layer.

  • As I mentioned, these are the building blocks.

  • These are the pieces that we know today where we talk about

  • storage, like I have disks.

  • We talk about computing resources.

  • I have machines.

  • We talk about databases.

  • I have databases.

  • And this is the layer where a very thin layer will come on

  • top of it, and allows you to access it over the cloud.

  • But it looks familiar.

  • I'll show you an example later of Compute Engine where we can

  • log into Google Compute Engine.

  • It's a machine in the cloud.

  • I get a UNIX prompt, and I can do stuff like configure web

  • servers, or run computation workloads-- whatever I

  • want to do on it.

  • There are some terms here, too, that you watch out for at

  • the infrastructure layer.

  • Because there is all this decision-making to be made,

  • there's this notion of zones.

  • Like, where are these resources?

  • And so, are they in the US?

  • Are they in Europe?

  • Where do I want them to be?

  • How do I have disaster recovery?

  • And so, yes, all of this complexity suddenly enters in

  • at this layer.

  • Platform as a service is different.

  • So this is things like Google App Engine, Salesforce.com,

  • Amazon Elastic Beanstalk, and so forth.

  • There's lots of examples of these out there.

  • But this is a different model.

  • This is where you can't just take the code that you've

  • always written and just suddenly put it in the cloud.

  • Some amount of rewrite, some amount of rework is going to

  • happen in the platform layer.

  • But what you get for it is, you get authentication models.

  • Somebody else will handle verifying who the user is

  • who's using your application.

  • You get services like Data Store, AKA Google BigTable,

  • where you have infinite key value pair look-up, and it's

  • redundant and reliable.

  • You have things like versioning control, where you

  • can do experiments.

  • Like I want to deploy only 1% of my users onto my new

  • software, versus 99% on my old software.

  • And as I'm confident that my new bits work, I can actually

  • kind of move the dial over.

  • And these systems in the background are automatically

  • moving users to these

  • different servers and services.

  • So you've given up some flexibility around

  • configuration and control, but you've gained a whole lot more

  • around scalability.

  • And then there are obviously test services and payment

  • systems, and things that you could

  • integrate in at this level.

  • Is all of this resonating so far with everybody?

  • OK, good.

  • Then we have the last layer, which is the software as a

  • service tier.

  • This is actually where Google started.

  • It's very interesting.

  • Different companies have taken different approaches to cloud.

  • Google has sort of come from the top down.

  • We started with software as a service, things like Gmail,

  • and search, and all these other things.

  • Moved down into the platform layer with Google App Engine,

  • and further down with these infrastructure

  • components I'll show.

  • Others in the industry started differently, and fundamentally

  • their design points are different, where they started

  • at the infrastructure layer, and moved up.

  • And you can sort of see this permeate through their

  • architectures.

  • And then, even a third one, which I can't name, that sort

  • of started in the middle, and sort of moved out from the

  • platform layer, kind of up and down, which is probably

  • perhaps one of the hardest points to start at.

  • So, software as a service layer--

  • you have a model around computation and code, but you

  • get capabilities like I want to run a query on a terabyte

  • of data, but the data has to be structured

  • a particular way.

  • For example, you have things like Google Translation

  • Service, and other things at this layer.

  • So reduced complexity--

  • the least complex--

  • but the least flexible.

  • So how does this look for Google?

  • And how do we sort of shape across these tiers?

  • Just one quick segue.

  • The language of cloud.

  • There's basically two languages of cloud.

  • There is REST and JSON, and SOAP and XML.

  • Every provider out there speaks one

  • of these two languages.

  • Google made a set of choices.

  • I'll talk about those in a minute.

  • Really, we're on REST and JSON.

  • We wanted simplicity, and less statefulness.

  • But there are different languages of the cloud.

  • These are important.

  • These will impact how you actually

  • develop your software.

  • So Google--

  • where are we?

  • Here's how Google layers across these services.

  • Yes, I realize this is a little bit of an eye chart for

  • those in the back of the room.

  • So I hope you have 20/10 vision.

  • But I will walk you through these, and give you a little

  • bit of an explanation of these technologies.

  • And certainly, we can talk more about these later.

  • I actually have demos of a couple of them, so I will

  • definitely be doing the demos.

  • So where do we start?

  • So at the bottom tier, we have the

  • infrastructure as a service.

  • So we have Google Cloud SQL.

  • This is hosted MySQL in the cloud.

  • So if you use MySQL, SQL Server, any other sort of

  • Oracle, SQL-based language, MySQL will look

  • very familiar to you.

  • The data types are familiar.

  • The syntax is familiar.

  • SQL-99, or whatever type syntax.

  • And in the cloud, you get this service.

  • And so whether you're programming in App Engine,

  • Compute Engine, or even outside of the Google Cloud,

  • you have a database.

  • It looks like a database.

  • And you can run queries, put data in, and run transactions,

  • and so forth.

  • What's unique about the Google Cloud SQL that's different

  • than others is our monetization model-- the way

  • you pay for it.

  • We have two models.

  • One is you can sort of get the subscription model.

  • The other is you can pay per use.

  • And so, for like $0.10 an hour, you could have 100

  • gigabyte database, or something like that.

  • And when you're not using it, it automatically

  • shuts down for you.

  • When you go and you run a query, we'll automatically

  • spin up your database, attach you to it, and

  • let your query complete.

  • I don't know of anybody else in the industry that gives you

  • that level flexibility for their SQL service.

  • So that's actually pretty cool.

  • By the way, I spent my whole career in these infrastructure

  • layers all the way up, so what's totally cool to me is

  • not always cool to others.

  • Like, that's my nerd humor.

  • I spent my whole time doing operating systems.

  • I think operating systems are cool.

  • Can we save one bit?

  • Most people are like, one bit?

  • What I do I even care, one bit?

  • I got more bits on my phone than like all

  • that existed in 1961.

  • Anyway.

  • So to the storage layer.

  • So Google Cloud Storage has been around for a while.

  • It's growing.

  • We've added capabilities, like static web hosting into the

  • service, so you can serve web pages out of here.

  • We are one of the first companies to support CORs.

  • This is like cross origin request, or whatever, where

  • you can say, oh, this server can request data from Google

  • Cloud Storage, and make it look as though it's coming

  • from my own site.

  • And this fixes a lot of permissions problems when you

  • build applications on the internet.

  • We have some really cool new features that I'm probably not

  • supposed to talk about that will be coming down the road

  • here with this, which makes it a lot easier.

  • We added a file API into App Engine, so it's easier to

  • manipulate files.

  • In Google Cloud Storage, we continue to innovate, quarter

  • after quarter, week after week.

  • New bits are shipped.

  • It's astonishing that we'll ship basically new kernels

  • every week in Compute Engine.

  • I don't know anybody else that does that at that speed.

  • And so with that, Compute Engine is our virtual machine

  • offering in the cloud.

  • We offer a couple different versions of OSes.

  • They look very familiar, and we'll log in.

  • You have different images you can choose.

  • At some point, we'll let you upload your own kernels, and

  • very interesting stuff.

  • But we have data centers all around the world that allow

  • you to host VMs, and I'll show you that.

  • So that's our infrastructure as a service layer.

  • Very cool, growing.

  • There's more that will be coming in the coming weeks,

  • months, and years in that layer.

  • The platform as a service is like Google App Engine, which

  • has been around for while.

  • And we continue to innovate services.

  • And we're always trying to find ways to better blend

  • programmability inside of the App Engine with Compute

  • Engine, so you can start to move your code more flexibly

  • between the different services.

  • We're getting better at that.

  • App Engine has different models of computation.

  • You can host Python code, Java code, Go code up in the cloud.

  • It'll scale up or scale down.

  • It integrates directly with the services down below.

  • So whether you're pulling objects or queries, you can

  • also have workloads run over inside of Compute Engine.

  • I'll show you a demo that uses Google App Engine as a front

  • end for hosting a website, and the interaction of the data

  • model with Compute Engine where we're doing some

  • scalable computation.

  • We have some other things that are coming.

  • I discovered that there is this thing out there called

  • Google End Points.

  • It's in Trusted Tester mode.

  • I thought it was safe to put it on this slide, because I

  • typed this in and it showed up on Google, so it must be OK.

  • So that's the engineering perspective, by the way.

  • But this is where we're trying to make it easier for you to

  • develop services.

  • So we give you this API layer that handles things like

  • authorization and authentication and payments,

  • and allows mobile phones and end browsers to send requests

  • into your service.

  • You handle the request, and you send back a response.

  • And so, if you've used Google Maps API, or these other APIs,

  • they are actually this type of Google End Points layer that

  • we use internally.

  • And now, we're going to offer that externally.

  • And then moving up the stack into the software as a service

  • tier, we have Google BigQuery.

  • And this is our service where you can upload terabytes of

  • data, run a SQL query, and the answer will

  • come back in seconds.

  • The first time you do this, it's actually quite

  • astonishing.

  • I grew up building SQL Server technology, and things like

  • that in years past.

  • And databases could never do this.

  • I was astonished the first time I did it.

  • And then, most people are probably familiar with thing

  • like Translate.

  • If you build front end web pages, and you want to get a

  • Google assessment on the performance of your page and

  • how to optimize it, we have services like Google PageSpeed

  • you can actually go use.

  • We have a set of sort of artificial brain models.

  • Anybody see that?

  • We took like 10,000 CPUs, went out and trained it on every

  • video on YouTube, and what did it learn to do?

  • Anybody know?

  • AUDIENCE: Cats.

  • TONY VOELLM: Yeah, it learned all about cats.

  • I'm a dog guy.

  • I was kind of offended.

  • But we have the Google Prediction API, which does a

  • lot of that.

  • It's these AI models in the cloud where you can train them

  • with sets of data, ask for answers, and then it will give

  • you your prediction of what's there.

  • It also does linear regression models, and lots of other

  • interesting things.

  • And of course, at the top is Docs.

  • I mentioned Docs on the slide because Docs can integrate

  • with all these other tiers.

  • To this Google End Points set of APIs that we expose out.

  • So everything we expose, we have is a RESTful API.

  • And Docs can actually program around these, whether it's in

  • the spreadsheets, or forms, or other areas.

  • It's actually pretty cool.

  • So how does Google view the cloud?

  • So this is-- by the way, this last side, this is how the

  • industry views cloud.

  • They think of it as this tiered model of

  • infrastructure, platform, and software as a service.

  • Google, we kind of think of it differently, though.

  • We think of it as apps.

  • There are a set of things that we built, that you use.

  • And then we have these other buckets of how we think about

  • the cloud platform.

  • We certainly think about how do you build your front end

  • stuff, like websites and apps, whether it's in your mobile

  • phone or Android apps.

  • Whatever they are, we kind of have this part that we think

  • about as apps.

  • We think about your data.

  • Data is very important.

  • And this is where privacy of your data

  • is extremely important.

  • We are one of the first cloud providers out there to have

  • encryption at [? REST ?].

  • So all data in Google Compute Engine is encrypted.

  • And we're taking that same mean and value to the rest of

  • our Google Cloud.

  • So that basically means there's a key.

  • We don't really have your key.

  • You lose your key, you're hosed because you're not going

  • to get your data back.

  • But there's probably one engineer who's highly audited

  • somewhere that could get the key.

  • But it's very, very, very controlled.

  • So we think about storing your data.

  • We think about computation at scale.

  • We're very interested in scientific workloads.

  • And we've done all kinds of interesting things, like

  • genomics, video rendering, and so forth.

  • There's been some really great articles that are written

  • about our computational capabilities, and

  • those are out there.

  • And then, of course, analyzing your big

  • data, which is a problem.

  • If you have terabytes of data, actually how

  • do you analyze it?

  • There's another interesting problem, which is I have

  • terabytes of data-- how do I get it into the Google Cloud?

  • Turns out, we--

  • I know, I don't really want to admit it.

  • You can send us a disk.

  • Not everybody's internet is like Google's internet.

  • But once it's in the Google Cloud, it's extremely fast.

  • But yes, if you send us a disk, we will actually

  • uploaded into the Cloud for you.

  • So this is more of the Google model.

  • We think about apps, computation, data, and

  • analyzing the data.

  • So what did we do to build Cloud?

  • So there are a bunch of principles and properties that

  • we've really held to, which are security.

  • And by security, I mean security and privacy.

  • We are highly, highly audited.

  • There's nothing that I can think of that can be touched

  • in cloud that isn't audited.

  • So if somebody has touched a piece of data, it's known who

  • it was, when they did it, how long they touched it, for what

  • purpose they did it.

  • And we have a very tight set of audited controls.

  • And you'll see more about audit processes and other

  • things later in the year.

  • We have a strong auth model.

  • So we chose this auth model called OAUTH2.

  • It's a little complicated from the developer end, which is

  • why we have upload to GitHub a bunch of code samples, like

  • how to program the Google Cloud.

  • So if you want to use cloud SQL or Compute Engine, or

  • whatever else, we put a bunch of example apps out there.

  • But there's some cool properties of OAUTH2, which is

  • like revocable access.

  • If you give somebody an access token, you could actually go

  • in and revoke access to that access token.

  • Which is something that's important.

  • Because once you're given somebody access to an app, you

  • want to be able to revoke it at some point.

  • You don't want them to have-- they've got your key, and they

  • always have your key.

  • That that's a problem.

  • OAUTH2 has other things, like time limits.

  • Like, you may look at the data for the next five minutes, and

  • after that, you can't get it.

  • And so you see things like that show up in our storage

  • tier where we'll give you a lease-- five minutes to pull

  • the imagine, and after the five minutes, you

  • can't see it anymore.

  • And so, while there's some slight complexity for

  • developers there, there's a lot of power that comes for

  • security around that.

  • Consistency was a second meme we've held.

  • I'm actually very happy to see the blogs picking up on this,

  • where they go out and they say things like, Google Compute

  • Engine is the most consistently performing

  • virtual machine in the cloud.

  • If I'm getting 100 IOPS to second the disk now, I see it

  • now, tomorrow, no matter how much seems to be running, the

  • workload seems to run exactly the same time and time again.

  • That is extremely hard to do, but something we value a lot

  • is consistency.

  • Open and flexible.

  • We're giving all the examples out.

  • In fact, all of our UI that I'm going to show you later,

  • you can develop all these UI yourself.

  • This is something that we've held as a principle.

  • We want people to build on cloud from all layers, whether

  • it's the management orchestration

  • layer, all the way up.

  • And then, yes, we have code.

  • And I'm a security guy, so I tried to use some

  • leetspeak on you there.

  • We haz code.

  • Sorry.

  • Tried to trick you.

  • And then proven.

  • So we actually do use all these services today ourselves

  • in production.

  • Google's one of the few companies I've ever been at

  • where we actually use all the things that you use.

  • We don't use anything different.

  • We use the same things, which is very cool.

  • So we prove the technology, and we won't ship it until

  • it's ready.

  • So I sometimes get the question,

  • why the Google Cloud?

  • Well, that sounds all great, Tony.

  • You have security.

  • You have consistency.

  • You have all these tiers, offerings everywhere.

  • I can write my mobile apps, and host them in

  • the back of the cloud.

  • Why Google, though?

  • Well, we know something about data centers.

  • If you haven't gone and done the virtual tour of the Google

  • data centers, you should.

  • Go out to Google, type "Google data centers." It's

  • phenomenal.

  • We have powerful machines, network services, geo-located

  • all over the world.

  • We've been running these services

  • obviously for a long time.

  • And we're really taking the best of what we have, and

  • really allowing the world to actually use

  • Google as their computer.

  • And so why Google?

  • Because we care.

  • We listen.

  • We take the feedback.

  • Every time I've gotten feedback, within weeks or

  • months, we've actually acted on the feedback.

  • So as you use the Google Cloud,

  • please do send us feedback.

  • We really do listen.

  • So why Google?

  • Because we listen, and we have the experience to run this big

  • data center.

  • So, some demos.

  • This is one of my favorite parts here.

  • So I'm going to do a couple demos.

  • First, I'm going to take you to the Cloud Console.

  • I'm going to click on a web page.

  • That's going to be a very sophisticated demo.

  • But then I'll take you through very briefly

  • some App Engine demos.

  • But I'm going to spend a bit more time on Compute Engine,

  • and show you some of the power of what fully generalized

  • computation in the cloud looks like.

  • We will spin virtual machines up in Europe.

  • So while we're getting close to sleeping, they're wide

  • awake, and so we're going to go spin up some

  • machines over there.

  • And then, I'll just kind of round things out after this.

  • So with that, the Cloud Console.

  • And by the way, I closed out my email.

  • So if anything personal show up, start laughing so I know

  • to close the window very fast.

  • Which is very likely to happen.

  • So Cloud.

  • It's all together now.

  • So if you go to cloud.google.com, you can

  • actually try all these things that I've been

  • talking about here.

  • You can see different sets of customers that we have.

  • You can click into computing at scale.

  • You can go and learn more.

  • You can dig through this.

  • I'm not going to spend very much time on this.

  • In fact, I'm just going to go dive into some

  • of the other demos.

  • But we have an amazing amount of help out here.

  • If there are examples you don't see, example you want--

  • really, like I said, please send us feedback, and I'll

  • take you through it.

  • So cloud.google.com.

  • That's your starting point for all of Cloud.

  • So if you walk out of here with anything, three things--

  • know we're helping you with your dating, because you need

  • to know this stuff.

  • cloud.google.com is what you want.

  • And we care about security and privacy.

  • So that's demo number one.

  • There's cloud.google.com Which a year ago didn't even exist.

  • So let me go back here.

  • And you can tell I've done demos before, because just in

  • case it didn't come up-- there it is.

  • OK, so Cloud.

  • So let's talk about App Engine.

  • App Engine is our platform as a service tier at Google.

  • This is the thing that automatically scales up and

  • scales down dynamically for you.

  • It's where you can host your data.

  • It has lots of services you can use to build interesting

  • things like--

  • we have an XMPP service where you can go do live chat.

  • And at the back of your chat, you can have a robot trying to

  • predict what the answer is.

  • And so, you've probably experienced this somewhere on

  • the internet where you're typing into, oh, I need help

  • with my flowers.

  • And it comes back, and it says, oh, here, have you done

  • these things?

  • It's very likely it could be an App Engine app using this

  • XMPP protocol out there.

  • App Engine, as I said, started out in 2008.

  • And you can just see the phenomenal growth that has

  • happened with App Engine.

  • And year after year after year after year, we've added

  • services to make it easier to develop interesting

  • applications that are out there.

  • We have customers across many different

  • tiers for App Engine.

  • And this is to give you an idea of the scope of how App

  • Engine is used.

  • Clearly, line of business apps at the bottom.

  • Business apps--

  • lots of businesses use us.

  • In fact, you've probably hit an App Engine app

  • and didn't know it.

  • At the consumer tier, we have companies like Khan Academy.

  • And there's a bunch of other really cool ones I wish I

  • could have talked about, but I can't.

  • But mobile gaming is out there.

  • Lots of mobile gaming.

  • In fact, the biggest mobile gaming vendors out there you

  • can think are all using the Google Cloud in one way,

  • shape, or fashion.

  • And then, on the mobile tier, we have news

  • readers, like Simperium.

  • You have Pulse, and others.

  • And so, App Engine obviously has big usage, in terms of the

  • number of users.

  • There's something like 7.5 billion hits over a million

  • applications on App Engine being hosted 24 by 7.

  • And we have a 99.95% reliability.

  • And we're trying to drive that up all the time.

  • So that's App Engine.

  • So let me go pop in and do my two little sample apps.

  • If you go, and you look, if you google "appspot.com," "app

  • spot," or actually "all App Engine hosted apps." So if you

  • want to see who else is sort of out hosting apps on App

  • Engine, you can go do an app spot search.

  • I found this one, which is kind of geeky, because we're

  • all developers here.

  • Which is called shell.appspot.com, which gives

  • you a thin interface to Python, so you can program

  • Python in the cloud, and do like everybody would want to

  • do, which is--

  • oh, I have a typo.

  • That's me.

  • I'm not a very--

  • whew.

  • There.

  • [? Printonia ?].

  • I have taken over the internet.

  • Thank you.

  • Thank you.

  • But that's an example app.

  • You may have apps that are programming apps.

  • There's other obviously much more sophisticated vendors out

  • there doing cloud development, and they're using App Engine.

  • I found this other site, which was called Beat the Boot.

  • This was a Chrome site that Google had built where you

  • could actually go and try to play games to see how fast you

  • can do things.

  • How much faster you were than Chrome in booting.

  • And I was failing miserably.

  • But this is all hosted on App Engine.

  • The back end data tier, the front end presentation.

  • And I believe that this one was I'm supposed to race it in

  • brushing teeth or something like that.

  • And clearly, Chrome beat me.

  • So there you have it.

  • You can write games on App Engine.

  • Yes, it's very sophisticated.

  • OK.

  • And then the demos that I wanted to spend a little bit

  • more time on are actually Compute Engine.

  • And so, Compute Engine is one of our newest offerings.

  • We launched this last year at Google I/O.

  • It's in limited preview.

  • And so a few short facts about Compute Engine, and then I'll

  • do the demos.

  • Which is these are some things that people are saying about

  • Compute Engine here.

  • And what I like about this site here is this is like a

  • genomics site.

  • And if you go back and watch the Google I/O launch, it

  • shows you differences between--

  • you're trying to do this locally on your work station

  • where gene association takes about 30 days per association

  • on powerful workstations in your office.

  • And on Compute Engine, we're doing associations like every

  • five seconds.

  • So it fundamentally changes the way research is done.

  • In the last six months since we shipped this, we've beaten

  • the TeraSort record on Compute Engine.

  • So a partner of ours, MapR, they did this analysis where

  • they are looking at like, why would you want

  • to move to the cloud?

  • Like, what's the economics of the cloud?

  • How is it different?

  • And so, they went out and they did a TeraSort.

  • And they said, well, what would it

  • take to build a TeraSort?

  • Like if I was going to go build a set of machines, what

  • would it look like?

  • And so they did this one thing where it's like, OK, we'll

  • need roughly 1,500 servers, 12,000 cores.

  • It takes a lot of planning and investment.

  • It's going to take us about $6 million roughly to build this

  • data center that, in a minute, could sort your data.

  • Or, optionally, in 53 seconds, you could sort your data on

  • Compute Engine.

  • And at a cost of $500.

  • And by the way, the $500 will let you do it 60 times.

  • Because it's like 13 and 1/2 cents a minute per core.

  • Multiply it out.

  • You pay by the hour there.

  • So you could sort 60 times for or $554.

  • Cloud is completely changing the economics of how we think

  • about building software, bar none.

  • Here's a video of the TeraSort happening.

  • This was produced by MapR.

  • And what they're doing here is they're actually kicking off a

  • script that's programming to the REST APIs on the back end

  • for Compute Engine.

  • They're pushing software into the nodes, telling the

  • MapReduce to go do the sort.

  • The red are CPUs that are fully engaged.

  • The green are the idle.

  • And this is what a sort actually looks like,

  • computational-wise.

  • It's pretty cool.

  • It almost looks like the game [? alive ?], where all the

  • CPUs are engaged, and subsets and subsets and subsets.

  • And there you have it--

  • a TeraSort in seconds.

  • So let me go give some actual live demos here quickly.

  • So one is this.

  • So we actually have a UI for spinning up and shutting down

  • machines in Compute Engine.

  • So here, what I'm going to do is I'm

  • actually create an instance.

  • I'm going to call it "NYC," because it's New York.

  • Like I said, I have a couple choices of where I can spin up

  • machines in the world.

  • I can go to the middle of the US, the east coast, or

  • somewhere in Europe.

  • So I'm going to go to Europe.

  • I'm going to ask for a two-core CPU.

  • And I'm going to tell it to go kick that off.

  • And so one of the things that you'll see in Compute Engine

  • is the first stage of booting a virtual machine is called

  • provisioning.

  • This is where we go look at the data center, figure out

  • what physical host we can host the virtual machine on.

  • Once we figure out where it can be placed, we place it.

  • Then we go into this mode which is called staging.

  • And that's what you see happening right now.

  • This is like, oh, I have to boot an image.

  • It's an operating system that I need.

  • Let me pull this operating system down to my host.

  • Let me tell my machine about the operating system.

  • And then, let me actually started booting

  • the operating system.

  • Now what's interesting about Europe is the image takes a

  • little bit longer, because currently the images are

  • mostly hosted in other places than next

  • to those data centers.

  • And so, it's booting.

  • And then now, it's transitioned into this mode

  • called Running.

  • This machine [? b, ?] this machine is on.

  • It's powered up now.

  • So somewhere in Europe right now, I started up a machine.

  • So I just used my $0.14.

  • And so, I'm going to make use of my $0.14 here.

  • So I'm going to switch into the command line here.

  • So let me do one thing.

  • I will attempt not to type my password in for you.

  • OK, so what I'm doing here is, I actually have a workstation.

  • I have a developer workstation in Kirkland that I actually

  • logged into here.

  • Because I needed some gateway to get into my machine.

  • Like I said, this is infrastructure

  • as a service layer.

  • Somehow, you have to get into the machine.

  • And you can get an x terminal to it, or whatever else.

  • I'm a developer.

  • I'm going to go to the command line.

  • So what I'm going to do is, I'm actually going to go SSH

  • into this machine that's in Europe.

  • And I'm going to use the set of local command line tools

  • that we have for Compute Engine.

  • So we have command line tools for all

  • these different services.

  • This is called gcutil So Google Compute util.

  • I have a project.

  • And so one of the first things I'm going to do is I'm going

  • to say, list all the instances that are out

  • running for me as a user.

  • And I wish I could--

  • AUDIENCE: [INAUDIBLE]

  • TONY VOELLM: Yeah, sorry.

  • Actually, what it really doesn't

  • like is the fact that--

  • AUDIENCE: [INAUDIBLE]

  • TONY VOELLM: I have two mistakes here.

  • No pressure, no pressure--

  • live typing.

  • OK, now I'm definitely sure I'm running.

  • And I know it's a little hard to see in the back, but here--

  • what you can see is I actually have a machine that is booted.

  • It is this [INAUDIBLE]

  • machine, two cores.

  • Tells you what image it is.

  • Gives me an external IP address.

  • One of the things I can do with the tool, though, is I

  • can just say, ssh into NYC, which is my machine.

  • And the tool will automatically look up the IP

  • address, and get me in.

  • And so, here I am actually logged in to a server in

  • Europe, running in a Google data center--

  • which like I said, for an infrastructure

  • guy, is very cool.

  • So, yay, it actually worked.

  • So now let me give you a more visual kind of depiction of

  • what this looks like.

  • So for Compute Engine, I showed you the front end UI.

  • So there's this UI here.

  • I can click on the machine, NYC.

  • It tells me where it is.

  • I showed it to you running inside the command line.

  • So I could actually get command line access, and now I

  • can configure apps, get, run all my software, have full

  • programmability.

  • But here's a slightly different demo.

  • What I'm going to do here is I'm going to spin up 16

  • virtual machines.

  • And this is an App Engine app front end hosted application.

  • And what it will do is, as it starts to go through the

  • provisioning stage, you'll see green.

  • So we have gone off.

  • We told 16 servers to boot.

  • We're 13 seconds in.

  • Machines are already in this provisioning stage, then they

  • get the staging.

  • And then when they're green, that means they're fully up

  • and ready to go.

  • My co-workers will often show this with like 1,000 nodes.

  • And it takes about the same amount of time, actually, to

  • start 1,000 nodes.

  • Roughly under a minute, we could start up 1,000 machines.

  • So there's now 12, 16 machines out there that are running on

  • my behalf somewhere in the world.

  • And what I'm going to do is I'm going to show you what

  • this looks like visually to have this powerful computation

  • at your fingertips.

  • So here's a fractal generator.

  • On the left side is a single CPU doing computations.

  • On the left side are the 16 virtual machines I just spun

  • up in the world.

  • And if I click in one of these, watch carefully--

  • they'll both zoom, but start to look at the details.

  • So I'm going to zoom fairly quickly into

  • this one on the left.

  • Watch the resolution differences of these two

  • images, OK?

  • Here we go.

  • So if you look, this one is very blocky and fuzzy.

  • That one over there with 16 nodes is already computed

  • seconds before this one over here.

  • And so here we actually have an application running in App

  • Engine using Compute Engine as a background processor,

  • literally with these CPUs in the background doing the

  • computation.

  • And you can see, with even 16 processors, it's maybe shaved,

  • I don't know, half the time off, or 3/4 of the time.

  • A thousand CPUs?

  • This would just be just like this.

  • I could literally fly right into the fractal.

  • So there you have an integrated demo of App Engine

  • and Compute Engine working together.

  • And since it's getting late, I'm going to start

  • to wrap this up.

  • So there's some demos of Cloud.

  • But as I said, I would give you the pitfalls.

  • So it's kind of like when the dog grabs your sock,

  • and won't come back.

  • There are pitfalls in Cloud.

  • Needless to say, when the dog gives back that pink sock, I'm

  • sure it's going to be wet.

  • So what to watch out for in Cloud.

  • One is, Cloud is interesting.

  • The monetization models are typically by the hour, by the

  • day, by the month.

  • You have to be a little careful.

  • It does feel a little bit like death by 1,000 cuts, where

  • every day you go spend just a dollar.

  • And at the end of the month, you're like, whoa, where did I

  • spend $3,000 on this credit card bill?

  • You have to watch out for this Cloud.

  • We are trying to do things at Google to make this easier,

  • more predictable.

  • But we're not there yet.

  • So just be thoughtful of the services you're using.

  • As you move from the infrastructure as a service

  • layer to the software as a service, there's more lock-in.

  • Remember, I said it's easier to do these things.

  • But it also means you're sort of more locked in.

  • All of us as developers worry about lock-in.

  • Google is worried about this for you, too.

  • And we've done a lot to make sure you can

  • get your data out.

  • We had this whole effort called the

  • Data Liberation Front.

  • Any data that Google has, you can export it.

  • There are other things that we're also doing here to make

  • people feel more comfortable about using the value added

  • services, about feeling locked in.

  • Location matters.

  • There are different geopolitical rules, different

  • governments.

  • Depending on what you build, you have to pay

  • attention to location.

  • If you start a virtual machine in Europe, you may be subject

  • to European laws.

  • So while it's very easy for me to start it and use this

  • thing, depending on data retention policies and so

  • forth, you have to be familiar with those laws.

  • So while it's super easy to do, it doesn't mean you're

  • sort of divorced from having to understand the

  • legal end of things.

  • So that's kind of a gotcha.

  • And then, disasters happen.

  • Like, Christmas we couldn't play videos.

  • And you hear about things.

  • But what I want to say is it's really hard to keep cloud

  • services running.

  • But it's even harder to run them yourselves.

  • I've been there.

  • And so you don't have to take my word for it, build your own

  • data center, run it for a year.

  • And then when you're done, please come to Google, and we

  • will help you use our data center.

  • It will be better for you.

  • So there are pitfalls out there.

  • So with that, people ask me a lot of times, where does the

  • team come from?

  • So where else would you build the Cloud than in the

  • cloudiest city in the world, which is Seattle/Kirkland.

  • In fact, almost all the cloud providers are up there.

  • We actually tell everybody it's really cloudy all the

  • time in Seattle.

  • So I had to sort of synthesize this picture with some clouds,

  • but this is actually Mount Rainier in the background

  • going over the I-90 bridge here.

  • Trust me, it doesn't look like this ever, ever.

  • It rains all the time.

  • Please stay in New York.

  • The second-biggest location for Cloud is San Francisco,

  • followed by Mountain View.

  • And as [? Ari ?] had mentioned, we actually have a

  • team here in New York that's actually building Cloud, which

  • is super cool.

  • So we have sort of 24 hours of coverage.

  • We actually have a team in India in Hyderabad.

  • We also have one in Sydney, Australia that are helping

  • keeping these services up 24 by 7 for you.

  • And with that, I hope you learned what Cloud is.

  • I hope your dating is successful.

  • And thank you for coming to Google.

  • [APPLAUSE]

  • TONY VOELLM: And I'm happy to take questions.

  • You don't have to stay, because this is probably the

  • last call for beer.

  • But if people have questions, I'll be up here.

  • Feel free.

  • We can answer questions, or not.

  • MALE SPEAKER 1: Yeah, I think we can do--

  • TONY VOELLM: So the question is, are there Google services

  • running on Google?

  • Yes.

  • In fact, a lot of the properties that are out there

  • are running on Google.

  • I actually checked with marketing, and unfortunately,

  • they threw these golden handcuffs on me.

  • So I'm not allowed to talk about it.

  • But there are open source build servers, and all kinds

  • of interesting things are built on Cloud now.

  • Inside of Google, there's not a single line of business app

  • that's isn't built on App Engine these days.

  • In fact, a lot of our external apps are built that way, too.

  • So yeah, we are definitely using the services we're

  • building, day in and day out.

  • Our business depends on it.

  • AUDIENCE: I have a question about the

  • redundancy and just pitfalls.

  • Because like you said, things do happen.

  • If I were to get a dedicated virtual host, if I would go to

  • the hosting company, and try to figure out what kind of

  • redundancy they have, all these back-up failover things.

  • How do you guys handle things like that?

  • I've used Amazon in the past.

  • So like you said, everything is pay as you go.

  • You start up your machines, but if something does happen,

  • are you guys pretty transparent in trying to

  • figure out what happened, and just work with customers?

  • TONY VOELLM: Yeah, so the question was, how

  • transparent is Google?

  • We're very transparent.

  • I'll actually show you a quick example.

  • So one of the things you can do with the API

  • that is super useful.

  • So let me get out of this VM.

  • We actually have something that's--

  • if you look at the tool, this tool here, we actually publish

  • things like our down time that we're going to have on our

  • servers through the API.

  • So if we think that there's a service outage that's coming,

  • we'll actually go list that.

  • And I was going to see if I could find it here.

  • Yeah, get PCR.

  • Let's see here-- delete.

  • AUDIENCE: I guess my question was more in particular, like

  • Amazon had-- some of their regions have been down in the

  • past unexpectedly.

  • So if my service is in the middle of doing a bunch of

  • computations and one of the region happens to the affect

  • that, do you guys work closely with customers specifically

  • try to figure out what happened?

  • TONY VOELLM: Yeah, so for failover and disaster

  • recovery, yes.

  • I mean, there are things that we're going to do that I can't

  • talk about today--

  • because they're not released yet-- that will help with

  • these disaster recovery scenarios.

  • We're very cognizant of this.

  • If you noticed, I was talking about oh, there are multiple

  • zones out there, in terms of where you can

  • actually start services.

  • And so, the fact that we have two zones in Europe tells you

  • something about redundancy.

  • So we're building this level of redundancy in.

  • There are certain things that customers will want at some

  • point, like load balancers and failover

  • and automatic restart.

  • Today, it's a little bit more self-service.

  • We will get there.

  • Specific individual customers, we're happy to share a Google

  • experience on how we've actually done it successfully.

  • And how we've like scaled up, and kept search, docs, and all

  • these services running 24 by 7.

  • That's one of the benefits of actually working with Google

  • is we give a lot of insight into actually how we do that,

  • and how to sort of create your environments.

  • Which is why I also talk about the pitfalls.

  • Disasters are going to happen.

  • If you put everything in a single zone, and there is a

  • planned outage, or an unplanned outage, yes, you're

  • going to have a problem with your application.

  • Don't do that, right?

  • Like today, don't do that.

  • The alternative is to use something like App Engine

  • which will automatically migrate and move around

  • failures for you.

  • Which is why you sort of move up to this next tier.

  • And the you might start asking, well, could you do

  • that for virtual machines?

  • And I can't say anything.

  • AUDIENCE: Thank you.

  • AUDIENCE: Can you hear me?

  • TONY VOELLM: Yeah.

  • AUDIENCE: So I serve large, evil enterprises, so my

  • question is two-fold.

  • First of all, are there plans for a private or hybrid

  • enterprise cloud?

  • TONY VOELLM: So the first question is, is there plans

  • for a hybrid cloud?

  • And what this means is typically part of the cloud is

  • in the Google data center, the other part is hosted locally.

  • This means that there's generally a VPN service, an

  • encrypted channel into Google.

  • We actually have some documentation today on how to

  • do this with things like App Engine apps.

  • Talk to us more.

  • I'll talk to you--

  • I can't say anything here, but we can talk more.

  • It makes a lot of sense to allow these

  • hybrid models exist.

  • AUDIENCE: So the second question is, what operating

  • systems are going to be supported in the future for

  • the Compute Engine?

  • TONY VOELLM: The question is, what operating systems will be

  • supported in the future?

  • So today, there's CentOS, and there's another derivative

  • that's very familiar to most people, which I'm

  • not allowed to say.

  • But you can go find it, we call it the Gcell but it looks

  • very familiar to something else.

  • We will be providing more operating systems in the

  • future outside of Linux derivatives at some point when

  • it makes sense.

  • Part of that's going to come from demand from customers.

  • But yes, it does make sense to support more in the future at

  • some point.

  • AUDIENCE: OK.

  • Thank you.

  • AUDIENCE: Hi.

  • I heard that Hadapoo?

  • Hadoop.

  • Can you comment a little bit that you use that?

  • Maybe you don't want to say that.

  • And the other question is I heard the Google App Engine

  • supports Python, so do you plan to support Perl?

  • TONY VOELLM: Support which?

  • AUDIENCE: Perl

  • TONY VOELLM: Perl.

  • Oh, well, that's a great question.

  • I can't say anything about future features.

  • But we have heard many times that PHP is

  • popular, Perl is popular.

  • Hey, could you just work your Google magic and do all this

  • automatic scaling with any language we bring.

  • We've heard it.

  • I can't say anything more than that right now.

  • But yes, we recognize that it's really important.

  • That was one question.

  • The other one is around Hadoop.

  • I believe we have some white papers around how to run

  • Hadoop really well on top of Compute Engine.

  • And it makes sense that a lot of cloud providers have

  • provided software that makes it even faster.

  • And we definitely recognize that, too.

  • But I can't say if something is coming or not.

  • But I can tell you, yes, we recognize a

  • need for both of those.

  • AUDIENCE: Thank you.

  • AUDIENCE: I'm just wondering--

  • a long time ago, there used to be grid computing,

  • and stuff like that.

  • And peer-to-peer, and things like this.

  • Is it possible that if one's doing peer-to-peer for gaming,

  • some of the assets could be in the cloud.

  • And if for some reason--

  • mainly it would be running peer-to-peer.

  • But say like, there's a [? fallout ?] then you can

  • [INAUDIBLE].

  • TONY VOELLM: So the question is, is will you start to see

  • models of how you leverage access from the phone, and

  • failover to the cloud, or using the cloud.

  • In fact, we actually do a lot to make this possible.

  • This is one of things for the Google Endpoints that make

  • programming easier.

  • We also have things in App Engine like task queues, XMPP.

  • We have this thing called the Channel API, which will call

  • back in to your cell phone to deliver notifications.

  • So there's a lot of social sites out there that use App

  • Engine, where you're walking around, and it's sending GPS

  • location or whatever to their app.

  • And then it's like, oh hey, you have three of your best

  • friends right next door.

  • And that's all built on top of the services in App Engine.

  • So absolutely what you're asking for is there.

  • And it's getting used.

  • AUDIENCE: Thank you.

  • AUDIENCE: This is a curiosity.

  • When you upload, or when you access a Street View image on

  • the computer, how many machines do you have working

  • in a typical browsing in like Street View image?

  • TONY VOELLM: Yeah, so the question is, when you're doing

  • Street View, how many machines.

  • are back there working on that I can't tell you.

  • But what's interesting is, is in the last year, Google used

  • to have this thing called tile servers, and we'd serve out

  • these tiles.

  • We actually moved away from that now.

  • And we're actually rendering now in the browser.

  • So we're sending the data.

  • And we're allowing you to cache it locally on the phone.

  • So if you suddenly drive out of Wi-Fi area, the map keeps

  • working for you.

  • On the back end of that, go look at

  • the Google data centers.

  • And we tell you where all the Google data centers are all

  • over the world, and you can just start to like-- wow,

  • there's probably a lot of servers out there handling

  • things like Gmail and Maps.

  • Maps in particular is heavy with data.

  • So I can't tell you.

  • But it's probably a big number.

  • MALE SPEAKER 1: All right.

  • So I want to thank everybody for coming.

  • Thank you, Tony, also, for taking your time to educate us

  • on all these things.

  • There are quite a few Google people around.

  • If any of you have questions for any of us, we're all

  • wearing Google shirts.

  • Tony's also going to make himself available for the next

  • few minutes to answer any questions.

  • Thanks again, everybody.

TONY VOELLM: I head up the Google Cloud Security

字幕與單字

單字即點即查 點擊單字可以查詢單字解釋

B1 中級

谷歌雲 (The Google Cloud)

  • 133 10
    richardwang 發佈於 2021 年 01 月 14 日
影片單字