Cloud’s analytics tools for collecting & visualizing game telemetry & data (Google Cloud Next ‘17)

[MUSIC PLAYING] INDRANIL CHAKRABORTY: Hello, everyone Thank you for coming to this session of IoT solution on Google Cloud I am Indranil Chakraborty, I’m a product guy at Google Cloud working on IoT What we’re going to cover today is three things One is how do we think about security when it comes to building IoT solution on Cloud? Second, how do we handle scale when we think about IoT solution on Cloud? And finally, how do we get actionable insight for when we build IoT solution? We’ll also have a demo at the end where one of our partner has built a pretty interesting solution for IoT on Google Cloud Let’s get started What is IoT? If you Google internet of things, one of the first simple definition that you see is interconnect via internet of computing devices embedded in everyday things Essentially, everyday things interconnected via the internet Hence, internet of things And one of the main application is the first application you see of internet of things is at home When you think about Nest thermostat, you don’t have to turn on your thermostat, or you don’t have to turn off your thermostat when you leave the home The Nest thermostat has a sensor It knows when you come inside your house, and it turns on the heater to a set temperature Google Home can also interact with your Nest thermostat It can even turn on or off your light In fact, you can use Nest thermostat to turn off the light when you leave the home So simple everyday things interacting with each other via the internet to create simple magic at home And when you think about internet of things, the application of IoT is much broader outside the context of home Let’s take example the parking spaces It is estimated that there are around 800 million parking spaces in the United States San Francisco city alone has close to half a million public parking spaces, but it’s still really hard to find parking in this area In fact the San Francisco city has done an interesting experiment with SF Park where they have embittered small sensors in a number of public spaces which can accurately detect whether a parking space is occupied or not And this information is sent to Cloud via the internet And so imagine if you are driving to meet your friend at a restaurant, your car knows exactly where to find a parking space because your car is interfacing via internet with all the parking space which in turn is sending real time update about the availability of parking space to your car Again, everyday things interacting with each other via internet to really make our life easier It will also be nice if while you are with your friend at the cafe, you get an alert about as the meter is running expiring, and so you can pay or extend the meter right from on your phone We also seeing a lot of interesting application in the industrial space, as well In fact, we have partners and customers who are working with manufacturing industries where they’re able to remotely monitor a lot of their machines across multiple factories You see, many of these machines already collect a lot of data, even today But unfortunately, they’re not able to use it Less than 10% of all the data is actually put to use So imagine if all these machines are connected over the internet, and as an Enterprise customer, or as a factory owner, you can monitor all your machines remotely across different regions, or even across the globe So we see that there’s a lot of applications of internet of things when it comes to industrial segment, as well

In fact, at Google, we think internet of things or IoT, as we call it, is at a point of inflection Depending on the analyst report you look into, it is estimated that by 2020 you will have anywhere from 20 billion to 40 billion devices connected to internet Now, each of these small devices might send a few bytes of data every second But when you look at it in a cumulative way, the overall amount of data of which these devices will generate is going to be massive And so it creates three challenges for developers, for partners, when they start thinking about building IoT solution on Cloud One is security And the question about security is, how do we make sure that none of the devices which are out on the field connected is compromised? The recent Dyn attack is a staunch reminder that there is a lot of vulnerability out there And how do we make sure that the data which these devices are generating is secured not just when it’s stored in the Cloud, but also during transit as it’s transiting from the device to the Cloud? And the second challenge is scale When you have so many millions or billions of devices, how do you make sure that all those devices is running the same version of operating system, same version of firmware, has the same security patch so it’s very hard to compromise? And then how do you make sure that the torrent of data which would be generated by so many devices can be easily handled by your application on the Cloud or on-prem? And finally, insights Well, you can have a torrent of data from all these devices, but what do you do with this data? How do you tease out really interesting, actionable insight from the data, which you can then apply on your application or for your business case? At Google, we have been thinking hard about these challenges for a number of years And in fact, when it comes to massive amount of data processing, ingesting, and then analyzing, we’ve been working on big data problems for over a decade in Google Cloud So we have Pub/Sub, Data Flow Bigtable, Cloud Email, to name a few of the services which can operate at scale and handle massive amount of data, which I will talk more about When it comes to internet of things, we not only think in the context of Cloud for security, scale, and insight, but we also think about how do we have the same at the Edge as well? And so it is my pleasure to invite Venkat Rapaka, who is director of product manager who leads IoT platform on Google to talk about how we think about Edge devices and Android things Venkat VENKAT RAPAKA: Thanks, Indranil Hi, everyone I’m Venkat Rapaka, as Indranil mentioned I want to talk a little bit today about how the notions of scale security and insight, obviously they’re as important on the Edge as on the Cloud side, and I’m excited to talk about our latest offering in this space that should be able to deliver all of this at the Edge as well I’ll talk about Android things Let me first talk about what it is It is Android, essentially, right? Android is one of the most popular operating systems out there today But it’s Android, but purpose built for IoT Think about the internet of things devices, there’s some pretty interesting characteristics to them A majority of them don’t have a display When they do have a display, it depicts purpose devices typically There is no user-installable applications, right? There’s no notion of the use that interacting with it to install new capabilities on the device So what we did is we removed a number of things that we didn’t think were essential in the IoT space Things like the applications that are Android comes with, notification support, and a bunch of things that are for UI, similarly to render a UI with installable applications And instead replace it with things that are, I think, very relevant for IoT Like peripheral support is a good example So you can think of the right type of sensors that people will be building into these devices

Actuators if you want to actually operate equipment So we added back some other things that we think are very important to the IoT space, but in the process we shrunk roughly the footprint of Android to about half the size Which basically means that this is going to be extensible to a lot more devices Still, we’re talking about, to be very clear, SoC class devices These are devices that typically run something like Linux, and they tend to be line powered devices But these are the types of devices that you usually expect to see at the edge of the network, and that run with a lot of intelligence on it So I’ll talk about Android things a little bit more like, it’s an interim solution, the multiple elements to it So I talked about the operating system a little bit It’s based on Android, it’s secure Android comes with monthly security updates We are committing to multiple years of monthly security updates here It’s, of course, updateable I talked about removing a number of things from Android, but the thing that preserved is a majority of the Android APIs All the APIs that are delivering for these types of devices are all preserved And what that means is we get all the Android APIs, and the developer environment and the test tools that are currently available, but also a million Android app developers that exist today could all become IoT device if they so choose So if you’re out there and you want to build a device, you no longer need deep OS expertise You don’t need people who understand firmware, and kernel, and OS level security You should be able to take Android things and be able to build a device just by building an Android app that can run on top of Android The other thing that we’re doing is we’re including APIs to enable these devices to be managed at scale So if you have hundreds or thousands of these devices in your installation, you should be able to use a standard off the shelf console and be able to manage all of these devices But the OS is only a part of what it takes to build a great device There’s hardware, right? And building hardware is incredibly complicated for a lot of people How do you pick the components, how do you assemble them all together? And to do that, we are working very closely with a number of major industry players, and we’re building with them pre-certified boards And these boards come with compute, they come with connectivity and storage So what this does is it takes both a lot of the complexity, and the guesswork, and the risk that it takes to build a great device, but also it takes out the time to market So these certified boards, they guarantee to run Android things They get regular updates every month, and they get you started out of the gate without having to know a lot about how to build a device And again, our attempt with these and with our partners is that if you’re trying to prototype, you should be able to buy one of these things, not have to commit to 10,000 of them just to get to some minimum order quantity to get scale efficiency So the cost will scale linearly, as well So these are all the ways we’re trying to take the friction out of the ability to build a great device The third thing that’s really interesting is that as we’re building this OS, we’re also looking at all the types of services that we can include in a turnkey fashion to make it easy to build a great product at the end of the day So a few examples up here The first symbol that you see is, of course, the Google Cloud platform So if you’re building a device that requires data to be connected, like sent back to a cloud, you should be able to use Google Cloud platform in a seamless way The second one is TensorFlow To be able to do on-device intelligence, of course Location services, location services come for free with Android The fourth one symbolizes connectivity Of course, there’s Wi-Fi and Bluetooth as you would expect, but also because it’s based on Android, you get cellular connectivity capability as well And this is, of course, one of the most commonly used operating systems by all the carriers in the world So to be able to build now an LTE capable device suddenly becomes much more accessible to people using something like Android Things And over time, we intend to support additional types of connectivity matters like thread, for example, for local low power connectivity with devices that are out there that are battery powered, for example So this is the sum total of what makes Android Things what it is, which is an end to end solution for people to be able to build a great device Let me talk about what this starts to enable And I’ll pick one specific example to talk about, a smart camera So imagine you’re trying to build something like a Nest cam, it’s a security camera You should be able to pick from among options of certified hardware

Something that meets your price and your performance characteristics of the device you want to build Should be able to get one of those Should be able to download Android Things so you have a secure operating system to go with it But it’s a camera you’re building, there’s a video feed coming out, and you might want to save it somewhere on the Cloud Or you might want to be able to access it using either a browser or a mobile device And I mentioned, of course, the Google Cloud Platform and, it’s built right in But also bandwidth is expensive And if you have installation of these devices, and you don’t want to be pumping data out all the time You want to be able to do something intelligent on the device itself to make sure that you only send data up to the Cloud when you want to And for that we have TensorFlow, right? So as an example, you should be able to look at what the camera is seeing on the device itself and decide, what is it that I’m looking at? Is it a cat running through the yard, or is it a person walking up? And is it a person that I trust and know, or is it someone that I don’t know, and therefore I’m going to alert? So suddenly you have the variety of options, starting all the way from doing nothing, log it but don’t wake somebody up, wake somebody up So you suddenly start to do all of that locally without having to send a ton of data out into the cloud to do that And I talked about one specific example of a security camera, but the thing that makes this interesting is this should be the same kind of a process Just by learning a different TensorFlow model, suddenly you have a traffic camera that can figure out the license plate number on device itself And take a photo of somebody who violates a red light and then sends that photo out You can imagine this camera sitting inside a building and doing occupancy detection and saying, this is the number of people who use this space at this time of the day, and only send those heuristics up, as opposed to actually sending the full data feed to have to be processed on the cloud But all of a sudden, you can imagine building a device like this And just by what you download to it, it suddenly changes in characteristics and what it can do I’m going to turn this around now and start talking about how as a device developer you can build this very easily But I’m going to turn this around and say, let’s imagine that you are an enterprise, and you have 1,000 of these cameras sitting all around your buildings worldwide, and you want to do premise monitoring I talked about the management at scale So you should be able to look at all of these cameras on one console, and you should be able to update the model that you use on it to maybe do facial recognition Maybe your employee database changes every day, because somebody new joins, somebody left Maybe you want to download that new database down to each of these cameras so they can do detection at the edge itself for you These are all possible from a console with one click And the other cool thing is, you might have cameras also running Android Things inside your building doing your occupancy detection, and your cameras that are doing premise monitoring Now they’re all based on Android Things You should be able to use the same console to look at all of them in one spot, which is exactly how we’re trying to take the friction out of an enterprise with all the management overhead you would expect So a lot more happening We’re right now in developer preview with Android Things Over the next few months, we’re going to have multiple developer preview releases coming out Towards the early part of the second half of this year, we’re going to go into commercial availability So please stay tuned, you’ll find more information at the tooling subdev Thank you again INDRANIL CHAKRABORTY: Thank you, Venkat So as Venkat mentioned, with Android Things, you can really build a device application which is secured, which can scale, and you can use TensorFlow to get insights on the edge itself Now let’s talk about Cloud, and how do we handle security scale and insight within Google Cloud? There are a number of services which I’ve listed here I’m going to talk about a couple of key services which can really help in building a secure, scalable, and insightful application when it comes to IoT First, let’s talk about security On Google Cloud, every data by default is encrypted It’s encrypted at rest as well as it’s encrypted when in transit Now, we understand that there could be a couple of very specific customer client needs, or user needs, because of which you would need multiple option depending on the level of protection

that the customer or you are more comfortable with So we do offer a wide spectrum of options when it comes to security The first, as I mentioned, is encryption by default. In which case you are using Google Cloud scale and velocity and our key management system without needing to implement any of your own In this case, you would be using the same key management system which is used by other Google products such as Apps, Google Maps, Search, and others And so you get the benefit of Google scale and infrastructure At the same time, you get a high degree of security when it comes to IoT data For reasons if you do not choose to trust with us for data key management, we do offer customer supplied encryption keys If you would like to maintain your keys and manage your keys on-prem, you could do that with this option In this option, you will have full control on creation and management of keys, and Google will use those keys only at the time of encryption, but will not retain it on the Cloud You will get the security, but the user will have to make sure that the keys are provisioned and managed on-prem We are also working with customers and partners, and we realized that we need to offer a middle ground, as well If you don’t trust us with the key management system, but you don’t want to have full control, then we have a middle ground which is called the Cloud Key Management System In this case, you have full control for key creation, deletion, rotation, revocation But the entire key management system is done on Google Cloud infrastructure So you get the benefit of Google Cloud, the velocity of Google Cloud infrastructure, but you have full control of how you want to manage your own keys So again, essentially we offer a high degree of security on Cloud with the option which meets your needs and is flexible, as well So that’s how we look at security, which is high security, but at the same time we do give you options to meet your own customer or user goals When it comes to scale, as I mentioned before, we have been working on this for quite a long time And if you take the example of the parking space, individual sensors might be sending small trickle of data But in aggregate, when you think about even if you have 10 million more parking spaces in US which are all connected with sensors to indicate whether the space is occupied or not, that’s a huge volume of data Even if those sensors are sending the data over a few seconds or a few minutes So you need a service which can ingest this massive scale of data, and then route this data to different services for processing and analysis And Cloud Pub/Sub is a great service for that What it essentially does is it has a topic which all its devices can publish do, either directly or via gateway And you can have other services to subscribe to these topics to ingest to consume the data The three key things which is unique to Cloud Pub/Sub Number one is the durable message persistence What does it mean? So if you think about these small constrained devices, they don’t have enough compute power to try to resend the data when it’s not getting an ack back from the ingestion service You want to be able to send the data out without much challenge, and without much computing power So Cloud pops up, makes sure that the data is stored and we didn’t send an ack back until the published from the device is consumed So that’s one key advantage A second is even though your devices might be globally distributed, you would want to be able to consume it from a central location so you can remotely monitor all those devices So Cloud Pub/Sub is a global service And you don’t have to think about creating separate shards,

or different regions You just publish your devices to one topic, and it handles it, and it scales In fact, Pub/Sub scales internally to the extent that whether it is a couple of thousands of messages or a couple of millions, it just scales automatically When it comes to internet of things, there are certain unpredictability in terms of load So sometimes the devices might be sending few data, and all of a sudden if there is in real world there’s increasing load, it might be sending a massive amount of data for ingestion And Cloud Pub/Sub scales automatically without you needing to provision or add any new infrastructure In fact, it can handle anywhere from a few thousand messages to millions of messages per second in a matter of minutes without you needing to do anything So that’s great about Pub/Sub Another thing it does well it is it separates the downstream application from the upstream deployment of your devices So you can continue to make changes or build new application which consumes the data sent by your devices, or you can send it to a different e-tail pipeline, you can have data flow or cloud function consume it And you can keep making changes without needing to change your device deployment So it creates an abstraction between your upstream device deployed and your downstream application So it allows you to iterate on your application at a much faster speed And finally, it also allows you to connect to some of the powerful real time services we have, such as Cloud Dataflow and Cloud Function, which I will just talk about So now that you’ve ingested this massive amount of data at scale, in case of IoT application, most of the typical time you would want to get some real time insights Real time information If you think about the parking lot example, you would want to know real time how many spaces are available near Moscone Center, or how many spaces are available near a park public garage space nearby So that as you are driving in, your car knows exactly which parking garage to get into And so Dataflow has the mechanism where it can real time process streams of data You can create windows to define boundaries and then it can aggregate, you can run some aggregation or average, or run other computation, as well So Dataflow can be used for a couple of things when you get streams of data from devices One is, as you get data from different devices in the field, the data format could be different And what you would want to do before you process or analyze the data is you would want to standardize the format across this different class of devices And so you could use that with Dataflow, where you can create a real time job to do format translation for data coming across different devices Second is in real time, it can create average You can compute an average, or you can do different transformation on the data itself In our simple example for parking space, one could compute average number of spaces available in a parking garage or in the whole region, and then you can use that to build some demand prediction, or even do pricing based on demand, as well And finally, Dataflow can also be used to not just work on the data which you’re receiving from the internet of things, you can also add data from other services So you can get weather data, you can get some traffic data, and combine all this information to do more rich analysis real time That’s the power of data flow, which we think is great to build IoT application Once you process the data, you want a simple way of moving the data or storing the data so that you can do subsequent analysis on that data And we recently announced Dataflow Templates, which is great for easy data movement between Pub/Sub to other cloud services, such as Bigtable or BigQuery for storage, DCS for storage, or even for a Cloud ML The unique advantage of Dataflow Template

is that as a developer, you can create your own environment You can create a template to connect data from Pub/Sub to, let’s say, BigQuery in this example Compile it, and then you save it And then a non developer operation user doesn’t have to compile it All they do is they’ll open the Google Cloud console as it’s shown here, they will select the template which just pops up to BigQuery It’s parametrized So they will enter the topic of the Pub/Sub on which the data from these devices are published They will select the table BigQuery, and then just run the job And in the background, the template automatically spins up a job and transfers your data as it comes in to the BigQuery table which you mentioned So really easy, and it helps you not only for storage, but also to scale So as a developer you create templates once, and the operations folks can then execute on those run jobs on this template without needing to compile every time You can also use templates for creating simple rules So as a developer, you can create a rule that says every time the number of parking space goes below 10 in a public garage, I want to send an alert, or I want to send SMS message to to these mobile phones You can also create an alert where you say every time something happens, I want to send a message to datastore All of that can be done using Templates As a developer, you create, you build the template, you code it, and you save it And not non-developer users can just execute the job without needing to compile every time So that works great for the purpose of scale, as well Now once you’ve analyzed the data at real time, you have ingested the data, you’ve analyzed it, there are a number of options to store the data on Google Cloud You can use Google Cloud Storage if you don’t care about the structure You can use Cloud Datastore if you have key value pair format and the data is more structured You can even use Cloud SQL for that matter But there’s one service which I will highlight which applies very well for the IoT use case, which is Bigtable If you think about the public parking example, there are use cases where the public garage, the garage owner might want to do some demand analysis to really understand what are some of the peak times when I see a peak demand for parking? And he or she can also use it to do predictive pricing based on demand Best way to do that is to do time series analysis on the data stream, which the parking garage and I might be collecting from all the parking spaces And this is where Bigtable shines Because Bigtable scales horizontally Whether you’re talking about few thousands of writes per second or millions of writes per second, Bigtable scales internally And it’s great for time series analysis, because it’s a column based, NoSQL database The challenge with Bigtable is you have to be very careful in how you design the schema when you’re thinking about time series analysis But what’s great is we have open TSDB, which is an open source we strongly support And it abstracts a time series schema on Bigtable So if you want to do time series analysis on a series of data coming from your IoT devices, and your IoT solution requires predictive processing, you can use open TSDB on Bigtable, and you should be up and running and designing the time series in a matter of no time Now you’ve ingested the data, you’ve processed data real time, you stored the data, as well There’s an aspect of reporting and visualization which is also important when you think about building IoT solution We do have two services which can help you build reports and dashboards There’s data and Data Studio what Datalab is great for, more sophisticated analysis If you want to get the data, and if you have statisticians with PhD’s who would really want to go deeper into the data use R to create some regression model, Datalab great It has a notebook interface, it leverages existing Jupiter modules, and is suitable for any data science and houses, or even building machine learning models as well It is closely integrated with BigQuery and Cloud ML, so it makes it really easy for you to take that data

and then build a cloud ML model for more sophisticated analysis down the road Data Studio, on the other hand, is primarily targeted for business analysts So if you want to create a simple dashboard where you want to show all your fleet or truck movements on a Google map, Data Studio is great And then you can use that to constantly monitor and generate report on an ongoing basis It doesn’t require any technical expertise You can drag, drop BigQuery table or Excel sheet, or some other source, even CSVs, onto Data Studio And you can create some really enriched dashboard for your reporting purposes So Datalab and Data Studio is great for generating reports to really tease out actionable insights from the data after you’ve processed and stored it So I talked about the different processes I’m sure you would want to see all this in action So I’m going invite Jose Ugia, who’s going to give a demo of a smart fleet which they have built on Google Cloud Jose JOSE UGIA: Thank, Indranil So every time I see the slide that Indranil just showed about the respected growth of the IoT devices in the industry, then I always get to this profound, yet simple thought And that is that we’ll be surrounded by them And because of that, in tech we have gotten things wrong a few times, especially when it comes to security, when it comes to scalability, when it comes to reliability And I think this time what we can do is try to learn from the previous mistakes Come up with a set of practices in all those things So by the time this happens, we are ready to do that And that’s something that in our company we have really present, and we love to keep it that way for the months to come So at Noa, what we do is, our vision is to transform the way we understand and utilize urban transportation And we do this in two main ways We want to do that in two main verticals Which are making transportation more convenient to you within urban fleets– that’s commuters, that students, that in police of campuses We want to make it easier to find actual means of transportation to move around But the most important part, and this is where we are putting most of our efforts, is on reducing the operational costs of managing those fleets Especially when it comes to a large number of them So there are things involved that grow with the number of devices linearly And what we want through automation and add in features that provide insights and knowledge, we want to reduce this cost And so one of the companies that we work with is Google This is why we have this bike over there This bike has something here which we call the donut, which is our hardware device that helps us sense So next time you go to the Google campus and you think about taking one of those bicycles as a souvenir, you may think twice, because we are tracking those things So if we can switch quickly to the wall, that sounds very powerful Switch to the walls This is how the device looks like So as you can see, it’s a small rounded element with machinery inside And you may imagine why we call it donut, that’s obvious reasons And it has this shape because it goes next to the wheel So the interesting thing is that you can simply take that wheel out, replace it with one that has this device, and then boom, you’re connected As easy as that What this device senses is things like humidity, temperature, gas, or pollution related particle And we also have antennas to connect to the cellular network We know the location of the device, because we have GPS antennas as well All together to provide us with the information that we need And if we can go back to the slides please, the information that we need to help these operators make those decisions that make them reduce costs There is already cost reduction But in some cases, this means viability It may mean the difference between having a program or not So as you may have heard, there are a few cities in the US in which this has really been a huge struggle In which the public systems cannot sustain those programs anymore because of their pricing

So in some cases that will mean having them or not And this is what really makes us excited about it Before we jump into matter, we have been running on Google for a few months now And I thought it was worth sharing some of the stats that we have gathered with you Googlers, they seem to be healthy people They cycle the world every month with their bicycles Everyone knows the diameter of our lovely blue planet, right? So that’s a lot of distance And we only have a subset of the devices equipped, we don’t have all of them So that’s something that came as a surprise, not because we thought that everyone was sedentary, but because we thought the activity was going to be lower It was really great to see that We have more than 100,000 trips every three months at Google, and there might be something going on at night Because we have seen that after 8 PM And before 6 AM, there are 1,000 trips every month So not sure if we will find out or not, but there’s some kind of secret project or fund that happens really late at night within the Google campus But how do we leverage Google products? And why this is really useful for us We base our data pipeline from our devices in two main principles One of them is real time We need to leave in real time because there are many things that depend on this Like security, like infrastructure, like availability, like the month of bicycles And we need to know about those as quickly as possible If someone is maybe trying to steal a bike, any of you maybe, we need to know but as quickly as possible So we want that to be milliseconds or seconds That’s why we ingest into Pub/Sub And once we do that, we transfer in parallel to all the different services That is Alert, that is location updates, that is insights, as well The other principle that we have is that our information is simple at the beginning, how we want to transform that as it comes So think of it as LEGO pieces, right? By themselves, they may be just colorful, beautiful pieces, but they mean almost nothing What we do with them, when we put them together, when we assemble them, then we can construct things like cars, rockets, houses, buildings, things that have meaning, right? So this is what we do with our information as soon as it comes, and we transform it as it arrives and then send it to the different services which are interested And that is once again within the order of milliseconds or seconds This is how it looks when we translate what I have said into Google services, or Google products from a really high level There are two things to highlight in this slide One of them is that we use MQTT, we have a set of machines that we use as a gateway to which our devices communicate They ingest We use MQTT as a protocol, and protocol buffers as a message format Just because it’s convenient, we are in the outside, we’re in a hostile environment with rain, with cold, with wind, with reduced cellular connectivity, with reduced GPS coverage And so having the chance to have a protocol that is meant thought for that is great, that’s why we use it The other one is that if you see the second part, this spurt of Pub/Sub plus Data Flow, this layers which seem to be repeated, this tax is what I just mentioned before This allows us to, once we ingest into Pub/Sub and we process in Data Flow, then some of those results will make it into storages, or into different services themselves Some of those will be republished into Pub/Sub so we can keep on transforming and analyzing that And I’ll show you a couple of examples of that in a second Lastly we’ll love to touch on something that has really changed the way we analyze our information We used to spend a lot of time to find out how our data was performing That is in terms of insights, in terms of business intelligence, but also monitoring Because we have our [INAUDIBLE] devices, and we have data about those So we can find out whether the GPS accuracy is good or bad, for example We take a longer time, now we leverage where we make use of BigQuery query so we can make decisions much faster And sometimes this really provides for a big difference in how our product is being run, and how to invest the little resources that we have as a small company to make the best out of it for our customers If we can switch to the product demo without further ado, the theory was great But things are much better than when we show them live And so let’s all invoke all the demo gods, and let’s challenge them right here on the stage Can we switch now? Yeah So what you will see here is just a reflection of what I just explained

In this case, I want to show you two things One is right where we are right now where this device is, and where one of our colleagues is cycling She must have gone really far, there she is The reason why it’s green is because she’s moving And this shows a reflection of the current snapshot This is where she is right now, this is the battery the device has, this is the status of the device You saw that I just moved this device that we have over here In a few seconds, if I don’t break it, it’s going to get green [INAUDIBLE] So this is just transmitting from here right now real time And you saw that quickly, we saw it on the system But things get more interesting if we go to the Google Fleet Because you just saw three bicycles But if we go in here, there’s plenty more This is right now in Mountain View after lunchtime, which is one of the peak times of activity So if we get to one of the most crowded areas in here, in the cross with the Shorebird and Charleston, then we have bicycles moving I don’t know if you can see them But as we move through the screen, they are moving right now So this is just the basic snapshot But why do we reaggregate and process information? What can we get out of it, right? If we go to Trips, which is one of the most interesting views for many of our operators, just with one query we can see what’s the path of that bicycle If we want to learn about what happened to specific bicycle, one of the back, or it may be suffering any suspicious behavior We can come here, now it’s not trackings, anymore it’s not little messages, it’s not the LEGO pieces Now we have that full trip So we have the start, the end, the velocity, the distance Things that make it much easier for us to consume And then here we can keep going back as we can see And we’ll go through the history of that bicycle Concentrate on the trips that we are interested in Finally, we have seen trackings, you have seen trips This is great The interesting part of this approach is that we can go from here into whatever we need So for example, once we have Trips, we do two things We run a Data Flow job to filter them out Because you know, GPS has errors, so you need to filter that out So you need to maybe go through certain APIs to make it a bit clearer So we do that, but another thing that we do is we calculate metrics, organization-wise So what you’re seeing is the aggregate metrics for distance in Google for the overall organization So there’s a time based serious metric And as you can see, as we change the actual date range, this is really quick, this is round trip, by the way The servers are in Ireland, so it’s round trip You see that’s really quick But that’s not the important thing The important thing is that this information is really pre-calculated That’s why it’s quick Which means that our servers don’t have to calculate it every single time that we run this query Not only that, but also that we can use all those things which had already denormalized and stored in our system, in our databases, we can use them then for further jobs Like analytics, machine learning, and so on and so forth And so with that, this is what I want to show you I hope this resembles something that you may be thinking of, some project that you may have already And I hope this is good insights for you for the actual months to come If you want to find out more, you can find us under they’re [INAUDIBLE] There’s a little section on the right that says something like jobs you may be interested in But other than that, I want to thank you, and would love to hand it back to Indranil INDRANIL CHAKRABORTY: Thank you, Jose So as you saw in Jose’s demo, he used Google Cloud not just to track real time the different bicycles, but he also used it to do post-analysis, and even those analyses were pretty fast, because he was using BigQuery and others And it’s interesting that this whole application can also be used for urban transportation, and even in the case of parking lot space We also do have a demo on the third floor in the IoT expert booth I would highly encourage that you guys should check it out Essentially what it does is it shows the simple application of Pub/Sub, Data Flow, and is compatible with Android

Things, as well So what does demo does, and as you go, you will find out It has a couple of temperature sensors on Raspberry Pi And these thermometer sensors are sending temperature measurement data, continuously sending the measurement data, to Pub/Sub, which then distributes the data to Data Flow And data flow does in real time compute the average temperature And then it goes through different pipeline And every time, as the average temperature reaches above a certain threshold, it fires a Firebase message to the controller which turns on the fan So it’s a very simple, nifty demo But it shows you the different pieces in a very simplistic way And how you can use Pub/Sub, Data Flow, Android Things, and other cloud services to build your IoT solution for your specific use case We also do have two other IoT session One tomorrow at around 11:00 in the morning at Marriott, which covers Nest, and how Nest is using Google Cloud Platform to build their thermostat intelligent application And so I think that’s something which would be really interesting if you’re interested in building IoT solution And then there is another session which is on Friday afternoon, at around 1:20 PM at Moscone West And there, we’re going to show how our customers and partners are building industrial IoT application using Google Cloud Platform So I think that’s something which will be interesting for you all, as well So to summarize, what we covered is how we use Android Things to make the edge secure, and how we use Android Things to manage millions and billions of devices at scale And also you were able to run TensorFlow to get insights at the edge itself We also talked about how Google Cloud, we offer multiple options to add the same degree of security so that it can meet your specific client or user needs as you’re building out the application And I also covered Pub/Sub Data Flow, and Google Bigtable, how that can be used to ingest data at scale, processing it and analyze it real time at scale And use some of our dashboarding applications, such as Data Flow or Data Studio to get actionable insights So we at Google are very excited with this space We are getting started, and we are working hard not just for security scale and insight, we are also working hard to make it really simple for developers and customers to make IoT applications on Google Cloud There’s a lot more you’ll hear from us throughout the year, so stay tuned And we’re excited, and I hope you guys are excited to join us on the journey, as well Thank you [APPLAUSE] [MUSIC PLAYING]