[SOUND] Economics really is the foundation of cloud computing. Cloud computing revolution really started because people were considering the costs involved in providing computational facilities. And whether there's a better way to manage all data center's computational infrastructure, networks, everything. We first look at utility pricing, and later, we will look at benefits of common infrastructure in a different video. This topic is based on a paper by Joe Weinman, Cloudonomics. And I suggest you take to look at that as well. There's more details in that paper and more benefits explained. We pick these two. So, looking at the value of utility pricing. We want to explain why you can have cloud services that are more expensive than owning your own infrastructure per hour or per day, whatever, and still be economical. Let's look at an example. Consider a car. You can buy or lease a car for, let's say, I don't know, $10 a day. Number's not that important, but let's say $10 a day. And let's say, if you want to rent a car for a day, it costs you $45 a day. So, very obviously, renting the car is more expensive than owning the car, per day. But let's say you go to a trip. Let's say you take two days, and for a vacation, and you go somewhere. You don't buy a car over there. You take an airplane trip. And then you rent a car for two days because buying wouldn't make much sense over there. So it totally depends on the type of demand that you have to figure out if it's more economical to rent or buy a car or services. So, let's look at this in a little bit more detail. We try to model this using a simple mathematical model and figure out how everything is going to work out. So let's say you have demand. Demand is shown here by demand of t that denotes that its demands per time. One day you might have a lot of computing demands, another day your demands could go down. We are looking at this for a certain time period between 0 and time capital T, let's say capital T is I don't know, a whole year, 365 days. During this time, your demand goes up and down. So P is the peak demand during one day throughout this 365 days, your maximum demand we can denote by P. What else do we have? We can talk about the average demand, you'll take the mathematical mean of all the demand during out the year. B would be the baseline cost of your computing unit. So for example, a computer costs $10 a day, that would be your B here. So that works, $10. Therefore, what is B of T? B of T would be the total cost of owning infrastructure throughout the year. And we will try to compute this value. What about cloud costs? Cloud unit cost is the cost of cloud resources, say again, per day. We want to compute what is the total cost of renting cloud computing resources throughout the year. Now finally, we have the utility premium, which is defined as the ratio of cloud cost over baseline cost. So for example, if your cloud cost is $45 then baseline cost is $10. Your utility premium would be 4.5. So, our mission here is trying to find out the total baseline cost per year, and total cloud cost per year, and then comparing them. Well, let's actually start from total baseline cost. This is easy. If your demand goes up and down, let's say this is your demand throughout the year, and goes up and down. You need to provide the services to your customers every day. And you don't go and buy new machines every day. So since this is your peak demand, you need that many machines throughout the year. You need 200 machines. That means that throughout the year you need all of 200 machines to be sitting there. And that's actually really the reason why we go through to the argument of cloud computing. But let's continue. So the total cost of owning would be the baseline cost, multiplied by the time period, 365 days, multiplied by how many machines do you have, P, the maximum number of machines that you need to answer your workload for peak demand. Now what about cloud cost? For cloud cost, it's actually a little bit more detailed. You can very quickly yen. Get and release resources from the cloud provider as you demand goes up and down. So now we can actually do an integration. Integral of the cost of the cloud that's actually you multiply by B, just look at here. Cloud cost is really you multiply by B. And you do an integral over time throughout the year from the first day to the last day. If you compute this, you get less value for the cloud costs. So now the question is, when is cloud's going to be cheaper than owning? CT less than BT. If we substitute these values that we just computed. In this argument, we want to find out when is A multiplied by U multiplied by B multiplied by T, less than P multiplied by B multiplied by T. If you do the math, implies that clouds are cheaper than baseline when the utility premium is less than the ratio of peak demand to average demand. So if your workload is very highly various, goes up and at some point goes down and goes up again and goes down, say here your average is a certain value and your peak is a certain value. If the ratio of peak to average is more than the utility premium that the cloud provider is asking it becomes more economical to use cloud computing. So in real world that was a simple mathematical model. What about in real world? So in real world and practice, demands are often very spiky which is good for cloud computing. You have news stories coming in, something happening, everybody's interested, goes to their news website, tries to find out, and then within a day or two, nobody cares anymore. The demand goes out. Marketing promotions. Manufacture wants to sell their things. There's a one week, two week time period. Everybody wants to use a certain sale, and then the sale is done. Product launch is the same thing. Internet flash floods. Your article or blog somehow made it through /.website. Everybody in the world comes and checks out your blog and you need to resources to respond to that sort of demand. And then within a few days again that demand goes down. Tax season, everybody has to file taxes within a certain time period throughout the year. The rest of the year nobody goes to that website. So in reality, lots of demands are spiky. Now often, a hybrid cloud computing model provide the best advantage. So going back to our car example, in your own town, you own a car, you use it for your daily commute, go around. Whenever you travel, you rent a car. For example, when you need to move from one house to another house, you rent a van. You don't own a van all the time. Even in your own hometown. So a key factor is really figuring out the ratio of your peak demand to the average demand that you have. And own part of your computing infrastructure yourself, for your average demand, and be able to move on to cloud computing and providers whenever there is a need to. Although of course, I should mention here that it’s not very simple, as I just implied. You should also consider the network costs. Because if you are providing a hybrid model, you have to have a good network that connects your infrastructure to the cloud infrastructure. You need to provide interoperability between your own cloud and the public cloud so that might cost a little bit. And of course, you need to consider reliability requirements, accessibility requirements that you have. So it's a little bit more detailed but what we see today is that a lot of entities grow towards hybrid cloud computing. So to summarize, utility pricing is good when demand varies over time. As is the case for startup companies, seasonal businesses, etc. When utility premium is less than the ratio of peak demands to average demand, cloud computing is beneficial and makes sense. In the next video, we will look at the possible savings on the other side for the cloud providers that they can create by using statistical multiplexing. That's putting different jobs from different customers together. [MUSIC]