Lecture 22: Economics Tutorial

Summary

This lecture from Cloud Computing delves into the economic considerations and models for utilizing cloud services effectively. It discusses when it might be economically advantageous for a business to opt for cloud solutions, especially in the context of variable demand and peak usage scenarios. The lecture also explores cost calculations and models to determine whether utilizing cloud infrastructure is more cost-effective compared to traditional in-house setups. Through various examples and problem-solving exercises, this lecture offers insights into understanding the complex cost dynamics involved in cloud computing.

Highlights

• Cloud infrastructure is beneficial in scenarios with variable demand and short-term high usage needs. 🖥️
• Choosing between cloud and in-house setups depends on cost efficiency calculations. 💸
• Predictive demand models are essential for optimizing cloud resource usage. 🔮
• The economic model of cloud deals with calculating utility premiums and understanding peak vs. average demand. 📉
• Penalties can occur with resource mismatch - either from under-utilized or unmet demand. ⚖️

Key Takeaways

• Understanding cloud economics helps determine when cloud solutions are financially viable. 💡
• Important factors include demand variability, cost models, and peak demand calculations. 📊
• Utilizing cloud services efficiently requires a grasp of economics and predictive demand models. 🌐
• Real-world cloud cost dynamics are complex, needing a deep dive into practical scenarios. 🔍
• Analyzing scenarios and conducting simple calculations can provide clarity on cloud benefits. 📈

Overview

The lecture kicks off with a focus on understanding when cloud computing is economically viable. It stresses the importance of analyzing different demand models to figure out if cloud solutions can potentially cut costs for a company. Throughout the lecture, there's an emphasis on the 'pay as you go' model, which is advantageous in scenarios with variable demand and short-term high usage needs.

Delving deeper into economic models, the lecture discusses key components like utility premiums and cost calculations. A significant portion is dedicated to understanding peak vs. average demand scenarios, and how these influence the decision to move to cloud infrastructure. By explaining the calculations required, the lecture provides clarity on cost efficiency and when it might be wise to choose cloud services over in-house solutions.

A series of practical examples and exercises are used to drive home these points. The practical section helps students grasp how to apply economic models to real-world situations, ultimately understanding the benefits and limitations of cloud computing. By highlighting the calculations, penalties, and considerations involved in cloud economics, the lecture provides an engaging and insightful overview of the topic.

Chapters

• 00:00 - 01:00: Introduction to Cloud Economics In this chapter, the discussion focuses on cloud computing, specifically addressing cloud economics. It invites readers to explore problems and exercises related to the financial aspects of cloud technology and its implications.
• 01:00 - 03:30: Overview of Cloud Cost Models The chapter 'Overview of Cloud Cost Models' continues from a previous lecture on cloud economics. It revisits when it is economical to use cloud services and examines problems or scenarios to help clarify the concepts through recap slides.
• 03:30 - 05:30: When to Use Cloud Solutions This chapter explores when it's appropriate to use cloud solutions, emphasizing the economic advantages, such as shared infrastructure, standardized resources, and benefits from statistical multiplexing. It discusses how demands are pooled to provide location-independent, online connectivity.
• 05:30 - 11:00: Cloud Cost Calculations The chapter 'Cloud Cost Calculations' discusses various pricing models for network and utility services, focusing on a pay-per-use or pay-as-you-go model. This approach is beneficial for environments with fluctuating demand levels, utilizing on-demand resources that can scale up or down without delay or significant cost implications. The flexibility and elasticity of these resources make it advantageous for efficiently managing expenses and optimizing resource allocation. Overall, the chapter suggests that adopting these models results in a mutually beneficial scenario for managing cloud costs.
• 11:00 - 16:00: Practical Considerations for Cloud Economics This chapter discusses the important considerations organizations must weigh when deciding to adopt cloud solutions. It highlights the decision-making process related to leveraging cloud for different needs, emphasizing scenarios where cloud deployment might be beneficial, such as when high-end systems are required temporarily. Conversely, if the need for resources is more permanent and consistent daily, other solutions might need consideration.
• 16:00 - 19:00: Example Problem: Calculating Penalty Costs The chapter discusses the economic considerations of using cloud computing services. It highlights that while cloud solutions can be efficient, they may not always be cost-effective, especially for usage periods extending beyond a few hours. The chapter refers to a cost modeling approach that incorporates variables such as demand (represented as dt), which is a function of time. The aim is to understand the overall costs associated with cloud usage, taking into account demand fluctuations and peak times to determine economic feasibility.
• 19:00 - 33:00: Example Problem: Comparing In-house vs. Cloud Costs The chapter explores the example problem of comparing costs between in-house infrastructure and cloud solutions. It introduces key variables like demand (a), baseline cost for ownership (b), and cloud unit cost (c). The chapter explains how to calculate total cost (ct) and baseline (bt), and introduces the concept of utility premium, represented by the ratio of cloud to baseline cost. Using a provided example, it illustrates a utility premium of 4.55, suggesting the cost-effectiveness of cloud solutions. This concept is further elaborated with calculations and previous slides.
• 33:00 - 34:30: Conclusion In the conclusion chapter, the focus is on calculating costs and understanding baseline design in relation to peak demand. It mentions calculating costs over a period from zero to t, multiplying various factors like units and baselines. The chapter emphasizes handling peak times in baseline calculations, ensuring that baseline costing aligns with peak demand scenarios. Overall, it underscores the importance of precise baseline and peak time management in cost calculations.

Lecture 22: Economics Tutorial Transcription

• 00:00 - 00:30 hello ah so we will continue our discussion on cloud computing ah today ah we will do some ah problem or some exercise on ah [clou/cloud] cloud economic right so that whether it is
• 00:30 - 01:00 economical to use cloud or ah when it is economical to use cloud we have already discussed this ah particular topic in our ah previous lecture on cloud economic and but ah we will see that ah we will look at some of the ah problems some troy problem so that it ah makes our um concept little more clear right so i just ah i have couple of slide ah recap
• 01:00 - 01:30 of over the earlier slide so that it will be easy um ah for ask to ah get tune to the problem right so as you have discussed ah in cloud ah property if you form the economic point of view its a common infrastructure like pooled standardized resource with benefits generated by statistical multiplexing right so we are multiplexing the ah demands so that is ah form a pooled resources we can do location independent online connectivity that is broad
• 01:30 - 02:00 ah network axis utility pricing like ah you uses sensitive or pay per use or pay as you go model this benefits applying environments with variable demand levels session so on and so forth and on demand resources so it is a scalable elastic resources provisioned de provisions without delay or cost in the associated with change so ah overall it ah looks like that ah its a win win situation that ah whatever we do
• 02:00 - 02:30 with ah is is all all positive things but whether it is ah a company or a organization needs to look into that whether it is a beneficial for its ah for to deploy cloud depending on its type of things as we discussed suppose you require system a some some high end system for a short period of time ah then it may be beneficial to go for cloud [bappose/suppose] suppose the requirement of the system is longer period of time daily ah you are using twelve
• 02:30 - 03:00 hours so or may be more than that and then it may not be always economical to go for cloud type of things so in economic point of view that will try to look at we have also seen this slides where we try to look at that ah overall ah costing modeled where ah cloud cost is ah like here the variables were there the dt demand dt is a demand a is function of time ah within a time periods zero to t t is the max of dt that is a peak
• 03:00 - 03:30 demand a is the average demand b is the baseline that is the unit cost for owning the things c is the cloud unit square the total ah cost is the denoted by ct total baseline bt so u or c by b is the utility premium right like the example you are shown in our earlier slide is utility premium is ah four point five five so that means it is all ah beneficial to use clouds so what is the utility similar so we can calculate ah we have shown that
• 03:30 - 04:00 ah we can calculate that zero to t u bdt as as a multiplication of this units similarly bt that is the total baseline cost is p into b into t because the baseline should be handled handle ah peak time right so peak demand rather so baseline ah whenever we do a baseline ah [ba/baseline] baseline costing of baseline ah design or what a whenever
• 04:00 - 04:30 you owned something then we usually look for a ah peak demand like suppose i want to big make system ah for our say mtech lab so what you have to say that if the number of students which can be admitted say in so number of system we think is is should be at least ten if not more to have some ah um redundancy into the things if there is a failure then ah recovery means we can replace the systems right but in actual things that the number of students strength may not we for a particular year of admission you may not attend that
• 04:30 - 05:00 thing right usually ah it may be less than that and so you have redundancy resources ah which are underutilized or not utilized so when the when cloud is cheaper if this cost of ct is less than bt or utility is p by a ah is less than p by a or here if you have the cost of cloud ah by this the what should be the utility premium right when you
• 05:00 - 05:30 utility premium is less than the ratio of the peak demand to the average demand then we can say it is cheaper now utility pricing in the real world is not that simple if they are lot of other things like in practice demands of in highly spikes are there like a there are maybe news stories or marketing promotion and so on and so forth of an hybrid model is based you own a car for daily use where are you for for a for whenever you are going out or traveling or for a longer distance you use those ah rented
• 05:30 - 06:00 things similarly you have your own system but when you are you require some when you are thinking there is some peak demand you go for this type of things so key factor is again the ratio to peak to average diamond right so that is the key factor but we should also consider other cost which are not considered here like network cost interoperability over head if to data are talking to each other then interoperability over head right ah then consider reliability accessibility and so on and so forth right
• 06:00 - 06:30 so those are other things which need to be accounted for when we calculate this cost so ah we also have seen this is a slide again where there is a ah on demand services value of on demand services when owning your own resource you pay penalty whenever your resource do not match with the instantaneous demand right either you penalty for unused resource or suffer for penalty for missing the service delivery right it can be there so if dt is
• 06:30 - 07:00 the ah instantaneous demand and rt is the instantaneous resource then the penalty cost is dt minus rt integral of that right like ah dt minus rt mode integral of that over ah the time period zero to capital t so that is the thing if the if demand is flat penalty is zero if the demand is linear periodic provisioning is acceptance so if the [d/demand] demand is linear periodic provisioning is accepted but most of the case there is a lag
• 07:00 - 07:30 between when the demand rises and the provisioning is done right and this sometimes creates a it may create a big challenge because of that if the growth of demand is exponential then chasing the demand with resource provisioning ah may be extremely difficult like here we have seen that if the demand is exponential dt is some form of e to the power t any fixed provisioning interval tp like it whenever you look for that extra provisioning it will
• 07:30 - 08:00 look for some provisioning interval tp so so ah arrow any according to the current demand will fall exponentially behind it right we are seen so rt equal to e of t minus tp so you have a lag between that right so dt minus rt if we calculate something constant in to the into e to the power t so that means the penalty is something c into k one e to
• 08:00 - 08:30 the power t that means it also grows exponentially like if you see the picture so that due to this provisioning delay that it goes on unserved demand goes on increasing so it is extremely difficult to chase this sort of situations right so these are all practical consideration so it is not that ah just ah doing a ah simple calculation but there are different other consideration for that you need to ah have a predictive model that whether you can say that this time the demand will increase like
• 08:30 - 09:00 these are true for our other commercial things also we do predict that during some seasonal things whether ah if the demand will increase on that time and have store accordingly here also you need to ah have some predictive model to do that if if at all required right now we will try to look at one or two problems ah which will allow us ah to have little more clarity on the whole ah thing so it is this is are very troy problems simple problem but
• 09:00 - 09:30 it ah by doing so it will give a some sort of a ah clarity or confidence into the things so what it says consider a peak computing demand of an organization is one twenty units right something some one twenty units the demand is a function of time that can be expressed at this dt equal to fifty sin of t from t zero to pi by two whereas twenty sin of t where pi by two to pi this is the some functional model is given the resource provisioning by
• 09:30 - 10:00 cloud to satisfy current demand t is ah rt equal to dt del of capital like a del of ah ddt of by dt right so this is the [reso/resource] resource provisioning where ah del is a delay in provisioning the extra computing resource in time right this way you consider now the cost of provision unit cloud resource of for unit time is zero
• 10:00 - 10:30 point nine units right so we need to calculate the penalty so our assumption is that the delay in provisioning is pi by twelve time units and minimum demand is zero there is no demand is the minimum case penalty either pay for the unused resource or ah you miss some del ah service delivery right so this is the consideration so we need to calculate
• 10:30 - 11:00 the penalty so let us just have a very ah state for work work ah working and see that how things works so if i have your r of zero to pi integral zero to pi from that ah given equation you can do right so if we look at this portion
• 11:00 - 11:30 so it is del of dt thirty so on other sense this we can have
• 11:30 - 12:00 so if you go on calculating so it is fifty minus cos of t zero to pi by two plus twenty
• 12:00 - 12:30 minus cos of t pi by two by pi so if you go calculate so it will come as seventy right
• 12:30 - 13:00 so that ah what we have finally finally if you look at so r equal to seventy plus thirty
• 13:00 - 13:30 pi by twelve and your d value is seventy so r minus d mode is thirty pi by twelve equal
• 13:30 - 14:00 to seven point eight six so this is the value what we answer will ah get so if you actually
• 14:00 - 14:30 this corresponds ah as as we ah if you remember as we are talking about penalty penalty equal
• 14:30 - 15:00 to is proportional to zero two pi demand minus mode of dt so we get these as the ah value
• 15:00 - 15:30 of the penalty because in this case what we say that is the ah your resource availability ah that provisioned is greater than your ah the diamond so we get a positive value right
• 15:30 - 16:00 so this is ah again from the equation we calculate again i should say this always calculation ah path we are not considered the other factors like network provisioning or other ah type of ah maintenance cost and all those things right we consider those are ah those we are not considered right so rather if we ah look at another problem like consider the peak computing demand of an organization is hundred units right the demand is a function that
• 16:00 - 16:30 can be ah expressed and that can be expressed as ah ct equal to zero to t u into b into dt this and like this so this is the function what we have discussed so what we say ah the cost of cloud ct equal to zero to t u cross b into this we have the already seen right
• 16:30 - 17:00 so rather here the this is the total cloud cost this we have already seen actually if
• 17:00 - 17:30 we ah please note the demand function which is ah there is ah it is not mentioned here there is some ah typo so demand expressed as a time can be expressed as ah five ah this fifty into one plus e to the power minus t ok
• 17:30 - 18:00 so let me reframe the question again consider a peak computing demand of an organization is hundred units ok so the demand as a function of time can be express at this so if i say the demand dt is expressed as this so this is not there in the question because there is a some ah missing terms so demand you considered one plus e to power minus t so this is a demand
• 18:00 - 18:30 is a function of time now what we additionally give baseline unit cost is one twenty so base line that you means in unit cost is one twenty right and ah cloud unit cost two hundred right so what we need to do in this situation is cloud cheaper than
• 18:30 - 19:00 owing for a period of hundred unit times right so you want to calculate if the cloud is cheaper then owing it into hundred we are obtain so what you need to at here this ah you need to calculate that whether if hundred unit time if it is a utilizing is there so it is again straight forward so if i see total baseline cost bt equal to p into b into t equal to
• 19:00 - 19:30 hundred into one twenty into this much right total cloud cost c of t equal to zero to hundred
• 19:30 - 20:00 c d of t dt equal to integral of zero to hundred two zero zero fifty one minus one plus minus this into dt so if you do a integration so if you do it
• 20:00 - 20:30 it approximately count as ten thousand one zero so much right if you just ah do that calculation so what we get ah here what we get one two zero one two three four five one
• 20:30 - 21:00 two three four five fine so what we do utility right so utility premium as if from our this
• 21:00 - 21:30 this zero point eight four or utility premium u what we get is u is less than one so that means at least in this case cloud will be cheaper right
• 21:30 - 22:00 ah so ah i i can have simple calculation to find out that what are the different for different scenarios what are the different ah cases are there right let us look at another ah problem right here what says a company x needs to support a spike in demand when it come when it become demand when it becomes popular followed by potentially a reduction once some
• 22:00 - 22:30 of the visitors turns away that means when when that there is a there can be spike and fall in the demand right this sort of scenarios right the company has two option to satisfy the requirement which are given in the a table so what it says purchase cost of in house is something and ah cloud server in is in a way there is no purchase cost ah number of cpu cores for in house server is twelve where at the cloud server is ah eight cost per hour over three year span for the cloud
• 22:30 - 23:00 sever as you are renting it is something forty two units per hour cost efficiency of the in house server because it is underutilized so it is forty percent whereas the cloud server is eighty percent when you because you are using it when the demands where power and cooling requirement in case of ah in house server is twenty two whereas for the cloud
• 23:00 - 23:30 server is nil and management cost per hour is six in case of in house server and cloud server it is one so again let me as repeat purchase cost is six lakh for in house and ah nil for cloud ah number of cpu core in in house is twelve cloud eight cost per hour in a three year span that in ah or other typo ah means for
• 23:30 - 24:00 cloud is forty two units whereas ah for in house there is no cost hour you can after on purchase the thing if you since is forty percent because most of the things are ah there is sixty means not that utilized and whereas cloud sever it is eighty percent power and cooling is twenty two and nil and management cost per hour in case of a cloud is much more that is six and it is one so what you need to calculate with this data calculate the price of a core hour on in house and cloud sever ok so that is straight forward
• 24:00 - 24:30 so what we have cost per hour for in house sever considering three years is divided by three cross three six five cross twenty four so if you calculate it comes around
• 24:30 - 25:00 twenty two point eight three inr if it is indian currency or units of cost right now if we have ah if we have a cost hour for in house
• 25:00 - 25:30 in house server it is twenty two point eight three by twelve right because the first question was core hour for in house server so core hour for in house server is this one is one
• 25:30 - 26:00 point nine zero unit or if we consider a new thing so core hour for cloud server is forty two by eight it directly comes from the question five point two five inr right now an question b find the cost per effective hour for the
• 26:00 - 26:30 both the option so cost per effective hour for house is twenty two point eight three and as it is forty percent efficient so forty by hundred right so it is fifty seven point
• 26:30 - 27:00 zero seven five so it is inr or something ah whatever the unit effective hour for cloud equal to it is eighty percentage efficiency so five right so this is the ah effective and now calculate the ratio of total cost per effective hour for in house to cloud deployment
• 27:00 - 27:30 right so i am need to calculate the ratio of total cost per effective hour for in house and cloud deployment so total cost per effective hour in house is fifty seven point zero seven
• 27:30 - 28:00 five if you you have calculated earlier ah that is for the in house sever plus twenty two because that is a power and cooling plus six for maintenance ah management cost and then we have eighty five point zero seven five same thing for cloud equal to fifty two
• 28:00 - 28:30 point five and it has only one management so it is fifty three point five and if it is inr find or whatever the unit right so ratio of effective hour of in house by cloud equal to eighty five point zero seven five by fifty three point five so one point
• 28:30 - 29:00 five nine and finally if the efficiency of in house sever units is to seventy percent which deployment have ah will become better in the total cost effective hour right so it has efficium efficiencies ah seven ah with initially now forty if you increase to seventy
• 29:00 - 29:30 percent so ah which will be better right so ah modified so efficiency modify not efficacy modified cost per plus effective hour for in house will be twenty two point eight
• 29:30 - 30:00 three by seventy percent now hundred so we get thirty two point six one so total cost plus effective hour for in house everything is for in house total cost per effective hour still it is for in house still it is thirty two point six one plus twenty
• 30:00 - 30:30 two plus six right equal to sixty point six one right but if we see that ah the cloud it is still fifty three point two five right now if it is instead of this is seventy percent increase if it is a ninety percent in house then if you do the same calculation we will
• 30:30 - 31:00 effectively what we will get instead of this value so twenty two point eight three by ninety by hundred right it is twenty five point three seven right so with this total cost for ninety percent will become twenty five point three seven plus twenty two plus six fifty three point three seven right inr so this is for if it is ninety percent right
• 31:00 - 31:30 so if you see ninety percent then it is better than this cloud so it is if if see if you look at the problem so the if the if if in your efficiency is less right so you are as we discuss the overall utilization of the resource of the in house is highly underutilized
• 31:30 - 32:00 so you lose on those things right so effectively we get a less ah it means we get a more benefit if we take form cloud however if your efficiency increases or in other sense that you are your ah infrastructure or your ah in house infrastructure is if it is heavily utilized like a we have seen up to ninety percent then it it may be better
• 32:00 - 32:30 then ah it will be performance wise better than cloud right this is a very synthetic for example to so that if the efficiency if you are more utilization are there then ah in this case is the in house will be better then cloud type or cloud purchasing a cloud so it all depends ah on overall ah um what what is your demand what sort of demand is
• 32:30 - 33:00 there how long duration things are there so we [nee/need] need to take a call that whether you go for a cloud type of ah um economically beneficial to go to cloud service provider or have in house ah infrastructure so this three is ah um small problems on economic model of cloud ah shows that that ah that how ah by simple calculation we can find out
• 33:00 - 33:30 and though in actual practice these are more complex with considering other for parameters as we have discussed ok so hope ah this will help you in ah clearing or having a better understanding of this economy behind this cloud ok thank you