Lecture 24: Resource Management - I

Summary

Highlights

• Cloud offers consumers as many resources as they need, functioning on a pay-as-you-go model. 💸
• Providers must manage resources well to achieve profitability and cover their environmental obligations. 🌱
• Different resource types require unique management strategies, such as infrastructure-as-a-service. 🛠️
• Green computing strategies like VM scheduling help reduce energy usage and carbon footprints. ♻️
• Optimizing data center designs can significantly mitigate power consumption. 📉

Key Takeaways

• Cloud computing provides a scalable, seemingly limitless resource pool. 🌤️
• Efficient resource management is crucial for both user satisfaction and environmental sustainability. 🌍
• Balancing profitability and maintaining quality service is a key goal for providers. 💡
• Infrastructure, platform, and application levels have different resource management challenges. 🔧
• Advanced VM scheduling and data center designs can improve energy efficiency. ⚡

Overview

Chapters

• 00:00 - 00:30: Introduction to Resource Management The chapter introduces the concept of resource management in cloud computing and similar service-oriented architectures. It emphasizes the importance of efficient management within distributed systems and cloud environments.
• 00:30 - 05:00: Cloud Computing Basics The chapter titled 'Cloud Computing Basics' introduces the concept of cloud computing. It explains that the cloud is conceived as an infinite resource pool that can be leveraged for resource management. The key benefit highlighted is its scalability and ability to provide services to multiple users simultaneously, presenting an advantage from both the consumer and provider perspectives.
• 05:00 - 10:00: Importance of Resource Management The chapter titled "Importance of Resource Management" discusses the concept of resource allocation based on a pay-as-you-go or metered service model. It emphasizes the significance for providers to utilize resources efficiently and effectively. This ensures better service to consumers and addresses the challenge of managing finite resources.
• 10:00 - 15:00: Types of Resources in Cloud Computing This chapter delves into the different types of resources available in cloud computing. It emphasizes the dual perspective of cloud resources, focusing on both providers' and consumers' needs. Providers aim to optimize their offerings to maximize profits while maintaining high-quality services and adherence to service level agreements (SLAs).
• 15:00 - 20:00: Challenges in Resource Management The chapter titled 'Challenges in Resource Management' discusses the consequences of violating Service Level Agreements (SLAs). Such violations can lead to significant penalties, highlighting the importance of adhering to agreed-upon terms. Another crucial aspect addressed is the environmental impact of resource usage. Even if a provider has sufficient financial resources or access to resources, there are environmental limitations, such as carbon footprint considerations, that must be taken into account in resource management strategies.
• 20:00 - 25:00: Environmental and Economic Aspects This chapter focuses on the intersection of environmental and economic aspects, particularly in the context of resource management in cloud computing. It highlights the necessity of managing energy resources efficiently due to restrictions and obligations associated with energy consumption. The chapter sets out to explore various resource management strategies and their implications in cloud computing, suggesting a series of lectures will further elaborate on these topics.
• 25:00 - 30:00: Importance of Green Data Centers The chapter discusses the various levels of cloud computing, beginning with the basic infrastructure which includes compute resources, storage resources, networking resources, and related resources. It then touches upon platform services that reside above the infrastructure layer, highlighting the concept of infrastructure as a service (IaaS).
• 30:00 - 40:00: Strategies for Energy Efficiency The chapter discusses various service models in cloud computing with a focus on Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). It emphasizes the concept of 'resource' which is primarily associated with infrastructure but can also manifest in different forms. The discussion centers around resource management within the cloud environment, highlighting the significance of infrastructure as a service within the overall operation of the cloud.
• 40:00 - 45:00: Example Scenario and Nonlinear Power Consumption The chapter discusses resource management in cloud computing, emphasizing the role of infrastructure as a key component. Customers access the cloud using various heterogeneous systems, including servers, laptops, and mobile devices such as smartphones and tablets. The focus is on how these different devices connect to and utilize cloud resources.
• 45:00 - 55:00: Virtual Machine Management The chapter titled 'Virtual Machine Management' discusses the concept of cloud services and how they are integrated into other services. It touches upon the idea of leveraging cloud services or service providers to enhance capabilities. The transcript encapsulates the process of managing and optimizing these services for improved efficiency and performance.
• 55:00 - 60:00: Energy Saving Techniques The chapter titled 'Energy Saving Techniques' delves into the complexities of overall operations management, emphasizing its challenging nature. The primary focus is on exploring different facets of resource management with an aim to utilize existing hardware and other available resources more efficiently. This pursuit of efficiency is identified as the fundamental objective. The chapter identifies resource types, such as physical resources, with a specific mention of computers, thereby underlining the importance of managing these resources efficiently.
• 60:00 - 64:00: Conclusion This chapter, titled 'Conclusion', delves into the integration of hardware systems designed for databases and encompasses other logical resources. It discusses the application of communication and other processes that execute on CPUs, emphasizing the importance of networking and management of scientific instruments. The chapter reiterates that databases are soft resources vital for various applications and system operations.

Lecture 24: Resource Management - I Transcription

• 00:00 - 00:30 hello so today we will discuss one of the one of the important aspect of a cloud computing or any type of this sort of service oriented ah um mechanisms like ah like cloud or distributed
• 00:30 - 01:00 system or agreed or a any type of systems that is the resource management right so what we are look what we are thinking of a cloud we are thinking that cloud is a infinite resource pool along with that ah we are ah basically um ah leveraging these resources to multiple users right so from from the customer point of view or the user point of or the consumer point of view that it is a something a scalable service
• 01:00 - 01:30 and a consumer can get as many as resources based on something ah of the concept of a pay as you go or meter services right so from the provider yield it is important to ah properly utilize the resource in both terms that it can serve the a consumer in a better way and in other sense also it which is which is a limited or a what we say finite amount of
• 01:30 - 02:00 resource it can serve as many as of the as many as customer ah to the as many as consumer of the customer the thing is that a from the providers its a there is a business angle also right he need to ah optimize the ah or it it to maximize his profit ah with ah without compromising on the quality of services or ah the sla um
• 02:00 - 02:30 violations right as we have seen if there are violation of slas it has a long implication where has to um ah give penalty and so and so forth there is a another aspect of the whole thing right even if i have ah even if the a a particular provider has ah say ah money or has a can a can afford number of resources but there is a limitation or there is a overall environmental aspects of it right how how much of of your carbon footprint how
• 02:30 - 03:00 much energy resource you are having so it is it is there is a restriction or there is a obligation on that mind of so keeping all this thing into mind this ah this resource management plays a a important role in the thing so what we will do in a couple of lectures we will to look at that what are the different aspects of a resource management ah in the ah in the context of cloud computing so if we look at different type of a resources ah
• 03:00 - 03:30 in in a typical cloud computing things so what we have as ah we have mentioned we have infrastructure ah right ah so basic infrastructure where is a compute resource storage resource networking resource or a related resource on the things there is a platform ah which ah is above the infrastructure or here if i say that infrastructure as a service is
• 03:30 - 04:00 the major thing then we have a platform as a service and then at the end we have the application as the resource right so the resource can manifest in a different way it can be the infrastructure though a primarily when we talk about resource or resource management we usually fall back to infrastructure as a service but considering ah the a overall ah cloud or overall operation of the cloud so basically at least we are having infrastructure as a service um a platform as a service or software as a service also we have infrastructure
• 04:00 - 04:30 as a resource management thing or platform as a resource management thing and application but though prominently a infrastructure plays a bigger role so these are the different type of resources and the customer basically ah hook into the cloud with different ah type of heterogeneous systems it can be servers it can be laptops maybe a smartphones tablets
• 04:30 - 05:00 desktop and so and so forth number of cases a cloud ah i can we can say that a cloud or a service can take the cloud service or service providers ah thing to leverage on other things right i ima ah ah i provide service taking some service from a service provider and so and so forth so overall this management and optimization
• 05:00 - 05:30 of this overall operations is a very tricky one so what we try to look at a today is that what are the different aspects of the things as at all whether with the same type of a hardware another things available is it possible to manage the resource in a efficient way that is our ah basic ah um call for this for for this particular resource management so as we have seen resource type can be physical resources like as we ah as we see that computer
• 05:30 - 06:00 disk ah memory databases ah network scientific instrument so database is again it is a soft resource ah so ah what we say that integrated this the means that hardware systems meant for the databases and also there are other logical resources like like what we say there are applications like communication applications and other applications executions of some ah processes which execute on the cpus monitoring
• 06:00 - 06:30 and management of the things right so these are monitoring tools management tools etcetera these are different other resources some resources are directly utilized right like if i say that ah cpu and a hard disk and other things some are basically meant to manage those ah resources so if you look at in a broad term like a what
• 06:30 - 07:00 is what what we are trying or what ah if we say the resource management what we want to do with a resource management the term a resource management refers to the operations ah used to control how capabilities provided by the cloud resources and services ah there is a typo can be made ah available to other entities ah whether users applications services in an efficient way right so what we want to do that ah whatever the resources and services
• 07:00 - 07:30 that means hard and soft resources are available with the cloud how it can be made available to the external entities like ah it may be a user it may be a application it may be a services in a efficient way so there is quote unquote the term efficient is ah tricky but ah efficient means it can be efficient in ah maximizing the profit of the isp it can be efficient in energy optimization or in ah efficient in giving or a respecting
• 07:30 - 08:00 slas and providing best quality of services right nevertheless the resource management plays a ah role in all those things now ah if we look at ah data center ah a power consumption so what we are ah referring that is a isp or a internet ah is a a csp or a cloud service
• 08:00 - 08:30 provider so they are ah they are having data centers right like major ah providers like ibm google yahoo or so there are several things like microsoft ah amazon and so and so forth they have a huge data centers across the world right as we have seen so based on the things that ah these resources are outsourced from this data sensor to be more specific so what is the power thing so there is some the report it is it may not
• 08:30 - 09:00 be the very recent devote it is much more higher than the whatever we are talking about so it is ah it is estimated that ah servers consumed around point five percent of the worlds total electricity uses right so its a huger node now if it is it is not only the server right if you look at a data center so data center is ah if you look at it there are two major component one is that compute compute infrastructure another is what we
• 09:00 - 09:30 say overall environmental infrastructure right or data center ah other logistics right ah what we say en viron ment or logistics of the data center right so what we want to do these basically gives you say servers or say storage and maybe adjoining
• 09:30 - 10:00 thing like network etcetera these are providing primary looking for your power management or power supply or air conditioning right and ah other ah um related stuff which is
• 10:00 - 10:30 related to this data center things so if i have a dc so it has two component incidentally this two have if not equally power hungry things right so if you look at the amount of power ah so here what we what we ah so if we make it so we have one if we have dc one component is towards compute one component is towards maintaining the overall environment
• 10:30 - 11:00 of the data center so here where what we having a servers storage network and other other whereas here ah primarily your power management maybe ups sort of power
• 11:00 - 11:30 then ah ac ah and other ah overall logistics arrangement or um other ah means a different type of setups and etcetera to have the house the data centers right now there is a power requirement out here there is a overall power requirement out here right so if you look at the and not only that this is not only power there is a power come place power comes
• 11:30 - 12:00 place so if you look at the overall things are if if i say the overall consumption here is x so it is it is also somewhat to towards x so it is the if it is not only the servers but the consumption by the other ah other type of other logistic units are equally high so when we talk about data centers where where is point five ah percent of the world total uses so if you if you look at if you basically
• 12:00 - 12:30 ah make provision for other things it may go above one one percent of the things right over ah over power and this is ah some data which is ah reported ah in some places but it may be much higher so and it is and it is also saying these servers demands ah when some report says that energy demand for the servers are increasing ah every five to six hour a six years maybe becoming double there is a large volume of carbon dioxide
• 12:30 - 13:00 footprint and which ah because of this fuel ah burning of this fossil fuels need to reduce energy use for the minimal a performance so what do you what is our goal that we need to reduce this energy utilization with minimal minimal ah effect or the compromising this performance right so so that should not be practically or that should not be very minimal impact on the performance so that is one of the goal
• 13:00 - 13:30 so as you are ah discussing so the overall motivation if we say that one to make quote unquote green data centers which will take very less energy perform at the highest level so there are different aspects so one is this economic aspect the new data centers run on megawatt scale require millions of dollars or millions of a a money to operate recently institutional looking for new ways to reduce different type of new ways to reduce cost
• 13:30 - 14:00 many facilities are their peak operating stage cannot explained without a new power source in some cases like ah if i have even thats why i was talking that even if i am having a fund available as a provider but i dont have any resource to do that right suppose we have a data centers at the kharagpur and we are taking power from the state electricity board and even tomorrow we say that we want to scale nothing to double and we are ready to pay but the state electricity board may not have that type of ah surplus power to
• 14:00 - 14:30 supply so it is it there is a ah it is not only that i whether i am having money or ah having ah um more space etcetera but whether that can be supplied there is a definitely environmental aspects and there are different type of legal ah obligations ah which ah for towards environment comes into play majority of energy sources are fossil fuels still to date huge volume of carbon dioxide emitted each year from power plants which creates
• 14:30 - 15:00 a ah environmental hazards sustainable energy sources are not ah till not ah ready to ah handle such huge requirement of the data centers need to reduce energy dependency so we need to have some mechanism to reduce the energy dependency so this is the ah two very very a broad aspects of the theme so whether ah we are talking about green
• 15:00 - 15:30 computing in this respect definitely so what what we can do what what a cloud providers are tries to look at ah is that ah whether these advanced scheduling schemes to reduce energy consumption is possible right so that ah which is which ah can be power aware right ah or ah ah or an ender or thermal aware both power aware and thermal aware these sort of things performance per watt is not follows the moores
• 15:30 - 16:00 law like our famous moores law but it is unlikely that performance per what is not following the moores law so data center design ah so that whether we can redesign or basically design the data center to reduce power uses effectiveness so that ah we can have effectively use the power like it can be cooling system or the rack design or the placement of the racks and type of things that is more of the data center design as we are talking just a couple of minutes back that one is that having the server its at the compute part
• 16:00 - 16:30 of it another is that data center design aspects of it that how efficiently you can design and if you those who are ah um ah visited some some sort of data center you know that there are different typical arrangement of the cooling it is not conventional cooling of the may not be the whole ah server room of the space it is it is a a separate enclosure made on the data on the major server racks and the server racks are cooled ah specifically
• 16:30 - 17:00 very ah on a a very ah confined area so that the energy required ah for this cooling purpose can be minimized so there can be different type of research direction and some of you ah who are interested may be doing ah some research on the cloud computing so there is a there is a important this ah this resource management looking at different aspects is
• 17:00 - 17:30 a extremely ah what we say um hot topic to do research there at different parts of the world people are working on this that how to how to optimize this ah resource ah how to have a optimal resource utilization for without compromising a services or without compromising the um performance of the cloud so how to conserve energy within a cloud environment so one may be scheduled vms to conserve energy whether we can have proper vm scheduling on
• 17:30 - 18:00 the conserve energy so management of both vms and underlining infrastructure minimize operating inefficiencies for non essential task right so that ah whether we can minimize that ah operational efficiencies or non essential like if you you take a vm with a particular flavor and lot of resource a lot of other ah packages etcetera running but practically
• 18:00 - 18:30 you dont require all those things right whether we can have a tailor made vm or tailor made ah os ah which can be run on the thing right optimize data center design the other aspects now ah now this these are so one is aspect aspect of scheduling the vm one aspect of a ease management another aspect is design design of the data centers and so so these are three broad aspects we will try to have a quick look ah on this aspect so that we can have ah we can have a feel of it right so if you look at ah this overall green cloud
• 18:30 - 19:00 framework so one ah one side is ah towards vm controls right how these ah vm scheduling management other side is data center design so this is more on the compute side of it this is more of the data center infrastructure side of it so here the infrastructure is the hard data center infrastructure the other come so if you look at the um vm controls
• 19:00 - 19:30 so one part is scheduling the view so which can be power aware or thermal aware looking at the things other is that vm management which is vm image designing right when whatever the vm image having whether we can have a efficient design of this ah vm image with um with the basically only those packages or the those ah services which are required
• 19:30 - 20:00 are uploaded at there so other is that ah that is whether you can look for vm migrations right so what what we are ah having ah actually if you look at it so what we are having a underlining hardware and they are vms are there suppose there are ah typically say sixteen blade server so every blade server is running say ah eight vms so sixteen into eight is the total number of vms are there right so now ah out of the sixteen into eight or say ten vm so sixteen into ten ah vms are runny or say ten maybe on the higher
• 20:00 - 20:30 side so it sixty into five say eighty vms are runny now eighty vms ah if there are at at any point of sign say ten or twenty vms are active whether ah it is it is efficient to distribute over the whole ah vm servers available or i can concentrate the ah vms into couple of servers right so ah because whether the whether the ah one
• 20:30 - 21:00 is that whether the server a a vm is running on a server or not it will go on consuming a base energy right whenever the whenever i run ah um a vm it may consume some incremental energy so based on that it requires some sort of a ah simple mathematics ah to work out that whether it is efficient so one the data center design side one is that server rack
• 21:00 - 21:30 design is a important aspects and there is separately if you look at do not the cloud computing part this is another ah research on the building of of data center or infrastructure type of things we are conditioning recirculation is a type of a another aspects of the things so there are if you look at that management side there is another aspect of dynamics shutdown if the things are ah not utilized so whether i can dynamically shutdown there so there are a lot of aspects and if you see these are all have lot of research potential especially
• 21:30 - 22:00 this part of the ah things on the compute side so this is a typical ah example scenario if you see there is a nonlinear relationship between the number of processor used and the power consumption so it is it is not like that that the number of processor used in the power consumption is a linear thing so if the number of processor goes on high the power consumption is basically somewhat going towards some saturation if not saturation
• 22:00 - 22:30 the increment is not linear so whether we can exploit this one whether we can schedule vms to take advantage of this relationship in order to ah conserve power right so that means i as i was talking that i want to concentrate my vms into the ah underutilized server so that the number the number of servers the vms running on the number of servers are reduced so that effectively ah i can ah put this ah non ah that is idle servers into sub
• 22:30 - 23:00 hibernate or mode or low power consumption mode so that the overall power consumption of my data center reduces so that schedule as many vms once on a multi core node greedy scheduling algorithms ah we can do there is a a snap shot of a greedy scheduling algorithm you can basically ah try out and see that how things works keep track of the core on
• 23:00 - 23:30 a given node so we have to see that ah at a way for it typically in a node that how this um ah codes are ah being ah busy or how many ah how much loading is there match vm requirements with the node capacity if it is there then whether we can migrate the thing right a a simple ah very vanilla type example so if i have
• 23:30 - 24:00 four nodes right so that ah it is ah in a idle condition or when a no load condition they consume one zero five watt and when when it is loaded fully loaded with eight vms it consumes say one seventy watt so if i have some eight vms then ah if we if we run on a one node the overall consumption is one seventy plus one naught five star three so
• 24:00 - 24:30 that is the overall consumption four watts right in this case this typically looking at the things when ah if i had it been it is distributed in thing so with the this sort of two vm lets say that is one thirty eight so i have four so in this case five fifty two watts right so that exactly what want to see that in doing
• 24:30 - 25:00 so we can basically ah reduce the things but there is there is a little catch in it now in order to achieve from this stage to this stage i need to migrate this migration also have some cost right migration has some cost not only that they here we are taking all vms to be the same ah category right so vm can be of different categories so it may not be able to ah put all the vms into a one node or two nodes etcetera so we need to check
• 25:00 - 25:30 and monitor that how the how much loading is there how much free vms are there so that we can basically migrate the vms right so it is not like straightforward of multiplying that to what age it is also this migration has some cost on doing that so what we need to do monitor cloud uses and load when a load decreases live migrate vms to more utilize nodes un ah known right shutdown unused node so that i can basically basically what we
• 25:30 - 26:00 are doing we are compacting those ah things into the server right so that is ah the live migration now number of ah these ah your hypervisor all say number of vmm ah do support this type of live migration though ah it may be commercially costly but you can do a live migration and it has lot of implication other way also like if they if if a server is at ah um having
• 26:00 - 26:30 problem then you like migrate to the other servers and so on but that is those are things are possible so when load increases so we can have ah some sort of wakeup call right wake up on land sort of things to start the waiting node so in the load increases i basically give a wakeup call to the nodes so which are slipping to wake up schedule new vms to the new nodes so there are they are technologies available and if we can be efficiently use effectively we can have a energy efficient thing right
• 26:30 - 27:00 so that it is node two it is migrated there the vm is ah put into service then the node two become idle then made into hibernate or offline mode right so effectively i am running on a node one so it is this though there is a cost of this sort of a migration but if if ah overall ah if i am achieving efficiency or in terms of power consumption then ah that may be a good option to look at so another ah aspects what
• 27:00 - 27:30 we have seen that management why is that whether i can ah minimize ah vm instances right vm machines as sometimes too loaded so lots of unwanted packages ah unneeded services are there right ah so you take some ah while vm it comes with a ah basic ah configuration there which may not be utilized by the ah consumer so ah our multi application behaves
• 27:30 - 28:00 on multi application oriented not service oriented in number of cases clouds are based on of a service oriented architecture so it is ah we should ah have service orientation need to customize lightweight linux vm for service oriented science or services right so that we can need to have a customized ah linux vm or the customized vm per say right need to keep vm image as much as possible to reduce a network latency so if we have
• 28:00 - 28:30 the vm image less than the ah means ah while carrying of the network the latency will be minimized so these are different aspects so this is a typical ah example scenario with a which starts with a ubuntu typically ah nine point zero four remove all packages which are not required like if i dont require x eleven or windows manager etcetera etcetera read ahead profiling utility reorder the boot sequence so that i can have a pre fetch boot
• 28:30 - 29:00 files on the dicks minimize cpu idle time due to io delays etcetera so i can have a read ahead ah things like if i have that ah the the steps to be followed i can do a a priori thing optimize linux kernel build on say in this typical case we dont know domu so there is no three d graphics no sound so that i can have a customized linux kernel which is primarily used for the things so based on the on the type of customer requirement or based on the services i can basically we can basically to optimize this type of stuff
• 29:00 - 29:30 so there are ah different ah energy saving some snaps or parameters like ah we can ah if if it is reduced the boot time from thirty eight to eight seconds so effectively we can have energy savings even if a small cloud where hundred images per hour so so much energy savings can be there in a production cloud where thousand images are created every time saves a lot of energy in other way a image size from four gb to
• 29:30 - 30:00 something six hundred plus mb so reduces ah life performance as we are telling that the life migration also is a comes with a cost and we can do better so what we try what it is trying to ah say that all those things ah makes its energy efficient or have a proper resource management ah resource manager without compromising the ah actual um performance
• 30:00 - 30:30 so if we try to summarize so its say emerging topic definitely so need to conserve energy wherever possible so one is that green cloud framework power aware scheduling of vms advanced vm and infrastructure management specialized vm images or customized vms images so small energy savings result in a large impact right so even it is a ah apparently vm wise of ah node wise small but considering the whole thing is a large impact
• 30:30 - 31:00 combining a number of different methods together can have a larger impact than when implemented separately so it is not only piecewise having a ah collective or cooperative way of ah looking at it and ah there are a lot of ah research interests and things combined concepts of power aware and scheduling power aware and thermal aware scheduling to minimize both energy and temperature ah integrated server rack cooling strategies these days as our
• 31:00 - 31:30 server racks are coming with integrated cooling power and so and so forth ah further improved vm image minimizes and looking at the customer needs and type of things profile customer profiling designing next generation cloud computing system to be more efficient ah both in terms of energy thermal and ah type of service provide provided so with this ah let us conclude today and as we understand there is a ah there is a
• 31:30 - 32:00 lot of scope of a doing research and for the studies ah in this particular way of resource management thank you