Lecture 25: Resource Management-II

Summary

Lecture 25 of the Cloud Computing series delves into the complex topic of Resource Management in cloud environments, emphasizing the importance of strategically managing both physical and logical resources to optimize usage, cost, and energy efficiency. The lecture outlines various resource management mechanisms such as provisioning, allocation, adaptation, and mapping while considering market demands, user requirements, and technical constraints. Special emphasis is placed on balancing optimal resource usage from both the provider's and the user's perspective, ensuring quality of service and financial viability.

Highlights

• The lecture emphasizes the dual importance of resource management for providers and consumers in the cloud 🌥️.
• Resource management must account for physical and logical resources as interconnected components 🔗.
• Efficient resource provisioning ensures scalable and cost-effective cloud services 📊.
• Challenges include balancing user demand with resource availability and energy consumption 🔋.
• Techniques like game theory and intelligent models aid in dynamic and real-time resource distribution 🌐.

Key Takeaways

• Resource management is crucial for cloud success, balancing efficiency and service quality 👍.
• Both physical (CPU, memory) and logical (OS, APIs) resources matter in management ⚙️.
• Approaches include provisioning, allocation, adaptation, and discovery techniques 🔍.
• Providers must ensure cost-effectiveness and energy efficiency while users focus on service quality 💡.
• Game theory and network queuing models are used for optimizing resource provisioning 📈.

Overview

Resource management within cloud computing is pivotal, playing a critical role for both service providers who aim to maximize their resource utilization while minimizing costs, and service consumers who seek assured service quality and contractual adherence. In this lecture, various strategies are discussed addressing the need for efficient use of resources through a harmonious balance of infrastructure and logical management.

Physical resources such as CPUs and storage, alongside logical resources like operating systems and APIs, are addressed for their integral roles in the effective functioning of cloud services. A core focus falls on IaaS, the Infrastructure as a Service, which is renowned for its vast array of resource management challenges, including energy efficiency and optimal resource allocation.

Discussed strategies encompass adaptive resource management, game theoretic approaches, and intelligent multi-agent models, each promising solutions to the challenges of scalability, cost-effectiveness, and service quality. These strategies are fundamental in guiding cloud service provisioning and ensuring that, despite variability in demand and available resources, both service providers and consumers achieve their operational and business objectives.

Chapters

• 00:00 - 01:00: Introduction to Resource Management in Cloud Computing The chapter introduces the topic of resource management within the realm of cloud computing. It follows up on previous discussions related to resource management, delving deeper into its principles and significance in the cloud context. While the transcript provided is brief, it signals a continuation from prior lectures, suggesting a progressive unveiling of more detailed aspects in resource management.
• 01:00 - 09:00: Aspects and Challenges of Resource Management The chapter discusses the critical role of resource management in cloud services, highlighting its importance for both service providers and consumers. Providers aim to maximize resource utilization with minimal energy costs to increase profits, while consumers desire guaranteed service performance.
• 09:00 - 19:30: Resource Management Mechanisms and Approaches This chapter emphasizes the critical role that resource management plays in ensuring the success of cloud computing. It highlights that quality of service and adherence to service level agreements (SLAs) are vital aspects. Despite the brevity of the transcript, it suggests a focus on exploring various aspects of resource management mechanisms and approaches.
• 19:30 - 24:15: Resource Provisioning Approaches This chapter discusses various approaches to resource provisioning within cloud computing environments. It emphasizes that the discussion is based on a review paper, which aims to cover some critical aspects of resource management, although not exhaustively. The lecturer clarifies that the goal is to address substantial areas pertinent to cloud computing platforms. The introductory part of the chapter suggests that definitions and clarifications about what constitutes 'resources' in this context will follow.
• 24:15 - 27:30: Resource Allocation Methods This chapter discusses the core elements of cloud computing platforms, specifically Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS). It also highlights the varied types of users including human users and automated processes or machines that utilize these cloud services.
• 27:30 - 29:00: Resource Mapping Techniques The chapter 'Resource Mapping Techniques' delves into the optimization and management of resources at the core level. It distinguishes between two main categories of resources: physical and logical. Physical resources refer to tangible assets, while logical resources pertain to non-physical aspects such as applications and monitoring systems. Both categories are crucial for effective resource management. The chapter also revisits previously covered concepts related to resource management.
• 29:00 - 30:00: Resource Adaptation and Scheduling This chapter delves into 'Resource Adaptation and Scheduling', focusing on the fundamental definition of resource management. Resource management involves operations that control how cloud resources and services are provided and made available to users and applications efficiently. The discussion covers resource management mechanisms and approaches.
• 30:00 - 32:30: Metrics for Evaluating Resource Management This chapter focuses on evaluating the management of IaaS (Infrastructure as a Service) resources. While management of other resources like PaaS (Platform as a Service) and SaaS (Software as a Service) is necessary, it is largely influenced by the available underlying hardware resources.
• 32:30 - 34:00: Conclusion and Considerations The chapter discusses different approaches for managing Infrastructure as a Service (IaaS) resources. It highlights that some techniques are applicable across various services, while others are specific to IaaS. The transcript emphasizes that IaaS is considered to be the most popular cloud service type.

Lecture 25: Resource Management-II Transcription

• 00:00 - 00:30 hello so we will continue our discussion on resource management in ah cloud so as as we ah discussed last ah lecture or last ah we lecture on the resource management what we
• 00:30 - 01:00 have ah tried to look at that it plays a important role in overall ah cloud service right so it is important not only from the service provider point of view it is also important for the service consumer point of view right so provider want to have a maximize its utilization of his resources with minimal energy cost and maximizing it ah profit right if you look at the from the consumer point of view it wants to have a guarantee or a particular
• 01:00 - 01:30 quality of service or support for its sla right so that the sla is not valid so nevertheless this whole resource management he plays the important role for this what we say quote unquote success of this ah cloud computing paradigm right so taught today what we will ah try to look at some of the aspects of these ah resource
• 01:30 - 02:00 management right we will try to look at a particular ah for a a review paper and take up ah some of the aspects of resource management so ah i dont want to claim that release the all the aspects but these are some of the important aspects where what a particular ah cloud computing environment or cloud computing ah platform so look at right so couple of slides ah maybe imputation from the other so what we made by resources just
• 02:00 - 02:30 to recap quickly so i have at the core ah infrastructure platform and application or iaas paas saas and there are different kind of ah user across the means user for this clouds means they can be they can be either ah human user or it can be some process or machine which are indirectly consuming cloud service to the for other services so what
• 02:30 - 03:00 we want to look at that how optimize these resources can be managed at the at the core now as we have seen there are two categories of user physical two category of ah resources one is physical resource another is a logical resource ah so the physically what is there and logically like ah applications monitoring and type of things so both plays a important role in the resource management and also we seen these ah we have gone through these particular
• 03:00 - 03:30 ah um underlining definition of resource management it refers to operation used to control over capabilities are provided by the cloud resources and services and can be made available to other entities ah users applications in an efficient manner now if we look at ah this resource management ah um mechanism or resource management approaches
• 03:30 - 04:00 the the maximum or the the maximum or the major thrust is on the management of the iaas type of resources or infrastructural resources right other resources like platform or ah saas though they are also management is necessary but those are mostly dictated by the amount of underlining backbone hard resources you are having right
• 04:00 - 04:30 so some of these type of techniques are applicable across the different type of services whereas some of the ah things are more good to the iaas so what we look at today is more about that what are the different approaches for iaas ah type of ah resource management right so infrastructure as a service is the most considered to be most popular or seen to be most popular cloud service among these different type of services so in iaas cloud providers
• 04:30 - 05:00 offers resources that include computer ah as virtual machines raw storage firewalls load balancer network devices and so and so forth so these are the different category of things which we consider as when we talk about infrastructure as the resource and one of the major challenge in iaas is the resource management how to optimally manage the resource and as we have seen as energy plays a important role ah for
• 05:00 - 05:30 ah for overall functioning of the cloud so it one of the aspect is with minimal or the energy consumption how i can give service at a particular level right so that is that is very important when we talk about iaas ah when we talk about resource management so i want to maximize profit from the provider point of view maximize utilization
• 05:30 - 06:00 of the resources and ah minimal ah requirement of energy right and of course on the other hand ah we have we need to satisfy these quality of services and ah um sla right rather there are several metrics which we will see when we discuss today but these are the aspects we need to look at when when we look at the resource management aspects so we will be following or we will be taking inputs mostly from a ah survey vapor which
• 06:00 - 06:30 where the link is provided ah you are free to download and look at the things and there you will get lot of other corresponding people who are interested in ah further research or further study on this type of resource management are welcome to look at it so if we look at the resource management if i broad objectives or the broad goal so ah to ah satisfy things that my scalability ah of that ah for which that cloud is one of
• 06:30 - 07:00 the properties that scalability should be preserved that i can i can scale up scale down and ideally infinite scalability quality of services should be preserved optimal ah utility reduced over its like in one or two do a resource management protocol or algorithms if the overhead is high then i loose on performance right so the it should be a optimal overhead
• 07:00 - 07:30 ah rather reduced over it and improved throughput so that overall throughput should be improved ah reduce latency so that ah it should not increase the time latency of the overall system specialized ah environments like whether i want to have a specialized environment in order to have say as last day we discussed as a specialized environment for rack management power and so and so forth cost effectiveness that overall it should
• 07:30 - 08:00 be cost effective ah it should be ah financially beneficial to the both provider and consumer ah the provider should not spend more and the consumers should not have to pay more subscription for that and it should be a simplified interface the interface should be again simple it it should not be very cumbersome interface so that ah it is or i can say that ah we should have a ease of use will be there so it it is easy to use that ah type of environment
• 08:00 - 08:30 so these are our broader ah goal or i can say broad an objective of ah cloud service provisioning without ah compromising this whether i can have a better resource management that is the objective of the things ok now there are several challenges like if you look at the hardware or the bare metal or the backbone of the things the one is that
• 08:30 - 09:00 ah cpu management memory management storage ah management workstation network element sensor actuators and so and so forth so these are the different ah components which are there which need to be properly managed and it is it these these are not isolated things right like cpu memory storage these are not isolated components so they have a intel linking while operations on the thing so you cannot have a very high power cpu in low memory and
• 09:00 - 09:30 type of things then the performance will not be there so if the coordination between the this bare metal or the backbone resources are important so when we manage resources we need to take care that those are preserved right i i cannot make a optimal management of storage without ignoring the other component of the like network component and other things a make a faster storage my network accessibility is still slow then purpose are not solved so those at needs to be looked into
• 09:30 - 10:00 so there are other logical ah resources so those are ah what we say physical hard resources there are logical resources like operating system ah energy management network throughput or bandwidth load balancing mechanisms information security which is coming up in a big ah way or which are looked into in a big way when you are leveraging lot of things on the cloud
• 10:00 - 10:30 and specially your sensitive or semi semi sensitive information on the cloud or privacy preserving things delays that how much delay or time delays are their application programming interface or api so the api is whether need to be redone on new type of api has to be there and there are various protocols so these are all what we say soft resources or logical resources which plays a important role and these hard resources and soft resources
• 10:30 - 11:00 are not ah separate to each other they are intermingled rather need to be looked into in a integrated way so this ah are the different challenges or broad objectives what you look at now there are different type of approaches ah researchers followed and ah or ah to have resource ah management aspects or we say the different resource management aspects one is resource provisioning right we will see
• 11:00 - 11:30 what are the resource allocation right so i need to provision then allocate the resources resource requirement mapping so whether i can map the resource requirements right or sometimes whether i can do a a a priority mapping of the resource requirement right like i am forcing that this sort of resource requirement are happening ah is there then i provision it accordingly right like ah resource requirement in different parts of the different
• 11:30 - 12:00 time scale of the day are different right different time period of the day are different different time period of the year are different maybe over the years things are different so and so forth then we have ah resource adaptation right how resource can be adapted resource discovery there should be a mechanism that how how can i discover resource ah faithfully like where do i find those resources and type of things and where ah how do i as a user can look for the resources so again i repeat that use and
• 12:00 - 12:30 may not be always human user it can be another process another ah set of processes all together so for optimal and that may be a part of a larger applications so in order to do that i need to discover resources and type of things so there should be some provisioning some cataloging registry type of things whether resources can be there there are resource brokering right so some sort of a brokerage or agent based things where those are can
• 12:30 - 13:00 be ah where which ah acts as agent to have ah provide ah me a optimal resource rather i should ah i should say that when a when a user request for a some resource then that initially it hits a agent or a broker which tries to look out that which are the which are the resource available which are less loaded ah nodes how vm can be allocated and
• 13:00 - 13:30 so and so forth so this is a important aspect that booker brokerage type of things resource estimation that is estimating that what sort of resource requirement will be there these are sometimes important when we do higher level of things like i am looking at a saas or paas label need to estimate that what sort of back backbone resources are required and ah there is a thing called resource modeling that how i can model the thing to the resources for considering estimation considering my present load and
• 13:30 - 14:00 type of things so these are different aspects and ah you can see that these are not all are independent aspects they have intel linking between them also right so these are different aspects and the emphasis may vary based on type of application or type of requirement you are there right so in some cases the some of the aspects may be higher priority and so and so forth nevertheless these are not isolated
• 14:00 - 14:30 ah components again they have a inter linking between the things so this ah slide what we are trying to look at that ah what are the different type of ah what are the different aspects and what are the different ah what they mean right so the um so what one aspects is resource provisioning a location of service provided
• 14:30 - 15:00 resources to a constant customers right the customer can be a user or a process right so resource allocation stands for distribution of resource economically among the computing groups of people or programs or processes resource adaptation ability or capacity of that system to adjust the resource dynamically to fulfill the requirement of the user so that ah based on the user requirement the overall the system adjusts itself how this
• 15:00 - 15:30 ah um resource the available resource can be optimally used among the users again i should say that without compromising the sla and ah um other quality of services and type of things then we have resource mapping which says that the correspondence ah between the resource required by the users and the resource available with the providers so based on the resource available and the require requirement how
• 15:30 - 16:00 we map resource modeling resource modeling is based on detailed information transmission ah network element resources entities participating in the ah network right so that means attributes of resource managements ah if we look at they are different states different transitions different outputs with a given environment so every ah resource management so if if i look at the resource management as a ah as a entity or a frame work so it
• 16:00 - 16:30 has it goes in two different state it has different transition from one state to another and every state has the output type of things so i can have realized some sort of a state chart ah diagram or type of things and based on that i i need to model that based on this this how this transition will go on so resource estimation so how i can closely guess the actual resource required by application usually with some thought or calculation involved
• 16:30 - 17:00 i can do some a priori i may have some a priori knowledge about the application or i can have some meta information at the application like this application may require so much memory so much displace so much ah so much threads and type of things and based on that i relocate the resource resource discovery and selection so as we are discussing identification of list of authenticated resource that are available for the job submission
• 17:00 - 17:30 and choose the best among them so it is always possible that you have multiple ah resources or multiple providers with resources are available so that discovering that which are the resources and which is the suitable thing and allocating the most optimal and based about thing and ah resource brokering so negotiation of the resources through an agent ensuring necessary ensuring that the necessary resources are available at the right time to complete the
• 17:30 - 18:00 objectives so i i broke up because i have a requirement as a user i have a requirement as a user process right and then i want to broker i want to um negotiate with the agent that which are the things available right how things will be available so that i my ah objective is fulfilled and the objective may be resource wise objective the objective sometimes can be on the pricing objective also i this much cost and this much things i have to choose so there is see there is a need for optimization
• 18:00 - 18:30 of the whole thing right so the i require a brokering service for that and finally resource scheduling right so is this ah a scheduling is a timetable of events and resources right so say our resources are available at certain times and events are planned during those times right so it can be ah so i have resources i have my operation procedure so time i require
• 18:30 - 19:00 some sort of timetable of sibling the resources that exactly a sibling problem per se so i may have lot ah lot of ah this may have lot of components like duration ha some predecessor activities some predecessor relationship resource allocated and so and so forth so there can be different component for ah to determine that ah start end and type of things so ah if we look at some of the ah approaches ah or some of the ah um different type of
• 19:00 - 19:30 aspects like as we discussed previously resource provisioning resource allocation and try to see that what sort of ah what sort of approaches people are following or researchers are following ah into things like first one let us see at the resource provisioning approaches like ah so what we have nash equilibrium approach for game theory
• 19:30 - 20:00 so using some sort of a game theoretic approach to find out that ah optimal uses of the resource right so what it does the runtime management and allocation of the iaas resources considering several criteria like heterogeneous distribution of resources ah rational exchange behavior of the cloud users in complete common information and dynamics successive allocation and so and so forth so that means i based on this different ah
• 20:00 - 20:30 components like heterogeneous duration of resources or ah users pattern of the cloud users and type of things i want to have a game theoretic approach ah to look at this so its a we can look at as a grame where ah one side is that the consumer which are a who are hungry for the resources or looking for the resources other side there is ah provider who are provisioning the resources and i want to find out a ah optimal way of allocating
• 20:30 - 21:00 the resources so that is a nash equilibrium based using game approach so there are there are ah research and or there are methods and approaches people are following there we have network queuing model so which ah a model based on network queue like ah those who have ah gone through network queuing model or network queue models in in data networks etcetera you can understand it is more of a ah again resource resource
• 21:00 - 21:30 provisioning ah mechanism queues where queues represent different tires of application so the model ah significantly or sufficiently captures the behavior of the tires with significantly different performance characteristics and application like ah session based workload concurrency limits and ah caching the intermediate tires and like that so based on so it is it is try to you what we do what we are what they are doing here to try to exploit network
• 21:30 - 22:00 queuing model there are approaches for prototype provisioning employs the k means clustering algorithm to automatically determine the workload mix and queuing model ah to predict the server capacity for a given workload mix so what it is trying to do so it it tries to cluster ah using say came in ah cluster k means cluster to automatically determine these what is this workload means of the different user ah and then to predict that what are
• 22:00 - 22:30 the what how can be provisioned there are other resource provisioning like vm provisioning ah things like user virtual machines that runs top of the xen hypervisor so the system provides say some sort of a scheduler like ah in some what they propose a simply simple earliest deadline first scheduler that implements the weighted fair sharing of the cpu capacity among the vms right so ah what it is doing it is taking the vm
• 22:30 - 23:00 which run over the hypervisor and ah scheduling it based on the type of ah type of the load it is it is getting that cr cpu cycles a particular vm can be changed on the runtime and so and so forth so if i if i if i have ah requirement from a from more resource requirement then i migrate from one vm to the other vms and type of ah things can be done
• 23:00 - 23:30 there are other ah methods and approaches like adaptive resource provisioning which tries to automatically ah detect the bottlenecks and residues and resolve that using dynamic resource management there are things called sea sla oriented resource ah methods handling process and dynamic provisioning to meet ah user slas in a ah automatic manner dynamic an automatic framework ah which adapt the adaptive parameters which adapt the parameters
• 23:30 - 24:00 to meet the specific users or accuracy goals so it goes on ah provisioning the recruit resources based on the based on the quality of services or the ah type of ah or the type of a sla it has to support and also there is optimal cloud provisioning mechanisms what ah which tries to look at the demand and price uncertainty ah considering those try to optimize
• 24:00 - 24:30 so we see there are there are several approaches which can be used for this sort of resource provisioning ah mechanisms similarly if we look at the resource allocation ah there are again ah several approaches few are ah listed here like market oriented resource allocation ah so which which are driven by the ah market requirement market demand on
• 24:30 - 25:00 the things so we try to do model predictive ah model predictive control to find its ah solution of the of that particular resource allocation there are intelligent multi agent model primarily ah looking for um resource view virtualization to automatically allocate services resource available specifically for devices which are mobile right
• 25:00 - 25:30 so it is i can have a intelligent multi agent model to allocate optimal resources energy energy aware resource allocation so this ah allocation is energy aware so that i can do a optimal energy provisioning measurement based analysis on performance so it allocation again based on different metrics or measurement parameters dynamic resource allocation methods
• 25:30 - 26:00 real time resource allocation mechanisms like if there is a real time demand on the things how how resource can be ah allocated so designed for helping small medium size iaas cloud providers to better utilize their hardware resource with minimal operational cost by a well designed underlining hardware infrastructure right so in order to help for ah in order to help this especially small and medium sized iaas cloud provider so ah how it can be ah allocated
• 26:00 - 26:30 in a real time and dynamic scheduling and consolidation mechanisms over and above i can have a dynamic scheduling and consolidation mechanisms of available resources there are ah again several approaches for ah resource mapping ah like symmetric mapping pattern that is ah for designing resource supply systems it divides the resource into three major functions
• 26:30 - 27:00 users and providers match and engage resource supply management agreements so that is ah ah users and the providers match that where the requirement are matching and do that ah type of ah matching before that users place tasks on the subscribe resource containers right so that ah it ah subscribe resource container place the tasks and the
• 27:00 - 27:30 mapping is done or provider place supplied resource container on physical resources and type of things so these are ah driven by ah container based services ah which ah which is a another ah type of ah another ah technology which is coming up in a bigger ways that this container classes and container things so user can subscribe ah resource continent place their tasks or providers can ah place supplied resource container
• 27:30 - 28:00 on the physical resources it can be a mapping of the load aware mapping so explore how simply vm ah image management and reduce image ah preparation over it by multicast file transferring and image ah caching and using so it is ah based on the load it ah does a load aware mapping to reduce deploying over a rate and make efficient use of the resources so based on the available load it does the load availability there can be technique
• 28:00 - 28:30 for iterated local search based request partitioning so whether i can ah partition request partitioning approach follow a based on iterated local search ah to facilitate a cost efficient and online splitting of is your request among eligible cloud service provider so user may be requesting on a user requests i can basically partition into smaller part if there is a
• 28:30 - 29:00 there is a way of partitioning it i can do a intelligent partitioning algorithm and then allocate the things into different ah csps like different cloud service providers and it that is that that that means that ah a large requests can be partitioned into smaller and look at it so there are other approaches like distributed ah in symbols of virtual applications like ah i have virtual applications ah in symbol
• 29:00 - 29:30 name or mapping a virtual network of a substrate network so i ah i have a underlining network and then i map a substrate network like why map a virtual network which is main for the user to this substrate network right so again this is a resource mapping from the network side of view there is a requirement from the user from the network and map it on the things and there are there are several adaptation approaches ah like reinforce learning
• 29:30 - 30:00 guided control policy so that is a learning mechanism to look at the adaptation there are web service based prototypes right so which can be used for ah resource ah adaptation ah there are several others like looking at virtual networks dnas based load balancing and of course i we can have hybrid approaches for ah having this sort of load ah resource adaptation so if i have this several type of techniques
• 30:00 - 30:30 like ah as we have seen here is like resource provisioning allocation we discussed few of them ah resource requirement mapping adaptation and so and so forth how to judge that they are performance finally ah what we are looking for is ah some matrix right so all those approaches ah need to be judged based on some metric like reliability ah ease
• 30:30 - 31:00 of deployment like i have a mechanisms and it takes lot of lot of overhead to deployment quality of services should not be compromised there should not be ah delay or much delay or the delays should be within the limit and control over it in order to manage these resource management things in order to control these resource management ah mechanisms or processes what are my control over it right so whenever whenever we are looking for any
• 31:00 - 31:30 resource management a play resource management tools or techniques we need to look at all those ah different aspects so otherwise the overall resource management may kill the basic purpose of ah this ah um cloud computing ah paradigm right so that ease of uses scalability and ah in finite resources and type of things we we may suffer
• 31:30 - 32:00 so we need to look at this different matrix and if you can you can see that these are the matrix may differ from different type of ah requirements right so different user may have different ah or different user processes the different requirements and where somewhere the reliability may be pretty high somewhere the quality of service somewhere some of the applications may be delay concerned some of the application may be accuracy concerned and we need to take up the actual ah take up the resource management process resource
• 32:00 - 32:30 management tools and technique based on those those parameters thank you