Taxonomy of resource allocation algorithms for inner IaaS cloud data centres Dang Minh Quan
Taxonomy of resource allocation algorithms
tải về 0.56 Mb.
|
- Điều hướng trang này:
- Execution phase of VMs
- Business model of providers
- Goal of the algorithm
Taxonomy of resource allocation algorithmsFrom our point of view, we classify IaaS’s resource allocation algorithms according to three main criteria: execution phase of VMs, business model of providers and goal of the algorithm. We build the hierarchy of those algorithms as presented in Fig. 5. In this section, we discuss in detail each criterion. Execution phase of VMsThis classification is based on activities of the cloud system during VM’s lifecycle. At the beginning, when a VM comes to the system, the system will do the initial placement. During the VM execution, the system will perform VM migration if necessary. The differences of task and workload parameters make the working mechanism of resource allocation algorithms different from each other at each execution phase. Initial placement - When users submit VMs to cloud, initial placement algorithms are executed. Their task is determining if the VM can be admitted and where to execute it. The algorithms need the resource information including static and dynamic information. The static information includes the computing node information such as number of computing nodes, number of CPUs, number of cores, memory capacity, storage capacity, etc. The dynamic information include the current usage of resource such as CPU load, memory load, network load, etc. At this phase, the user’s preference is strictly respected. The workload in this case is a group of VMs with their required resource information. It usually includes the number of vCPUs, amount of memory, amount of storage and bandwidth. In some cases, this information also includes the bid price of the user. 
Figure 5. Taxonomy of resource allocation algorithms VM migration - VMs migration algorithms are triggered by internal data centre policy during the operation process of data centre. Unlike the initial placement algorithms, migration algorithms do not care about discarding some VMs. They try to find new place for running VMs to satisfy data centre’s goal. To do this task, the resource information is the same as in the previous case. But, the workload in this case is all VMs running in the data centre with their current resource information usage. It usually includes the CPU usage rate, amount of memory usage, amount of storage usage and bandwidth usage. Business model of providersThis classification is based on the way IaaS cloud providers distribute cloud resource to users. The IaaS cloud providers can distribute resource under spot market model, resource reservation model, game theory model or resource on demand model. The difference in business model makes the working mechanism of the underlying resource allocation algorithms different from each other. Spot market - In the spot market model, user has to bid the maximum price he/she is willing to pay per resource hour. If the user’s maximum price bid exceeds the current Spot Price, the user can run VMs on the gained resource within that period. After that, user has to bid again to gain the right of using resource for the next period. It is noted that the resource allocation mechanism for spot market includes two separate steps. The first step is determining which VM with its correlated bid price could be admitted to the system. The second step is determining which admitted VM running on which physical machine (PM). Resource reservation - In the resource reservation model, user choose the VMs instance types and the period to use those VMs. Reserved Instances offer a capacity reservation, so that user can launch several instances he/she has reserved when he/she needs them. For long-term resource reservation, such as in Amazone EC2, Reserved Instances give user the option to make a low, one-time payment for each instance he/she wants to reserve. In turn, he/she receives a significant discount on the hourly charge for that instance. In the elastic cloud environment, the provider must make sure they have enough resources to meet demand. Otherwise, the provider will need to pay compensation to those customers whose reserved VM instances have not been executed when they wanted. Game theory - In the game theory model, the complex interactions between multiple clients using the cloud simultaneously are considered. At each scheduling round, the system considers all the existing workloads and new workloads. The existing workloads are prioritized to be scheduled first. After that, users having new requests interactively select the right resources from the available pool. This selection process takes place through several rounds until it reaches equilibrium. This equilibrium ensures that clients are charged (near) optimal prices for their resource usage and resources of the cloud are used near their optimal capacity. Resource on demand – This is the current most popular resource distribution model for IaaS cloud. In the resource on demand model, when users need resource to run their VMs, they can get them from the cloud. They can use them as long as they wish. The price of using resource for each VM instance is usually fixed. Goal of the algorithmThis classification is based on the expectancy of the cloud providers with their resource usage. Some main focuses are energy-efficient, profit maximization, load balance or number of active PMs minimization. Besides that, some other goals include fault tolerance, ensuring SLA, etc. It is clear that different goals lead to different resource allocation algorithms. Profit maximization - Profit maximization is an important goal for commercial cloud providers. There are two ways to gain profit maximization. If the cloud providers have many prices for a single resource, in the spot market for example, selling the same amount of resource with highest price will reach profit maximization. If the cloud providers have fixed price for a single resource, maximizing the workload running by the same amount of resource also ensures profit maximization. Social welfare maximization – Social welfare maximization is a goal of economic that tried to apply to cloud environment. The mechanisms manage to distribute resources in a way that reaches equilibrium. This equilibrium ensures that clients are charged (near) optimal prices for their resource usage and resources of the cloud are used near their optimal capacity. Energy efficient - Saving money in the energy budget of a data centre, without sacrificing SLAs is an excellent incentive for data centre owners. It would at the same time be a great success for environmental sustainability. Usually the power consumption of a cloud data centre is calculated through the power consumption of the IT equipment and the PUE (Power usage effectiveness) value of the data centre.  (1)
With a data centre, its PUE is relatively stable. Thus, efforts to cut energy consumption of a data centre focuses on reducing the energy consumption of servers. The power consumption model of a server is further divided to power consumption of CPU, disk, memory, etc. Detail server power consumption model can be seen in [13]. Allocating the same amount of workload using minimal amount of energy will assure energy-efficient. Number of active PMs minimization – This goal derived from the goal of the online bin-packing problem. Objects of different volumes must be packed into a finite number of bins of capacity V in a way that minimizes the number of bins used. In the IaaS cloud environment, the resource allocation module has to allocate a set of VMs into a finite number of PMs in a way that minimizes the number of PMs used. Load balance - Load balance is one of the main challenges in cloud computing. It requires distributing the dynamic workload across multiple nodes to make sure that no single node is overwhelmed. It helps in optimal use of resources and hence in enhancing the performance of the system. For the IaaS cloud, the load of each host is calculated with:  (2)
Usually, the VM entitlement is the occupied CPU resource or memory resource of the VM running on the host. The host capacity is the available resource of the host, which can be provided to VMs. Miscellaneous – Besides main goals as stated above, the literature also recorded some research works about IaaS cloud resource allocation with other goals such as ensuring SLA, fault tolerance, multi objectives, etc. Каталог: Uploads -> files files -> VIỆn chăn nuôi trịnh hồng sơn khả NĂng sản xuất và giá trị giống của dòng lợN ĐỰc vcn03 luậN Án tiến sĩ NÔng nghiệp hà NỘI 2014 files -> Btl bộ ĐỘi biên phòng phòng tài chíNH files -> Bch đOÀn tỉnh đIỆn biên số: 60 -hd/TĐtn-tg đOÀn tncs hồ chí minh files -> BỘ NÔng nghiệP files -> PHỤ LỤC 13 MẪU ĐƠN ĐỀ nghị HỌC, SÁt hạch đỂ CẤp giấy phép lái xe (Ban hành kèm theo Thông tư số 46/2012/tt-bgtvt ngày 07 tháng 11 năm 2012 của Bộ trưởng Bộ gtvt) CỘng hòa xã HỘi chủ nghĩa việt nam độc lập Tự do Hạnh phúc files -> TRƯỜng cao đẲng kinh tế KỸ thuật phú LÂm cộng hòa xã HỘi chủ nghĩa việt nam files -> CHƯƠng trình hoạT ĐỘng lễ HỘi trưỜng yên năM 2016 Từ ngày 14 17/04/2016 files -> Nghị định số 79/2006/NĐ-cp, ngày 09/8/2006 của Chính phủ Quy định chi tiết thi hành một số điều của Luật Dược files -> MỤc lục mở ĐẦU 1 phần I. ĐIỀu kiện tự nhiêN, kinh tế, XÃ HỘI 5 files -> LỜi cam đoan tôi xin cam đoan đây là công trình nghiên cứu khoa học của riêng tôi. Các số liệu, kết quả nghiên cứu nêu trong luận án này là trung thực, khách quan và chưa được ai bảo vệ ở bất kỳ học vị nào tải về 0.56 Mb. Chia sẻ với bạn bè của bạn:
|