Our work differs from the above systems in that we’ve de- veloped some elaborate heat transfer models for data centers. Our model is a trade-off between the complex CFD model and the other on-line scheduling algorithms. Our model, therefore, is less complex than the CFD models and can be used for on- line scheduling in data centers. It can provide more accurate description of datacenterthermal maps than , . In detail, we study a temperature–based workload model and a thermal-based datacenter model. This paper then defines the thermalawareworkloadscheduling problem for data centers and presents a thermal-awarescheduling algorithm for datacenter workloads. We use simulations to evaluate thermal- awareworkloadscheduling algorithms and discuss the trade- off between throughput, cooling cost other performance met- rics. Our unique contribution is shown as follows. We propose a general framework for thermalaware resource management for data centers. Our framework is not bound to any specific model, such as the RC-thermal model, the CFD model, or a task-termparature profile. A new heuristic for thermalawareworkloadscheduling is developed and evaluated, in terms of performance loss, cooling cost and reliability.
• Increase power density and improve operation effi- ciency. Compute servers with lower temperatures can be accommodated in smaller spaces, thus increasing power density and operation efficiency of a datacenter. In this paper, we develop a thermal-awareworkloadscheduling concept framework and algorithm in a datacenter. The goal of our implementation is to reduce tem- peratures of compute nodes in a datacenter without sig- nificantly increasing job execution times. The key idea of the implementation is to distribute workloads to ‘‘cool’’ computing nodes, thus making a thermal balancing. We first develop a workload model and compute resource model for data centers in Sect. 3. Then a task scheduling concept framework and a thermal-awarescheduling algo- rithm (TASA) are described in Sects. 5 and 7. In TASA, workloads are distributed to ‘‘cool’’ computing nodes, which is predicted by the artificial neural network (ANN) technique. In Sect. 6, we present the implementation of ANN-based temperature prediction. Section 8 discusses the simulation and performance evaluation of TASA and Sect. 9 concludes our work.
Dynamic voltage and frequency scaling (DVFS) has been proven to be a feasible solution to reduce processor power con- sumption , . By lowering processor clock frequency and supply voltage during some time slots, for example, idle or communication phases, large reductions in power consumption can be achieved with only modest performance losses. A DVFS-enabled cluster  is a compute cluster where compute nodes can run at multiple power/performance operating points. The DVFS techniques have been applied in the high perfor- mance computing fields, for example, in large data centers, to reduce power consumption and achieve high reliability and availability , , . Popular DVFS-based software solutions for high end computing include:
which is typically a two-step scheduling process. The matching resolves the contention for paths and output ports using a cen- tralized scheduler. In a bufferless architecture, packets arrive to their destinations in an ordered way which obviate the needs for re-sequencing mechanisms. Still, the need for a central scheduler rises the system complexity and makes the architec- ture less appealing for large-scale switches. Buffered structures provide higher performance than bufferless switches. They need simple control, but backpressure mechanisms must be implemented to prevent buffers overflowing. Packets of the same flow are likely to experience variable delays depending on their sejourn in the middle stage SEs. This results in an out- of-sequence packets delivery. Some new interesting proposals suggested building scalable high-performance switches/routers using the Networks-on-Chip paradigm   . A NoC- based switch brings several advantages over classic crossbars, such as a flexible design, a pipelined scheduling and a sub- quadratic growth of the fabric’s cost.
is known as Cloud computing.  Cloud Computing promises reliable services delivered through next- generation data centers that are built on virtualized compute and storage technologies. Consumers will be able to access applications and data from a “Cloud” anywhere in the world on demand. The consumers are assured that the Cloud infrastructure is very robust and will always be available at any time. Computing services need to be highly reliable, scalable, and autonomic to support ubiquitous access, dynamic discovery and composability. In particular, consumers indicate the required service level through Quality of Service (QoS) parameters, which are noted in SLAs established with providers. Of all these paradigms, the recently emerged Cloud computing paradigm appears to be the most promising one to leverage and build on
combination of Throttled (TVLB) and ESCE (AVLB) algorithm. TVLB algorithm makes use of states of VMs. A virtual machine state may be either AVAILABLE or BUSY. AVAILABLE state indicates that the virtual machine is idle/free and ready for cloudlet allotment, where BUSY state indicates that the current virtual machine is busy in execution of previous cloudlets and is not available to handle any new cloudlet request. This current load state of a VM helps in taking decision whether to allocate cloudlets to virtual machines or not. Active VM Load Balancing algorithms continuously monitor the job queue for new cloudlets and allot them to the bunch of idle/free VMs. It also maintains the list of cloudlets allocated to each virtual machine. This allocated cloudlet list helps in determining whether a VM is overloaded or under loaded at particular moment of time. On the basis of this information, VM load Balancer moves some load from overloaded VMs to the VM having minimum number of cloudlets, so as to maintain a high degree of balance among virtual machines. Knowledge of VM States and allocated cloudlets are the two main features of Throttled and Active VM Load Balancer Scheduling algorithms respectively. These features combined together, make the proposed scheduling algorithm more efficient & effective and help in fair distribution of the load. Hybrid VM Load Balancing Algorithm Input- Userbases/Cloudlets UB1, UB2,... UBn. - Available VMs VM1, VM2, VM3,…,VMn within datacenter. Step1: Hybrid VM Load Balancer maintains a list of VMs, their states (AVAILABLE/BUSY) and allocated cloudlets list. Initially state of every VM is AVAILABLE and allocated cloudlet list is empty. Step 2: DataCenter Controller gets cloudlet requests from cloud clients. Step 3: DataCenter Controller asks the Hybrid VM load Balancer for available VM. Step 4: Hybrid VM load Balancer do a) Find the next available VM using VM State List. b) Check if present allocation count is less than maximum VM list length and length of virtual machine list is greater than zero, then allocate the VM. c) Determine the current cloudlet load on every VM. d) Return vmId of those VM which have minimum load. Step 5: Hybrid VM Load Balancer allocates the cloudlet over available VM. Step 6: If a VM get overloaded then hybrid VM Load balancer moves some workload on VM that have minimum workload Step 7: The DCC get the reply of sent cloudlet & then allots a waiting request from job pool to hybrid VM Load Balancer. Step 8: continue with step 4. Output- Userbases/cloulets are allocated on the available VMs and completed with minimum response time and processing time at DC.
In  a context aware scheduler for hadoop (CASH) is suggested which utilizes the cluster heterogeneity and workload mix. By making the scheduler aware of cluster heterogeneity can improve the throughput of the system. Cash algorithm classifies the jobs and nodes as CPU or I/O bound and maps the tasks with different demands to the nodes which can fulfill these demands. The algorithm is implemented and tested on Mimuk simulator. When compared with FIFO scheduler, performance is improved by 20-36%. For small jobs the simulator showed an average improvement which needs to be improved further. Nguyen et al. in  proposed a hybrid scheduler algorithm based on dynamic priority in order to reduce the response time for variable length jobs. The dynamic priorities can accommodate multiple task length, job size and job waiting time by applying an algorithm named greedy fractional knapsack for job task processor assignment. A reordering of task processor assignment is implemented for job execution to account of data availability and preserving data locality. It improves the average response time by 2.1x faster than hadoop fair scheduler. In future, effects of service level will be evaluated on total workload completion time.
High temperature has become a major problem for system-on-chip testing. In order to reduce the test time while keeping the temperature of the chip under test within a safe range, a thermal-aware test scheduling technique is required. This paper presents an approach to minimize the test time and, at the same time, prevent the temperature of cores under test going over the given upper limit. We employ test set partitioning to divide test sets into shorter test sequences, and add cooling spans between test sequences, so that overheating can be avoided. Moreover, test sequences from different test sets are interleaved, in order to utilize the cooling spans and the test bus bandwidth for test data transportation, hence the total test time is reduced. The test scheduling problem is formulated as a combinatorial optimization problem, and we use constraint logic programming (CLP) to solve it in order to obtain the optimal solution. As the CLP approach needs relatively long time for execution, we have also developed a heuristic to generate the near-optimal test schedule with much shorter computation time. Experimental results have shown the efficiency of both the CLP and heuristic approach.
For cooling energy, results in Fig. 7(c) show that thermal-aware algorithms TASA, Ella-W and Ella-B consume 64.2% less cooling energy than the other non-thermal-aware algorithms Random, MaxUtil, MTTE and OBFIT. Cooling energy is affected by removed volume of heat produced by CRACs and corresponding CoP. Higher CRAC temperature setting leads to smaller CoP, indicating better cooling efficiency. Thermal-aware algorithms allocate workloads allowing for maximized CRAC temperature setting. In order to analyze this result, we further demonstrate the workload distribution (with standard error of mean, SEM) of each rack in Fig. 9(a). The indicator rack utilization is defined as the average utilization of all servers within a rack. We present the rack utilization of Rack #1 - #6, from which we can learn the workload distribution of all the 24 racks since our modelled datacenter is symmetric. For Random algorithm, rack utilization follows a uniform distribution as any server has the same probability being selected. MaxUtil selects servers in ascending order (i.e. from the rack to the last), making workloads consolidate more within the first rack and follow a roughly descended distribution. MTTE and OBFIT do not show any obvious features since their allocation is affected by the server reliability that is unrelated to rack location. However, temperature-aware algorithms TASA, Ella-W and Ella-B allocate more workloads to Racks #2 - #5, and less workloads to the bilateral racks (i.e. Rack#1 and #6). This is because the middle racks are closer to the CRAC (datacenter layout is illustrated in Fig. 5(a)), leading to higher cooling efficiency compared with bilateral racks. This results in thermal- aware algorithms performing better within the context of enhanced cooling efficiency.
Vinothina et al.  survey on resource allocation strategies focuses on Service level agreement (SLA) to identify utilization of resources. It is an in-depth discussion on optimal allocation to strengthen cloud services. Tinghuai et al.  conducted a review of existing resource allocation and scheduling techniques to attain SLA by providing cost- effective services. The scheduling strategies are considered locality-aware, reliability-aware and energy-aware for resource management. Manvi et al.  review resource provisioning, allocation and scheduling techniques for efficient management of infrastructure resources. The QoS requirements and SLA violation concerns were also discussed to provide cost-effective services. Singh et al.  presented a survey on QoS aware resource management techniques in cloud computing. The importance of optimization-aware self- management techniques also discussed. It shows the impact of QoS parameters on SLA, it also gives suggestions further development of standardizing resource management technique. Mustafa et al.  present a review and taxonomy of resource management techniques. It discusses challenges in energy-aware, workload-aware and network resources for managing SLA and profit in hybrid and mobile cloud. Alkhank et al.  detailed taxonomy on workflow scheduling for efficient resource management. It discusses challenges in resource allocation and scheduling to provide cost-aware services while managing data-intensive applications. Zhan et al.  presented, taxonomy on resource scheduling schemes to provide application-aware
resource flow to another process flow of dynamic resource in the cloud environment. In addition, it extends the resource model to consider the data transfer cost between data centers so that nodes can be deployed on different regions. Extending the algorithm to include heuristics that ensure a task is assigned to a node with sufficient memory to execute it will be included in the algorithm. Also, it assigns different options for the selection of the initial resource pool. In addition, data transfer cost between data centers are also calculated so as to minimize the cost of execution in multi-cloud service provider environment. The main contribution of paper, the following problem solve in the existing system, they contribution are
Today the case of big data archetype in distributed computing is to store the huge datasets. Significant angle for cloud client is cloud information security and protection. In this paper we have given an execution of approved evaluating and proficient fine grained updates. We have additionally actualized auditability mindful information booking which attempts to use the cloud assets to serve the customers with most extreme throughput. Our proposed framework has three noteworthy segments which give a protected, approved and effective evaluating of information and alteration of information in cloud condition. Security is given by confining produced TPA from inspecting user‟s information without user‟s concern. Effectiveness is accomplished by supporting little updates. This paper executed a need based planning calculation to give legitimate use of cloud assets among the errands coming to CSS. Our proposed work‟s result demonstrates the outcome for content record. We can transfer content document and give fine-grained updates to those record as it is an all around acknowledged arrangement. In future the work will center for various sorts of document. For security reason more layer of verification to TPA will be given.
Andrew J. Younge et al.  developed a new framework that provided efficient green enhancements within a scalable Cloud computing architecture. In this paper the author defined that using power-awarescheduling techniques, resource management, live migration, and a minimal virtual machine design, overall system efficiency will be greatly improved in a data centers. The results show that if the scheduler distributes the VMs with the aim to fully utilize all processing cores within each node, the power consumption is decreased dramatically.
WLAN networks have emerged as the most deployed last stage component of internet connectivity to mobile users. Growing subscribers of WLANs for Wi-Fi accessing devices is driving the traffic in WLANs. Accordingly, WLAN services are prone to severe declines in performance amid channel interference and contention. To address rising traffic issues and performance degradation, a novel channel scheduling strategy is proposed. Unlike the most of the contemporary models, the proposed model schedules the channels based on multi-objective QoS factors, moreover it balances the load by transmitting buffered data packets as transmission-window. The contention state of channel availability also addressed in this proposal. The research work depicts that the recommended algorithm signifies the improved throughput, defused drop ratio, and also distributes the user traffic based on optimizing channel scheduling.
Containerized clusters of machines at scale that provision Cloud services are encountering substantive difficulties with stragglers – whereby a small subset of task execution negatively degrades system performance. Stragglers are an unsolved challenge due to a wide variety of root-causes and stochastic behavior. While there have been efforts to mitigate their effects, few works have attempted to empirically ascertain how system operational scenarios precisely influence straggler occurrence and severity. This challenge is further compounded with the difficulties of conducting experiments within real-world containerized clusters. System maintenance and experiment design are often error-prone and time-consuming processes, and a large portion of tools created for workload submission and straggler injection are bespoke to specific clusters, limiting experiment replicability. In this paper we propose PRISM, a framework that automates containerized cluster setup, experiment design, and experiment execution. Our framework is capable of deployment, configuration, execution, performance trace transformation and aggregation of containerized application frameworks, enabling scripted execution of diverse workloads and cluster configurations. The framework reduces time required for cluster setup and experiment execution (from hours to minutes). We use PRISM to conduct automated experimentation of system operational conditions and identify straggler manifestation is affected by resource contention, input data size and scheduler architecture limitations.
In Chapter 4 we combine MPC with meeting scheduling and propose a MIP-based solution. This model achieves significant energy reduction when compared to ap- proaches similar to those presented in Goyal et al. ; Kwak et al. ; Majum- dar et al. . The drawback of this approach is that it does not scale well, which is why we developed a hybrid solution that combines MIP with LNS [Lim et al., 2015b]. Also limited by the scalability of MIP, existing work on energy-aware schedul- ing only consider a small number of meetings or rooms. For example, Chai et al.  consider only 30 meetings in 9 rooms. Kwak et al.  solves a total of 300 meetings per day in 35 rooms, but in an online manner where the MIP model needs to be solved at every session is relatively smaller. The other work resorts to heuristics-based approaches that generate a feasible solution in a reasonably short period of time but do not guarantee optimality. For instance, Pan et al.  ex- amine up to 800 meetings in 150 rooms with their greedy scheduling algorithm. Likewise, Majumdar et al.  only consider up to 12 meetings in 4 rooms due to the computational overhead incurred by feeding their schedules into building energy simulation software Energy+ [Crawley et al., 2000] for the calculation of HVAC con- sumption. Our work overcomes the limitations of these approaches by combining MIP with LNS, and by solving each sub-problem to optimality or near-optimality using a MIP-based integrated HVAC control and occupancy scheduling model.
K-Tier datacenter scheduler starts with initial abstract job obtained from input job by using job duration abstraction. Initial abstract datacenter is obtained from input data structure by collapsing all computation nodes into a single node. Scheduler keeps job abstraction constant but refines datacenter abstraction as required. Memory allocator maintains a partition of the memory in order to find best suitable free memory block. Each refinement step splits some block into two new blocks. Partition is represented as a binary tree . When an allocated memory is freed then compaction easily done by collapsing the tree. Best-fit allocation is used to schedule tasks from one job to nodes close to each other.Representation of datacenter changes with each allocation.