Energy-Aware Learning Agent (EALA) for Disaggregated Cloud Scheduling

Backfilling refers to the practice of allowing small jobs to be completed ahead of schedule as long as they do not cause the first job in the line to wait. Users are expected to offer estimates of how long jobs will take to complete in order to make these decisions possible, and these projections are often based on historical data. However, predictions are very hard and may not be accurate, particularly in cloud computing scenarios where jobs or applications run on Virtual Machines (VMs). In addition, scheduling and consolidation techniques can improve the energy efficiency and performance of applications. Consolidation involves VM migrations that can have a negative impact on workload performance and users’ costs. Backfilling can be used as an alternative technique for consolidation (short-term) and/or can be used along with consolidation (long-term). Backfilling methods are well-utilised in single computing systems, but are relatively unexplored in cloud resource allocation. A backfilling-based resource allocation and consolidation technique is proposed. Using real workloads from the Google cluster traces, we investigate the impact of backfilling on infrastructure energy efficiency and performance. For 12583 heterogeneous servers and approximately three million jobs that belong to three different applications, we observed that approximately 19% energy savings and 6% workload performance improvements are achievable using the backfilling approach. Furthermore, our evaluation suggests that using VM runtime as a criterion for the backfilling approach is approximately 3.56%–7.78% more energy and 1.91%–3.38% more performance efficient than using priority as a backfilling criterion.

In Infrastructure-as-a-Service (IaaS) clouds, users manage the systems in virtual machines (VMs) called user VMs through remote management systems (RMSes). To allow users to manage their VMs during failures inside the VMs, IaaS usually provides out-of-band remote management. This management is performed indirectly via an RMS server in a privileged VM called the management VM. However, it is discontinued when a user VM is migrated. This is because an RMS server in the management VM at the source host is terminated on VM migration. Even worse, pending data is lost between an RMS client and a user VM. In this paper, we propose D-MORE for continuing out-of-band remote management across VM migration. D-MORE provides a privileged and migratable VM called DomR and performs out-of-band remote management of a user VM via DomR. During VM migration, it synchronously co-migrates DomR and its target VM and transparently maintains the connections between an RMS client, DomR, and its target VM. We have implemented D-MORE in Xen and confirmed that a remote user could manage his VM via DomR after the VM has been migrated. Our experiments showed that input data was not lost during VM migration and the overhead of D-MORE was acceptable.

Cloud providers generally run one or more Virtual Machine (VM) instances on the same physical machine. Though it increases data center utilization, it exposes VM to a co-location attack. VM placement and migration are the two strategies adopted for mitigating co-locations. Current methods for VM placement or VM migration consider only security as decision criteria and do not consider other factors like Quality-of-Service degradation, data center utilization, etc. This work proposes a placement and migration strategy for mitigation of co-location attacks with joint consideration of multi objectives like QoS, data center utilization, energy consumption, and security risks. A security-driven multi-criteria optimization -based VM placement policy is proposed. A joint consideration of multiobjective performance optimization along with co-location security risk minimization is done to design a novel VM placement policy based on user categorization. The policy can reduce the likelihood of co-location target VM with attacker VM without much degradation to the performance of VM and data center utilization. The solution mitigates co-location risks without much compromise to the performance of VM and data center resource utilization. The co-residence risk is mitigated by the categorization of users into three levels i.e. unlabeled, risky, and safe, and physical machines into two groups as safe and unsafe. The PMs available in data center is grouped into three different VM placement policies, they are undecided pool, safe pool and unsafe pool.

Servers have different energy efficiency and energy proportionality (EP) due to their hardware configuration (i.e., CPU generation and memory installation) and workload. However, current virtual machine (VM) scheduling in virtualized environments will saturate servers without considering their energy efficiency and EP differences. This article will discuss EASE, the energy efficiency and proportionality aware VM scheduling approach. EASE first executes customized computing intensive, memory intensive, and hybrid benchmarks to calculate a server's energy efficiency and EP. Then it schedules VMs to servers to keep them working at their peak energy efficiency point (or optimal working range). This step improves the overall energy efficiency of the cluster and the data center. For performance guarantee, EASE migrates VMs from servers under highly contending conditions. The experimental results on real clusters show that power consumption can be saved 37.07% ~ 49.98% in the homogeneous cluster. The average completion time of the computing intensive VMs increases only 0.31 % ~ 8.49%. In the heterogeneous nodes, the power consumption of the computing intensive VMs can be reduced by 44.22 %. The job completion time can be saved by 53.80%.

We propose two VM migration methods that improve the total sum of program correct by dynamically moving VMs to appropriate edge node. We assume a multi-hierarchy information processing system that the cloud data center and the edge node VM (Virtual Machine) perform processing at the same time, the data center completes all processing, and the VM has the required response time and returns a result that the processing has progressed (called accuracy). The first method is “VM sweeping method” that allocates more CPU time to VMs that perform high accuracy processing. The second methods is “VM sweeping method” that allocates more CPU time to VMs that perform high accuracy processing that improves per unit CPU time (AIR), and “VM number averaging method” that attempts to move VMs so that the assignment status of the CPU time in all edge nodes is equal. The accuracy improvement effect of the proposed VM migration method was evaluated by a simulator. Evaluations changed the AIR from the Early-blooming type where the accuracy greatly increases in the first half to the Late-blooming type where the accuracy greatly increases in the second half, and comparing it with the case where the VM is not moved. It was clarified that a big effect can be expected in the VM sweeping method, and the VM number averaging method in the Earlyblooming type program.

As a cloud computing model have led clusters to the large-scale data centers, reducing of the energy consumption which imposes a crucial part of the whole operating expense for data centers has received a lot of attention of a wide public. At cluster-level viewpoint, the most popular method for energy efficient cloud is Dynamic Right Sizing (DRS), which turns off idle servers those do not have any of running virtual resources. To maximize the energy efficiency through DRS, one of primary adaptive resource management strategies is a Virtual Machine (VM) consolidation which integrates VM instances into as few servers as possible. In this paper, we propose Virtual machine Consolidation based Size Decision (VC-SD) approach migrates VM instances from under-utilized servers which are supposed to be turned off to sustaining ones according to their monitored resource utilizations in real time. In addition, we design a Self Adjusting Workload Prediction (SAWP) method to improve a forecasting accuracy of resource utilization even under irregular demand patterns. Through experimental results based on real cloud servers, we show various metrics such as resource utilization, energy consumption and switching overhead caused by application processing, VM migration and DRS execution to verify a necessity of our proposed methodologies.

Energy consumption is significant in cloud data centres even when some virtual machines (VMs) are idle. Taking into account both the energy efficiency and the response performance, we propose a VM allocation strategy with speed switch and VM consolidation in the cloud data centre. When the traffic is lighter, the processing speed of VMs in baseline module will be switched between a low speed and a high speed. When the traffic is much heavier, the VMs in reserved module will be consolidated with the VMs in baseline module, and operates at a high speed. We establish a two-dimensional Markov chain to mathematically estimate the energy-saving degree and the response performance. Numerical experiments with analysis and simulation show that our proposed VM allocation strategy can effectively reduce the energy consumption as well as guarantee the response performance.

Modern multi-core, multi-socket hardware powering the "cloud" is designed with memory and cache resources that have local associativity to a group of cores. Operating systems have characterized these associativities as part of Non-Uniform Memory Architecture (NUMA) optimizations within their virtual memory manager (VMM) and scheduler which improves application performance. Mature NUMA optimizations are prevalent for an OS running on bare-hardware. However, their benefits are reduced within virtual machines(VM) in the cloud. Even though cloud applications are executed by standard multi-core multi-socket hardware, the abstraction brought in by the virtualization layer makes it challenging to optimize resource management within a VM. This paper describes new paradigms in hypervisor technology-- scheduling VMs on cores and managing their memory resources which streamlines their execution on underlying multicore hardware. Being Linux kernel developers, we base our work on Linux operating system and Kernel Virtual Machine (KVM) --a subsystem which turns the Linux kernel into a scalable hypervisor. We evaluate few approaches for optimal resource allocation with KVM. We summarize our findings with a comparative study of how different scheduling algorithms can be employed with KVM for various systems to support efficient resource placement while running cloud workloads.

Impact of Factors Affecting Pre-copy Virtual Machine Migration Technique for Cloud Computing

Live Virtual Machine (VM) migration has significantly improved the flexibility of modern Data Centers (DC). However, seamless live migration of a VM between geo-distributed DCs faces several challenges due to difficulties in preserving the network configuration after the migration paired with a large network convergence time. Although SDN-based approaches can speed up network convergence time, these techniques have two limitations. First, they typically react to the new topology by installing new flow rules once the migration is finished. Second, because the WAN is typically not under SDN control, they result in sub-optimal routing thus severely degrading the network performance once the VM is attached at the new location. In this paper, we identify networking challenges for VM migration across geo-distributed DCs. Based on those observations, we design a novel long-haul VM migration scheme that overcomes those limitations. First, instead of reactively restoring connectivity after the migration, our SDN-based approach proactively restores flows across the WAN towards the new location with the help of EVPN and VXLAN overlay technologies. Second, the SDN controller accelerates the network convergence by announcing the migration to other controllers using MP-BGP control plane messages. Finally, the SDN controller resolves the sub-optimal routing problem that arises as a result of migration implementing a distributed anycast gateway. We implement our approach as extensions to the OpenDaylight controller. Our evaluation shows that our approach outperforms existing approaches in reducing the downtime by 400 ms and increasing the application performance up to 12 times.

More than bin packing: Dynamic resource allocation strategies in cloud data centers

This paper addresses the problem of simultaneously migrating a group of co-located and virtual machines (VMs), i.e, VMs executing on the same physical machine. We refer to such a mass simultaneous migration of active VMs as live gang migration. Cluster administrators may often need to perform gang migration for load balancing, system maintenance, or power savings. Application performance requirements may dictate that the total migration time, network traffic overhead, and service downtime, be kept minimal when migrating multiple VMs. State-of-the-art migration techniques optimize the migration of a single VM. In this paper, we optimize the simultaneous migration of multiple co-located VMs. We present the design, implementation, and evaluation of a de-duplication based approach to perform concurrent migration of co-located VMs. Our approach transmits memory content that is identical across VMs only once during migration to significantly reduce both the total migration time and network traffic. Using the QEMU/KVM platform, we detail a proof-of-concept prototype implementation of two types of de-duplication strategies (at page level and sub-page level) and a differential compression approach to exploit content similarity across VMs. Evaluations over Gigabit Ethernet with various types of VM workloads demonstrate that our prototype for gang migration can achieve significant reductions in both network traffic and total migration time.

As cloud computing models have evolved from clusters to large-scale data centers, reducing the energy consumption, which is a large part of the overall operating expense of data centers, has received much attention lately. From a cluster-level viewpoint, the most popular method for an energy efficient cloud is Dynamic Right Sizing (DRS), which turns off idle servers that do not have any virtual resources running. To maximize the energy efficiency with DRS, one of the primary adaptive resource management strategies is a Virtual Machine (VM) migration which consolidates VM instances into as few servers as possible. In this paper, we propose a Two Phase based Adaptive Resource Management (TP-ARM) scheme that migrates VM instances from under-utilized servers that are supposed to be turned off to sustainable ones based on their monitored resource utilizations in real time. In addition, we designed a Self-Adjusting Workload Prediction (SAWP) method to improve the forecasting accuracy of resource utilization even under irregular demand patterns. From the experimental results using real cloud servers, we show that our proposed schemes provide the superior performance of energy consumption, resource utilization and job completion time over existing resource allocation schemes.

The expeditious development in information technology visualizes that facilities for computing are one of the prime utilities of life after the other main essential utilities i.e., water, gas, and electricity. With the help of virtualization, cloud computing helps in transparently sharing of data and services between the cloud users and provides access to more than thousands of computers as a single instance. The virtual machine allocation is a challenging task in virtualization which is regulated as a crucial part of the virtual machine (VM) migration. It is done to discover the best assignment of VMs on physical machines (PMs) because it has explicit effects on resource utilization, energy efficiency, the performance of running applications, etc. So, there is a need to design an effective method for the VM placement problem. The research paper presents a virtual machine allocation technique implementing Enhanced-Modified Best Fit Decreasing (E-MBFD) Algorithm. The outcome of the proposed technique is the cross-validation of allocated virtual machines on physical machines using Artificial Neural Network (ANN). Also, the proposed technique is implemented with the merits of finding the false allocations which happen due to inefficient utilization of resources and it helps in re-allocation of these VMs. The empirical result shows that the E-MBFD technique outperforms in terms of minimized power consumption and a lesser number of service-level agreement (SLA) violations in contrast with the conventional technique.

As more and more data centers embrace end host virtualization and virtual machine (VM) mobility becomes commonplace, we explore its implications on data center networks. Live VM migrations are considered expensive operations because of the additional network traffic they generate, which can impact the network performance of other applications in the network, and because of the downtime that applications running on a migrating VM may experience. Most virtualization vendors currently recommend a separate network for VM mobility. However, setting up an alternate network just for VM migrations can be extremely costly and thus presents a barrier to seamless VM mobility. Therefore, it is apparent that VM migrations should be orchestrated in a network-aware manner with appropriate QoS controls such that they do not degrade network performance of other flows in the network while still being allocated the bandwidth they require for successful completion within the specified time lines. In this context, we present VMPatrol - a QoS framework for VM migrations. VMPatrol uses a cost of migration model to allocate a minimal bandwidth for a migration flow such that it completes within the specified time limit while causing minimal interference to other flows in the network. Our implementation and experimental evaluation of VMPatrol on real and virtual software testbeds demonstrates that automated bandwidth reservation can reduce the impact of migrations on other flows in the network to a negligible level.