After we knew it

Virtual Infrastructure-as-a-Service (IaaS) clouds are emerging for secondary cloud service providers to manage their own IaaS clouds on top of existing IaaS clouds. In virtual IaaS clouds, guest virtual machines (VMs) run inside cloud VMs provided by existing IaaS clouds. Unlike traditional IaaS clouds, they can be migrated between cloud VMs co-located at the same host. However, the performance of such VM migration is low due to slow virtual networks and doubled system loads. To optimize VM migration between co-located cloud VMs, we propose zero-copy migration for virtual IaaS clouds. Zero-copy migration just relocates the memory image of a guest VM without any copy. To enable live migration with negligible downtime, it first makes the memory of a guest VM share with the destination cloud VM and thereafter completes memory relocation. We have implemented a system called VMBeam for enabling zero-copy migration in Xen. According to our experimental results, zero-copy migration could achieve high migration performance and low system loads.

Datacenters have played an increasingly essential role as the underlying infrastructure in cloud computing. As implied by the essence of cloud computing, resources in these datacenters are shared by multiple competing entities, which can be either tenants that rent virtual machines (VMs) in a public cloud such as Amazon EC2, or applications that embrace data parallel frameworks like MapReduce in a private cloud maintained by Google. It has been generally observed that with traditional transport-layer protocols allocating link bandwidth in datacenters, network traffic from competing applications interferes with each other, resulting in a severe lack of predictability and fairness of application performance. Such a critical issue has drawn a substantial amount of recent research attention on bandwidth allocation in datacenter networks, with a number of new mechanisms proposed to efficiently and fairly share a datacenter network among competing entities. In this article, we present an extensive survey of existing bandwidth allocation mechanisms in the literature, covering the scenarios of both public and private clouds. We thoroughly investigate their underlying design principles, evaluate the trade-off involved in their design choices and summarize them in a unified design space, with the hope of conveying some meaningful insights for better designs in the future.

Cloud computing enables users to perform distributed computing tasks on many virtual machines, without owning a physical cluster. Recently, various distributed computing tasks such as scientific applications are being moved from supercomputers and private clusters to public clouds. Message passing interface (MPI) is a key and common component in distributed computing tasks. The virtualized computing environment of the public cloud hides the network topology information from the users, and existing topology-aware optimizations for MPI are no longer feasible in the cloud environment. We propose a network performance aware MPI library named CMPI. CMPI embraces a new model for capturing the network performance among different virtual machines in the cloud. Based on the network performance model, we develop novel network performance aware algorithms for communication operations. This poster gives an overview of CMPI design, and presents some preliminary results on collective operations such as broadcast.We demonstrate the effectiveness of our network performance aware optimizations on Amazon EC2.

Virtual machine (VM) management is one of the key problems in cloud data centers. Many efforts have been put on finding an (approximate) optimal placement of VMs at physical machines (PMs) to achieve high resource utility. In this paper, we consider to reduce the startup delay of VMs, which is a key issue for improving user experience and resource utility. The basic idea of our work is to cache VM images at PMs so as to reduce the time of VM image retrieval. By handling VM placement and image placement jointly and de-centrally, we design a joint placement algorithm and an associated layered transport VM selection algorithm, so as to start VMs quickly in the meanwhile save some calculation resources. The key point of our algorithms lies in how to accelerate the startup of VM with limited information in a distributed fat-tree network. The proposed algorithms are evaluated by extensive simulations using a distributed system simulator SimGrid. The results show that our algorithms just lost a little performance in most cases, compared with the centralized algorithm and a classical cache placement strategy.

Infrastructure-as-a-service (IaaS) Clouds concurrently accommodate diverse sets of user requests, requiring an efficient strategy for storing and retrieving virtual machine images (VMIs) at a large scale. The VMI storage management require dealing with multiple VMIs, typically in the magnitude of gigabytes, which entails VMI sprawl issues hindering the elastic resource management and provisioning. Nevertheless, existing techniques to facilitate VMI management overlook VMI semantics (i.e at the level of base image and software packages) with either restricted possibility to identify and extract reusable functionalities or with higher VMI publish and retrieval overheads. In this paper, we design, implement and evaluate Expelliarmus, a novel VMI management system that helps to minimize storage, publish and retrieval overheads. To achieve this goal, Expelliarmus incorporates three complementary features. First, it makes use of VMIs modelled as semantic graphs to expedite the similarity computation between multiple VMIs. Second, Expelliarmus provides a semantic aware VMI decomposition and base image selection to extract and store non-redundant base image and software packages. Third, Expelliarmus can also assemble VMIs based on the required software packages upon user request. We evaluate Expelliarmus through a representative set of synthetic Cloud VMIs on the real test-bed. Experimental results show that our semantic-centric approach is able to optimize repository size by 2.2 - 16 times compared to state-of-the-art systems (e.g. IBM's Mirage and Hemera) with significant VMI publish and slight retrieval performance improvement.

Cost issue always is one of most important issue for enterprise in building their IT platform. With the advance of various cloud environments, enterprise’s IT executive may have different consideration in constructing their IT platform. For example, constructing an own cloud environment or renting a computing platform from cloud providers. This paper the authors, based on the Net Present Value in Finance field, propose a cost model for evaluating the tradeoff between building a self-own computing platform and renting one from public provider. The cost evaluation formulas are derived based on some cost factor, such as human salary, platform purchasing cost, energy consuming, maintenance fee, and cooling cost, for evaluating required cost on building and operating of both public and private clouds. Based on the cost model, the authors applied a well-known public cloud instance, e.g. Chunghwa Telecom’s HiCloud platform in Taiwan, to compute renting cost, and make a comparison with the derived cost of private cloud for a variety of enterprises’ settings. According to the experiments, different size of enterprises could have different options in building private cloud or renting from public cloud to make optimum cost saving.

By employing the virtual machines (VMs) consolidation technique at a virtualized data center, optimal mapping of VMs to physical machines (PMs) can be performed. The type of optimization approach and the policy of detecting the appropriate time to implement the consolidation process are influential in the performance of the consolidation technique. In a majority of researches, the consolidation approach merely focuses on the management of underloaded or overloaded PMs, while a number of VMs could also be in an underload or overload state. Managing an abnormal state of VM results in the postponement of PM getting into an abnormal state as well and affects the implementation time of the consolidation process. For the aim of optimal VM consolidation in this research, a self-adaptive architecture is presented to detect and manage underloaded and overloaded VMs /PMs in reaction to workload changes in the data center. The goal of consolidation process is employing the minimum number of active VMs and PMs, while guaranteeing the quality of service (QoS). Assessment criteria of QoS are two parameters including average number of requests in the PM buffer and average waiting time in the VM. To evaluate these two parameters, a probabilistic model of the data center is proposed by applying the queuing theory. The assessment results of the probabilistic model form a basis for decision-making in the modules of the proposed architecture. Numerical results obtained from the assessment of the probabilistic model via discrete-event simulator under various parameter settings confirm the efficiency of the proposed architecture in achieving the aims of the consolidation process.

Infrastructure-as-a-Service (IaaS) clouds have become more popular enabling users to run applications under virtual machines. This paper investigates the energy-aware virtual machine (VM) allocation problems in IaaS clouds along characteristics: multiple resources, and fixed interval times and non-preemption of virtual machines. Many previous works proposed to use a minimum number of physical machines, however, this is not necessarily a good solution to minimize total energy consumption in the VM placement with multiple resources, fixed interval times and non-preemption. We observed that minimizing total energy consumption of physical machines is equivalent to minimize the sum of total completion time of all physical machines. Based on the observation, we propose EMinRET algorithm. The EMinRET algorithm swaps an allocating VM with a suitable overlapped VM, which is of the same VM type and is allocated on the same physical machine, to minimize total completion time of all physical machines. The EMinRET uses resource utilization during executing time period of a physical machine as the evaluation metric, and will then choose a host that minimizes the metric to allocate a new VM. In addition, this work studies some heuristics for sorting the list of virtual machines (e.g., sorting by the earliest starting time, or the longest duration time first, etc.) to allocate VM. Using the realistic log-trace in the Feitelson's Parallel Workloads Archive, our simulation results show that the EMinRET algorithm could reduce from 25% to 45% energy consumption compared with power-aware best-fit decreasing (PABFD) [1]) and vector binpacking norm-based greedy algorithms (VBP-Norm-L1/L2 [2]). Moreover, the EMinRET heuristic has also less total energy consumption than our previous heuristics (e.g. MinDFT and EPOBF) in the simulations (using same virtual machines sorting method).

In Infrastructure-as-a-Service (IaaS) clouds, stepping-stone attacks via hosted virtual machines (VMs) are critical for the credibility. This type of attack uses compromised VMs as stepping stones for attacking the outside hosts. For self-protection, IaaS clouds should perform active responses against stepping-stone attacks. However, it is difficult to stop only outgoing attacks at edge firewalls, which can only use packet headers. In this paper, we propose a new self-protection mechanism against stepping-stone attacks, which is called xFilter. xFilter is a packet filter running in the virtual machine monitor (VMM) underlying VMs and achieves pinpoint active responses by using VM introspection. VM introspection enables xFilter to directly obtain information on packet senders inside VMs. On attack detection, xFilter automatically generates filtering rules based on packet senders. To make packet filtering with VM introspection efficient, we introduced several optimization techniques. Our experiments showed that the performance degradation due to xFilter was usually less than 16%.

Infrastructure-as-a-service (IaaS) clouds enable customers to allocate computing resources in a flexible manner to satisfy their needs, and pay only for the allocated resources. One of the challenges for IaaS customers is the correct provisioning of their resources. Many users end up under provisioning, hurting application performance, or over provisioning, paying for resources that are not really necessary. Memory is an essential resource for any computing system, and is frequently a performance-limiting factor in cloud environments. Our work uses monitoring to enable a cloud customer to determine if the memory allocated to his virtual machines is correctly provisioned, under provisioned, or over provisioned. Experimental results with the Xen platform demonstrate the effectiveness of the proposed approach.

Infrastructure-as-a-Service (IaaS) clouds provide infrastructure as a pool of virtual resources, and the public IaaS clouds, e.g. Amazon Web Service (AWS) and private IaaS cloud toolkits, e.g. OpenStack, CloudStack, etc. provide their own application programming interfaces (APIs) for managing the cloud resources they offer. The heterogeneity of the APIs, however, makes it difficult to access and use the multiple cloud services concurrently and collectively. In this paper, we explore previous efforts to solve this problem and present our own implementation of an integrated cloud API, which can make it possible to access and use multiple clouds collectively in a uniform way. The implemented API provides a RESTful access and hides underlying cloud infrastructures from users or applications. We show the implementation details of the integrated API and performance evaluation of it comparing the proprietary APIs based on our cloud testbed. From the evaluation results, we could conclude that the overhead imposed by our interface is negligibly small and can be successfully used for multi-cloud access.

The emergence of academic and commercial Infrastructure-as-a-Service (IaaS) clouds is opening access to new resources for the HEP community. In this paper we will describe a system we have developed for creating a single dynamic batch environment spanning multiple IaaS clouds of different types (e.g. Nimbus, OpenNebula, Amazon EC2). A HEP user interacting with the system submits a job description file with a pointer to their VM image. VM images can either be created by users directly or provided to the users. We have created a new software component called Cloud Scheduler that detects waiting jobs and boots the user VM required on any one of the available cloud resources. As the user VMs appear, they are attached to the job queues of a central Condor job scheduler, the job scheduler then submits the jobs to the VMs. The number of VMs available to the user is expanded and contracted dynamically depending on the number of user jobs. We present the motivation and design of the system with particular emphasis on Cloud Scheduler. We show that the system provides the ability to exploit academic and commercial cloud sites in a transparent fashion.

Virtual machine (VM) images (VMIs) often share common parts of significant size as they are stored individually. Using existing de-duplication techniques for such images are non-trivial, impose serious technical challenges, and requires direct access to clouds' proprietary image storages, which is not always feasible. We propose an alternative approach to split images into shared parts, called fragments, which are stored only once. Our solution requires a reasonably small set of base images available in the cloud, and additionally only the increments will be stored without the contents of base images, providing significant storage space savings. Composite images consisting of a base image and one or more fragments are assembled on-demand at VM deployment. Our technique can be used in conjunction with practically any popular cloud solution, and the storage of fragments is independent of the proprietary image storage of the cloud provider.

Cloud computing is characterized by its essentially pay-per-use pricing with elasticity. Typically, the granularity of usage for such pricing is at virtual machine (VM) level in IaaS clouds, e.g., a multiple of machine hours. The elasticity and cost effectiveness in these clouds are primarily achieved through the exploitation of resource virtualization and sharing. However, a majority of applications running on VMs in clouds struggle to fully utilize resources allocated to them. Since co-location granularity is strictly restricted to VM level and resources allocated to VMs are space-shared, the unused resources are apt to be wasted while users are still charged for such wastage. In this paper, we address the problem of fine-grained and adaptive resource sharing for real pay-per-use pricing. To this end, we devise a resource management mechanism as a cost efficiency solution for both users and providers of clouds. The mechanism consists of a container-based resource allocator and a real-usage based pricing scheme. We demonstrate the efficacy of this mechanism via experiments, in Amazon EC2, using two typical workloads in clouds, web services and database services, and a compute-intensive high energy physics application. Our results show that the mechanism can achieve near-optimal cost efficiency.

After years of development, cloud computing has developed into public cloud, private cloud, community cloud, hybrid cloud and other cloud forms. However, the government, financial and other industries have more concerns about security, operation and maintenance, cost and other issues in the process of cloud. In this regard, the industry, academia and other parties have launched the frontier exploration, give different solutions. To meet users’ new needs, we proposed a new model that combines the advantages of public and private clouds. It is called proprietary cloud. Proprietary cloud provides users with more efficient management by incorporating years of experience of public cloud providers. On the one hand, it meets the rigid needs of users’ independent control and exclusive resources, on the other hand, it reduces the maintenance cost and downtime probability of users, and achieves the desired effect of users. It mainly includes self-built model and exclusive model deployment modes. The main difference is whether to accept public cloud management. This paper introduces and analyzes proprietary cloud in detail from the aspects of deployment architecture, mode differences and common deployment patterns. It concludes that proprietary cloud is an indispensable cloud computing deployment pattern in the future.KeywordsCloud computingDigital transformationProprietary cloud