Enabling Cloud Computing for Seamless Research Computation

使用云计算交付模式向研究人员提供计算资源可以降低研究人员的成本和障碍，同时提高基础设施提供商的利用率和接受度。然而，这种访问基础设施的模式仍然需要技术专业知识、定制软件开发或重新设计，而且通常还需要在不熟悉的环境中远程工作。这项提议寻求将云视为研究人员首选计算设备的自然延伸，而不需要专业知识。这是云资源演变的下一步。**所有学科和背景的研究人员长期以来一直使用计算资源来支持和增强他们的研究。云计算作为一种新兴的资源交付机制，其广泛的保护伞已经改变了研究人员获取和利用计算资源的方式。尤其令人感兴趣的是对计算基础设施(基础设施即服务，IaaS)的按需、服务驱动的访问。代替购买物理机器，研究人员被授予临时访问虚拟或物理资源的权限，只使用他们需要的资源，并且(如果适用)只为他们使用的资源付费。对于那些有零星但密集的计算需求的人，这种方法降低了成本。除了各种公共云提供商(Amazon、IBM)外，在提供高性能计算方面有很长历史的机构和联盟也在以类似的模式提供资源。**尽管IaaS在研究方面具有优势，但利用这些资源仍然需要技术专业知识(通过培训、招聘或外包)，这会带来时间和成本上的障碍。例如，研究软件可能会被重新设计为使用检查点以批处理模式工作，或者可能会将MatLab任务移植到MatLab分布式计算中，或者重新编写分析任务以使用Map-Reduce范例。**该提议寻求将云作为个人设备和工作空间的无缝扩展。例如，研究人员可能会像往常一样使用他们喜欢的软件包继续进行，直到更长的计算开始。此任务将被推送到远程资源执行，然后在完成一段时间后拉回本地计算机。为了实现这一目标，该项目将a)系统地了解当前的研究工作流程，使用观察、仪器和访谈。目标是了解研究人员目前如何使用计算资源，以及该过程如何满足他们的期望。B)利用虚拟机迁移、进程迁移和机器克隆，在将任务从一台物理机移动到另一台物理机方面推进最先进的技术。C)创建一个以可用技术增强的工作流程，使研究人员能够将工作推送到云端。D)从该实施中提取并详细分析针对一般用户的类似方法的要求和设计。**这些目标将推动云式计算资源访问的最先进水平，解决可用性问题。所描述的方法超越了以服务形式提供的特定于任务的软件，允许研究人员使用他们首选的软件，而不会受到远程桌面解决方案的延迟影响。减少使用计算来扩展和加强研究的障碍也将对其他学科产生更广泛的影响。

The provision of computing resources to researchers using a cloud computing delivery model can lower costs and reduce barriers for researchers, while increasing utilization and uptake for infrastructure providers. However, this model of access to infrastructure still requires technical expertise, custom software development or re-engineering, and often the need to work remotely in an unfamiliar environment. This proposal seeks to instead treat the cloud as a natural extension of a researcher's preferred computing device, with no specialized knowledge required. This is the next step in the evolution of cloud resources.**Researchers of all disciplines and backgrounds have long used computing resources to support and augment their research. The broad umbrella of "cloud computing" as an emerging mechanism for resource delivery has shifted how researchers acquire and utilize computing resources. Of particular interest is on-demand, service-driven access to computing infrastructure (infrastructure-as-a-service, IaaS). In lieu of acquiring physical machines, a researcher is granted temporary access to virtual or physical resources, using only what they need, and (if applicable) paying only for what they use. For those with sporadic but intensive computation needs, this approach reduces costs. In addition to various public cloud providers (Amazon, IBM), institutions and coalitions with a long history of providing high-performance computing are making their resources available in a similar model.**Despite the advantages of IaaS for research, making use of these resources still requires technical expertise (through training, hiring, or outsourcing) which imposes barriers of time and cost. For example, research software might be re-engineered to work in batch mode with checkpoints, or Matlab jobs might be ported to Matlab Distributed Computing, or an analysis task re-written to use the Map-Reduce paradigm.**This proposal seeks to enable the cloud as a seamless extension of a personal device and workspace. For example, a researcher might proceed as normal using their preferred software packages, until a longer computation begins. This task would be pushed to remote resources for execution, then pulled back to the local machine some time after completing. To achieve this goal, this project will a) Understand current research workflows systematically, using observation, instrumentation, and interviews. The objective is to understand how researchers currently use computation resources and how that process meets their expectations. b) Advance the state of the art in moving tasks from one physical machine to another, drawing on virtual machine migration, process migration, and machine cloning. c) Create a workflow augmented with usable technology to enable researchers to push work to the cloud. d) Extract from this implementation and detailed analysis the requirements and design for a similar approach to general users.**These objectives will advance the state-of-the-art in cloud-style access to computing resources, addressing usability concerns. The described approach moves beyond task-specific software offered as a service, allowing researchers to use their preferred software, without the latency impact of remote desktop solutions. Reducing barriers to using computation to extend and enhance research will also have a broader impact on other disciplines.