دانلود رایگان ترجمه مقاله چالش ها در مجازی سازی در زمان واقعی و محاسبات ابری قابل پیش بینی (نشریه الزویر 2014)

چالش ها در مجازی سازی در زمان واقعی و محاسبات ابری قابل پیش بینی

Challenges in real-time virtualization and predictable cloud computing

Abstract

Cloud computing and virtualization technology have revolutionized general-purpose computing applications in the past decade. The cloud paradigm offers advantages through reduction of operation costs, server consolidation, flexible system configuration and elastic resource provisioning. However, despite the success of cloud computing for general-purpose computing, existing cloud computing and virtualization technology face tremendous challenges in supporting emerging soft real-time applications such as online video streaming, cloud-based gaming, and telecommunication management. These applications demand real-time performance in open, shared and virtualized computing environments. This paper identifies the technical challenges in supporting real-time applications in the cloud, surveys recent advancement in real-time virtualization and cloud computing technology, and offers research directions to enable cloud-based real-time applications in the future.

1 Introduction

The widespread availability for the masses of high-speed Internet connections at affordable rates, by means of DSL and more recently optical technologies, paired with an unprecedented connectivity through cellular and wireless technologies, is enabling an inescapable shift towards distributed computing models. Applications relying merely on physical resources and data available in the local personal computer (PC) are slowly but consistently becoming part of the history, as the PC declines leaving the way to a new era of distributed computing. This is subsumed into the recently expanding paradigm of Cloud Computing [87], in which resources are rented in an on-demand and pay-per-use fashion from cloud providers. Just as a huge hardware machine, cloud computing data centres deliver an infrastructure, platform, and software applications as services that are available to consumers. Such services are referred to as IaaS (infrastructure as a service), PaaS (platform as a service), and SaaS (software as a service), respectively [88]. Cloud applications are capable of running and spreading out their computations and data on as many nodes as needed, and they can access huge amounts of data directly available within the premises of cloud data centers. Shortly, cloud computing is enabling the next generation of computing services, heavily geared towards massively distributed and on-line computing, as well as enabling a new model of on-demand high performance computing (HPC) accessible to anyone from anywhere, whenever needed. As new application domains enter the cloud world progressively, real-time systems are also expected to move in this direction due to the tremendous possibilities and augmented utility that this paradigm could bring about. Examples are both on the hard and soft sides of real-time systems, such as military distributed control systems applied to remote surveillance, early response and warning systems, unmanned vehicles with augmented intelligence from the sensor cloud, or cloud gaming, among others. Cloud computing, and particularly the use of public clouds, brings advantages on the technical, environmental and business sides, allowing multiple under-utilized systems to be consolidated within fewer physical servers hosting them. A cloud provider can manage physical resources in a very efficient way by scaling on the several hundreds and thousands of customers (a.k.a., tenants) with dynamically changing workload requirements, by re-optimizing the infrastructure in a completely automated (or semiautomated) fashion whenever needed, providing high levels of availability and reliability. One of the most important technologies that enabled this paradigm shift in computing is virtualization, and particularly machine virtualization. Machine virtualization (also referred to as processor virtualization) allows a single physical machine to emulate the behavior of multiple machines, with the possibility to host multiple and heterogeneous operating systems (called guest operating systems or guest OSs) on the same hardware. A virtual machine monitor (VMM), or hypervisor, is the software infrastructure running on (and having full control of) the physical host and which is capable of running such emulation. Virtualization allows for server consolidation in data centers, where multiple operating systems that would leave their underlying hosts under-utilized can be moved to the same physical resources. This enables the achievement of a reduction of the number of required physical hosts, and their improved exploitation at higher saturation levels, thus saving costs and energy [86]. The multi-tenant nature of cloud computing has a great influence on the increasingly rich and challenging user requirements on cloud infrastructures. Users demand not only access to on-line storage, but also to real-time and interactive applications and services. This is also witnessed by visionary products already on the market, such as lightweight computers which are almost incapable of doing anything locally, unless they are connected to the ‘‘Cloud’’. In a high-performance cloud computing (HPCC) environment, applications have much stronger temporal requirements; as such, the characteristic of performance, including resource guarantees and timely provisioning of results, becomes critical. It is actually an open research area to match such requirements with virtualized environments due to I/O overhead and jitter of the required duration of executed instructions. Moreover, activities or jobs are mostly allocated to a specific core, but they often have synchronization dependencies with respect to other activities. Merging cloud computing with real-time is a complex problem that requires to also focus on (among others) the efficient access to the physical platform. Although real-time hypervisors typically may allow applications to access to the physical machine, in virtualized environments for cloud computing, it is clear that the hardware is typically not directly accessible by the user-level application software layers. With the current available technology, it could be possible to improve service response times from a cloud platform using high performance techniques. The main problems lie on the multi-tenancy of the cloud computing platforms that execute on heavy loaded servers the requests of several independent users. Currently, there are a number of commercial real-time hypervisors (some providing hierarchical scheduling) for safety critical systems of different origins such as WindRiver, Acontis Technology, SysGO, OpenSynergy, LynuxWorks, or Real Time Systems GmbH. However, it is not likely to see them among the mainstream cloud technology in the immediate future due to performance levels and compatibility problems at the low level execution layer. Mainstream and real-time ones were created with different objectives. For example, real-time hypervisors were not invented for maximizing throughput of user requests and providing statistical guarantees on service contracts, but to preserve temporal isolation and determinism.