دانلود مقاله ترجمه شده کاربرد برنامه نویسی پویا برای رفع مشکل جداسازی عملکرد در PaaS – مجله IEEE

• فهرست مطالب:

 چکیده
۱ مقدمه
۲ کارهای مرتبط
۳ راه حل مبتنی بر برنامه نویسی پویا و پیش بینی
۴ آزمایش
۵ نتیجه گیری

• بخشی از ترجمه:

در این مقاله روشی برای حل مسئله تفکیک عملکرد در PaaS طراحی و پیاده می کنیم. در این راستا از روش پیش بینی برازش منحنی برای محاسبه منابع مصرفی VM ها از قبل و تعیین VM های مورد نیاز جهت مهاجرت و انتقال و از روش برنامه نویسی پویا برای تصمیم گیری در مورد بهترین استراتژی برای مهاجرت VM، استفاده می کنیم. سپس برای اثبات درست بودن روش حل مسئله تفکیک عملکرد در PaaS، یک آزمایش طراحی کرده و آن را به پایان می رسانیم.
در کار آتی، مدل ریاضی بهتری برای برازش منحنی جهت دستیابی به نتیجه پیش بینی درست تر طراحی خواهیم نمود. در این رابطه روش بهتری برای محاسبه منابع جهت پوشش صحنه های بیشتر در PaaS طراحی خواهیم نمود.

• بخشی از مقاله انگلیسی:

Cloud computing is now not only a concept in computer science field but also a great business model in many other fields. But nowadays lots of people have got a misconception that resources in a cloud computing environment are infinite or unlimited. In [1], authors remind us that we should avoid this misconception and provide customers the resources as they really need. A service-level agreement (SLA) is a part of a service contract where the level of service is formally defined. [2] In SLA customer and services provider will make multiple agreements include how many resources services providers may provide to customer. A SLA is a contract. So it is very important to services providers that they should not breach the SLA at any time or they will bear the liability for breach of contract. As we all know, an application may not use resources in upper bound all the time which is set in SLA. So one of the basic ideas is to allocate more resources to virtual machines (VM) in a real machine host and total of these resources in all VMs are more than the real machine host has. To ensure a services provider won’t breach SLA, we should make sure that if one of VMs’ workload increases fast in a host, other VMs’ performance may not be influenced. This problem is called Performance Isolation problem. To solve this problem, we could move such kind of VMs to other hosts which have more idle resources. As moving a VM to another host may cost lots of time, we should know which VM’s workload may increase fast in advance. This forecast needs not to be most accurate because we just want to know the increasing risk or trend of VM’s resources demands. The strategy to move VMs is another problem. If we move a VM with high workload to a host which has fewer resources, the performance isolation problem still exists. If we move the VM to a host which has lots of resources with none other VMs, it’s very wasteful. We need find a suitable way to move these VMs. II. RELATED WORK In 2005, Chen Guang, et al. made a research in Curve Fitting [3] and give an algorithm. If we got history data of every kind of resource of a VM, we could use this algorithm to get the curve of this VM and then we could forecast the trend of the resource incensement. In 2010, Chen Xu had a survey of performance management in his master thesis [4], the authors combine the SLA provided for tenants of the largest number of active users to detect the offending tenants, once the monitor found the active users are too many for a long time, then determine the tenant for unauthorized users. Due to the presence of unauthorized users will lead to the performance of other tenants damaged; it will be re-consultation with the tenants and the development of a new SLA. In 2011, Javier Espadas, et al. had a survey of tenant-based resource allocation model [5]. The authors believe that the VM resources are allocated to tenants depending on the number of active users of the tenant. Therefore, the authors propose a set of active users based multi-tenant virtual machine resource allocation algorithm. In 2008, Zhi Hu Wang, et al. had a survey of performance evaluation [6]. The authors used x-axis and y-axis. The x-axis includes isolation, security, customization, and scalability. The y-axis includes performance, manageability, and development efficiency. The authors think that a good multi-tenant application should be as much as possible to improve the quality attributes. Allocation of resources should meet the demand for resources for these quality attributes. III. FORECAST AND DYNAMIC PROGRAMMING BASED SOLUTION In Forecast and Dynamic Programming Based Solution (FDPBS), we designed the architecture as the simple class diagram shows in figure 1. As we need to know which VM’s resources demand may increase fast, we should monitor every VM in all hosts to get their history monitor data at first. In this paper we take CPU and memory for example. We don’t care how to monitor these VMs. What we need is the data. In this paper, we used XML files to save these VMs’ monitor data. In this XML file, we record a VM’s id, IP address, host, CPU resource upper bound; memory resource upper bound and history monitor data. We don’t have to know these data in real time because we just want to know the trend of VMs’ resource demands. So it is needless for us to design an Observer-Listener architecture, we just let the VMMonitor class to monitor all the VMs at short intervals. As we all know, operations to VMs may cost lots of time. If we monitor VMs too many times in a short period, last VMs’ operations may not finish. By our experience, 10-15 times an hour is proper. It depends on your platform.