Article Preview
TopIntroduction
In most cloud services, virtual machine instances normally share physical processors and I/O interfaces with other instances. It is expected that virtualization can impact the computation and communication performance of cloud services. Currently, most of the research work related to cloud computing has dealt with implementation issues, while performance-related issues have received much less attention (Maguluri et al., 2012). Walker (2008) compares the application level performance of Amazon EC2 against another high performance computing cluster, NCSA, and shows that Amazon EC2 has much worse performance than traditional scientific clusters. To overcome the major delays during data exchange for large data sets in the cloud, Broberg et al. (2009) and Fedak et al. (2009) propose MetaCDN and BitDew, respectively. MetaCDN reduces the complexity of dealing with multiple storage providers on the cloud and provides a unified name space, and BitDew addresses the issue of large-scale data management in the cloud and the grid environments. Based on empirical measurements, Wang and Eugene Ng (2010) make the first study focusing on the end-to-end (e2e) networking performance of Amazon EC2 instances and on understanding the impact of virtualization on the data center network performance. They analyze the abnormal large delay variations and unstable TCP/UDP throughput caused by end host virtualization. Their findings are helpful to explain the observations in (Walker, 2008). Mei et al.(2013) first show that current implementation of virtual machine monitor does not provide sufficient performance isolation to guarantee the effectiveness of resource sharing across multiple virtual machine instances running on a single physical host machine, especially when applications running on neighboring VMs are competing for computing and communication resources. Then we present the detailed analysis on different factors that can impact the throughput performance and resource sharing effectiveness.
To study the response time in terms of various metrics, such as the overhead of acquiring and realizing the virtual computing resources, Yigitbasi et al. (2009) design and implement C-Meter, a portable, extensible, and easy-to-use framework for generating and submitting test workloads to computing clouds. Using C-Meter, researchers can be easy to evaluate different scheduling algorithms. In addition, theoretical analyses on cloud services quality mostly rely on performance evaluation of M/G/m queuing systems (Miyazawa, 1986; Yao, 1985).