Analysis, Modelling and Characterisation of Zombie Servers in Large-Scale Cloud Datacentres

作者：Panneerselvam John^*; Liu Lu; Hardy James; Antonopoulos Nick

来源：IEEE Access, 2017, 5: 15040-15054.

DOI：10.1109/ACCESS.2017.2725898

• Energy-aware systems
• data exploration
• power management
• ubiquitous computing

Cloud datacentres are acknowledged as being massive energy consumers, which may have significant environment impacts. Service providers have an ethical responsibility to reduce the environmental impact of server resources and a simultaneous and complementary commercial desire to reduce energy costs. Zombie servers in the datacenters are one of the primary sources of undesirable energy expenditures by incurring idle resources during task execution. This paper investigates the cause, impact and energy-related implications of zombie servers. Important outcomes of this paper are the characterization of the diversity among the workload behaviors in resource consumption and the quantification of the presence of idle CPU and memory resources during task execution causing server zombieness. The undesirable power consumption of zombie servers is determined based on the profiles of currently available servers and their corresponding environmental implications are illustrated in this paper. Empirical analysis shows that cloud workloads are highly heterogeneous in resource consumption pattern and CPU resources may display 75.6% of idleness relative to their allocated level, while memory is 25.5% idle. The report concludes that significant reductions in power consumption and CO2 emission can be achieved by provisioning a realistic level of resources to servers, which are scaled to suit the anticipated workloads.

• 出版日期2017