Potential effects on server power metering and modeling

Abstract

Cloud datacenters are compute facilities formed by hundreds or even thousands of servers. With the increasing demand of cloud services, energy efficiency of servers in data center has become a significant issue. The knowledge of the energy consumption corresponding to hardware and software configuration is important for operators to optimize energy efficiency of a data center. We currently work on a predictive model for energy consumption of a server, with inputs as service provided, hardware material equipped (type and quantity of processor, memory and hard drive) and technical environment (energy conversion and cooling). In this article, we characterize some potential factors on the power variation of the servers, such as: original fabrication, position in the rack, voltage variation and temperature of components on motherboard. The results show that certain factors, such as original fabrication, ambient temperature and CPU temperature, have noticeable effects on the power consumption of servers. The experimental results emphasize the importance of adding these external factors into the metric, so as to build an energy predictive model adaptable in real situations.

Taming Energy Consumption Variations In Systems Benchmarking

“…Moreover, changing hardware components, such as the hard drive, the memory or even the power supply, does not affect the energy variation of a node, making it mainly related to the processor. This result was recently assessed by [26], where the rack placement and power supply introduced a maximum of 2.8 % variation in the observed energy consumption.…”

“…Kistowski et al showed in [15] that identical processors can exhibit significant energy consumption variation with no close correlation with the processor temperature and performance. On the other hand, the authors of [26] claimed that the processor thermal effect is one of the most contributing factors to the energy variation, and the correlation between the CPU temperature and the energy consumption variation is very tight. This makes the processor temperature a delicate factor to consider while comparing energy consumption variations across a set of homogeneous processors.…”

Experimental Characterization of Variation in Power Consumption for Processors of Different Generations

“…As the power increases with the load, processor works harder and dissipates more heat, if the heat is not evacuated in time by the cooling system, the temperature surrounding processor becomes higher and leads to the rise of leakage current, which will in reverse increase the power of CPU [25] [26]. In the study of [6], the authors observed that leakage current of processor can contribute to over 30% power variation for server. However, there is doubt whether the samples have the same ability to evacuate the heat.…”

“…On the contrary to dynamic power, static power is temperature sensitive value, but cannot be affected by frequency scaling. The consumption of the whole server can be then simply represented by the equation (5) and (6):…”

“…The question is, if we build the power model for one server and validate the model precision based on the model outputs and the real measurements, can we trust the model and rely on the same precision to other servers in the homogeneous cluster or even in the whole data centers? In the experimentation of [6], a same test suite composed of several CPU intensive workloads is executed one by one to 12 identical servers in a rack, 7.8% power variation is observed among them. Thermal effect is turned out to be one of the causes: the rise of the ambient temperature increases the power consumption of servers in two ways: the consumption of cooling system (integrated fans) and leakage current.…”