Publication No 40744


Kühn, P.J.; Mashaly, M.


DVFS-Power Management and Performance Engineering of Cloud Data Center Server Clusters


Dynamic Voltage and Frequency Scaling (DVFS) is amethod to save energy consumption of electronic devices and toprotect them against overheating by automatic sensing andadaptation of their energy consumption. This can be accomplishedeither on the program instruction level for electronic devices or onthe task or job level for server clusters. This paper models DVFS onthe job level and through which Service Levels Objectives can beguaranteed with respect to prescribed mean or quantiles of servicedelays according to given Service Level Agreements (SLA) betweenuser and service provider. The two parameters V (voltage) and f(frequency) cannot be changed independently of each other;typically only several combinations of V and f values areimplemented in hardware for several power states. In this paper anovel analysis of operating DVFS is suggested for Server Clustersof Cloud Data Centers (CDC) under prescribed bounds of servicelevel objectives which are defined by SLAs. The method is based onthe theory of queuing models of the type GI/G/n for a server clusterto establish a relationship between SLA parameters and the powerconsumption and is performed for the example of the Intel PentiumM Processor with Enhanced SpeedStep Power Management. Asresult of this method precise boundsare provided for the loadranges of service request rates ? for each power mode whichguarantee minimum power consumption dependent on given SLAvalues and job arrival and service statistics. As the instantaneousload in a CDC can be highly volatile the current load level isusually monitored by periodic sensing which may result in a ratherhigh frequency of DVFS range changes and correspondingoverhead. For that reason an automated smoothing method issuggested which reduces the frequency of DVFS range changessignificantly. This method is based on a Finite State Machine(FSM) with hysteresis levels.



Reference entry

Kühn, P.J.; Mashaly, M.
DVFS-Power Management and Performance Engineering of Cloud Data Center Server Clusters
Proceedings of the 15th Annual Conference on Wireless On-demand Network Systems and Services (WONS), Wengen, January 2019, pp. 91-98

BibTex file

Download  [BIBTEX]

Full Text

Download  [PDF]

Authors marked with an asterisk (*) were IKR staff members at the time the publication has been written.