Agile, efficient virtualization power management with low-latency server power states

Isci, Canturk; McIntosh, Suzanne; Kephart, Jeffrey O.; Das, Rajarshi; Hanson, James E.; Piper, Scott; Wolford, Robert; Brey, Tom; Kantner, Robert M.; Ng, Allen L.; Norris, James R.; Traoré, Abdoulaye; Frissora, Michael

doi:10.1145/2485922.2485931

Cited by 37 publications

(17 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the case of ERWS, the lower panel of Figure 1 clearly shows that the optimal value of k may lie elsewhere. In line with the findings in [10] suspend gives the best ERP and ERWS amongst the sleep options. This is because setup delay from the suspend sleep state is significantly lower while the power consumption is still kept reasonably low.…”

Section: Numerical Resultssupporting

confidence: 83%

“…Moreover, mean setup delays of 100 s, 50 s and 10 s will be used for the off, hibernate and suspend states, respectively. All values are taken from experimental measurements performed in [7,10] while the service times obey an exponential distribution, where the mean is exaggerated to E[S] = 1 s for illustration. Weighting factors of w1 = 1 and w2 = 0.75 are used for all the ERWS plots.…”

Section: Numerical Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Optimal sleep-state control of energy-aware M/G/1 queues

Gebrehiwot¹,

Aalto²,

Lassila³

2015

EAI Endorsed Trans IoT

View full text Add to dashboard Cite

We study the problem of optimally controlling the use of sleep states in an energy-aware M/G/1 queue. In our model, we consider a family of policies where the server upon becoming idle can wait for a random period before entering, potentially randomly, any of a finite number of possible sleep states to save energy. The server becomes busy again after a possibly random number of jobs have arrived. However, jobs are served only after a random setup time. This kind of an energy-aware queuing system has been analyzed in recent papers under specific assumptions regarding the cost metrics and the distributions of the random variables. In this paper, we consider an essentially more general model. Notably we show that the optimal control of the idle time and sleep states is deterministic and does not benefit from randomization: either the system only uses the idle state and no sleep states, or the idle state is not used at all and the server immediately goes to some fixed sleep state and waits until a fixed number of jobs have arrived before starting the setup. We prove this result for two popular cost metrics, namely weighted sum of energy and response time (ERWS) and their product ERP.

show abstract

Section: Numerical Resultssupporting

confidence: 83%

Section: Numerical Resultsmentioning

confidence: 99%

Optimal sleep-state control of energy-aware M/G/1 queues

Gebrehiwot¹,

Aalto²,

Lassila³

2015

EAI Endorsed Trans IoT

View full text Add to dashboard Cite

show abstract

“…Sleep power for all nodes is calculated as a fixed percentage of static power for each node type, assumed to be 16% based on a recent study of node power states [19]. The average node utilization factor used during FDLD allocations, φ E , is set as Algorithm 2 Pseudo-code for GALD-CL heuristic 1. create an initial population of chromosomes 2. while within time limit do 3. perform selection, crossover and mutation to create new chromosomes 4. for each new chromosome, evaluate: 5.…”

Section: A Experimental Setupmentioning

confidence: 99%

“…modify chromosome, return to step 5 18. trim population (with elitism) 19. end while 20. take final allocation from the best chromosome 21. calculate final execution rates (CERi(τ ), ∀i ∈ I, ∀τ ∈ N τ ) 22. calculate final cost from sum of power costs and allocation costs (P C d (τ ) and AC d (τ ), ∀d ∈ D, ∀τ ∈ N τ ) Each of five task types is representative of a different benchmark from the PARSEC benchmark suite.…”

Section: A Experimental Setupmentioning

confidence: 99%

Energy cost optimization for geographically distributed heterogeneous data centers

Jonardi

Oxley

Pasricha

et al. 2015

2015 Sixth International Green and Sustainable Computing Conference (IGSC)

View full text Add to dashboard Cite

The proliferation of distributed data centers has recently motivated researchers to study energy cost minimization at the geo-distributed level. Researchers have been using models for time-of-use (TOU) electricity pricing and renewable energy sources to help reduce energy costs when performing geographical workload distribution, but have made oversimplifying assumptions at the data center level. Important considerations such as the thermal, power, and co-location interference effects within each data center have a large impact on the performance of workload management techniques. By designing three techniques that possess varying amounts of knowledge of such information, we compare and quantify the benefits of considering detailed models at the data center level, and demonstrate that our best heuristic can on average achieve a cost reduction of 37% compared to state of the art prior work.

show abstract

“…DVFS), system-wide approaches that reduce servers' idle powers [13], per-core power gating [14], consolidation of servers during low workload periods (e.g. [15]), virtualization techniques to obtain energy savings [16], [17] or the use of low-latency power state changes in servers to improve the impact of virtualization [18]. To complement these, there are many contributions that focus on the characterization of workloads in DCs, such as [19].…”

Section: Related Workmentioning

confidence: 99%

On the feasibility of collaborative green data center ecosystems

et al. 2015

View full text Add to dashboard Cite

The increasing awareness of the impact of the IT sector on the environment, together with economic factors, have fueled many research efforts to reduce the energy expenditure of data centers. Recent work proposes to achieve additional energy savings by exploiting, in concert with customers, services' workloads and to reduce data centers' carbon footprints by adopting demandresponse mechanisms between data centers and their energy providers. In this paper, we debate about the incentives that customers and data centers should have to adopt such measures and propose a new service type and pricing scheme that is economically attractive and technically realizable. Simulation results based on real measurements confirm that our scheme can achieve additional energy savings while preserving service performance.

show abstract

Agile, efficient virtualization power management with low-latency server power states

Cited by 37 publications

References 31 publications

Optimal sleep-state control of energy-aware M/G/1 queues

Optimal sleep-state control of energy-aware M/G/1 queues

Energy cost optimization for geographically distributed heterogeneous data centers

On the feasibility of collaborative green data center ecosystems

Contact Info

Product

Resources

About