Dynamic right-sizing for power-proportional data centers

@article{Lin2011DynamicRF,
  title={Dynamic right-sizing for power-proportional data centers},
  author={Minghong Lin and Adam Wierman and Lachlan L. H. Andrew and Eno Thereska},
  journal={2011 Proceedings IEEE INFOCOM},
  year={2011},
  pages={1098-1106}
}
Power consumption imposes a significant cost for data centers implementing cloud services, yet much of that power is used to maintain excess service capacity during periods of predictably low load. This paper investigates how much can be saved by dynamically ‘right-sizing’ the data center by turning off servers during such periods, and how to achieve that saving via an online algorithm. We prove that the optimal offline algorithm for dynamic right-sizing has a simple structure when viewed in… 

Figures from this paper

Simple and effective dynamic provisioning for power-proportional data centers
TLDR
This paper develops online dynamic provisioning solutions with and without future workload information available, and designs two online algorithms with competitive ratios 2 - α and e/ (e - 1 + α), respectively, where 0 ≤ α ≤ 1 is the normalized size of a look-ahead window in which future workload Information is available.
An Online Algorithm for Power-Proportional Data Centers with Switching Cost
TLDR
This paper proposes a new online algorithm based on a regularization technique, which achieves a better competitive ratio compared to the state-of-the-art greedy algorithm in [17], and introduces a switching cost offset into the model and extends the algorithm to this new setting.
Online dynamic capacity provisioning in data centers
TLDR
This work overviews recent results showing that the optimal offline algorithm for dynamic capacity provisioning has a simple structure when viewed in reverse time, and this structure can be exploited to develop a new ‘lazy’ online algorithm which is 3-competitive.
Simple and E ective Dynamic Provisioning for Power-proportional Data Centers
  • Lu, Tan
  • Computer Science
  • 2012
TLDR
This thesis reveals an elegant structure of the dynamic provisioning problem, which allows for the optimal solution in a divide-and-conquer manner, and designs two online algorithms with competitive ratios 2 − α and e/ (e− 1 + α), respectively, where 0 ≤ α ≤ 1 is the normalized size of a look-ahead window in which future workload information is available.
Algorithmic Issues in Green Data Centers
TLDR
This thesis develops an online algorithm to make data center more power-proportional by dynamically adapting the number of active servers to match the current workload and proposes a framework to explore the Diversity of power prices and the diversity of propagation delays given geographically distributed data centers.
On Power Management Policies for Data Centers
  • Z. Haas, Shuyang Gu
  • Computer Science
    2015 IEEE International Conference on Data Science and Data Intensive Systems
  • 2015
TLDR
An efficient offline right-sizing heuristic is developed, and it is demonstrated that its performance is close to optimal, and an online algorithm is designed that computes the real-time servers' right-Sizing according to the arriving workload.
Dynamic Server Provisioning for Data Center Power Management
TLDR
This thesis demonstrates that dynamic server provisioning can successfully meet typical response time guarantees while significantly lowering power consumption, and addresses several long-standing open questions in queueing theory, including the analysis of multi-server systems with switching costs.
Quantifying the impact of shutdown techniques for energy‐efficient data centers
TLDR
The potential gain of shutdown techniques is evaluated by taking into account shutdown and boot up costs in time and energy, made on recent server architectures and future energy‐aware architectures.
Dynamic resource management in virtualized data centers with bursty traffic
TLDR
This paper develops a queueing model with a controllable service rate that accounts for switching frictions within a setting that has a changing arrival rate, and solves for the optimal control policy using dynamic programming, and finds it has intuitive structural properties.
Increasing large-scale data center capacity by statistical power control
TLDR
Ampere is a novel power management system for data centers to increase the computing capacity by over-provisioning the number of servers by using a statistical control approach to implement dynamic power management by indirectly affecting the workload scheduling.
...
1
2
3
4
5
...

References

SHOWING 1-10 OF 72 REFERENCES
Simple and effective dynamic provisioning for power-proportional data centers
TLDR
This paper develops online dynamic provisioning solutions with and without future workload information available, and designs two online algorithms with competitive ratios 2 - α and e/ (e - 1 + α), respectively, where 0 ≤ α ≤ 1 is the normalized size of a look-ahead window in which future workload Information is available.
Sierra: a power-proportional, distributed storage system
TLDR
Sierra is presented: a power-proportional, distributed storage system that achieves all these through a set of techniques including power-aware layout, predictive gear scheduling, and a replicated shortterm versioned store.
Multi-mode energy management for multi-tier server clusters
  • T. Horvath, K. Skadron
  • Computer Science
    2008 International Conference on Parallel Architectures and Compilation Techniques (PACT)
  • 2008
TLDR
Experimental results using realistic dynamic workloads based on the TPC-W benchmark show that exploiting multiple sleep states results in significant additional cluster-wide energy savings up to 23% with little or no performance degradation.
Power-Aware Speed Scaling in Processor Sharing Systems
TLDR
Study of how to optimally scale speed to balance mean response time and mean energy consumption under processor sharing scheduling shows that a simple scheme that halts when the system is idle and uses a static rate while the system are busy provides nearly the same performance as the optimal dynamic speed scaling scheme.
Energy-Aware Server Provisioning and Load Dispatching for Connection-Intensive Internet Services
TLDR
This paper characterize unique properties, performance, and power models of connection servers, based on a real data trace collected from the deployed Windows Live Messenger, and shows that these algorithms can save a significant amount of energy without sacrificing user experiences.
Power and performance management of virtualized computing environments via lookahead control
TLDR
This work implements and validate a dynamic resource provisioning framework for virtualized server environments wherein the provisioning problem is posed as one of sequential optimization under uncertainty and solved using a lookahead control scheme.
Autonomic power and performance management for computing systems
TLDR
This paper presents a theoretical framework and methodology for autonomic power and performance management in e-business data centers, and adopts mathematically-rigorous optimization approach to minimize power while meeting performance constraints.
Robust and flexible power-proportional storage
TLDR
Rabbit is a distributed file system that arranges its data-layout to provide ideal power-proportionality down to very low minimum number of powered-up nodes (enough to store a primary replica of available datasets).
Autonomic Power and Performance Management for Computing Systems
TLDR
This paper presents a theoretical framework and methodology for autonomic power and performance management in e-business data centers and adopts mathematically-rigorous optimization approach minimizing power while meeting performance constraints.
A Case For Adaptive Datacenters To Conserve Energy and Improve Reliability
TLDR
This paper identifies a new approach based on the synergy between virtual machines and statistical machine learning, and observes that constrained energy conservation can improve hardware reliability, and gives initial results on a cluster that reduces energy costs, reduces integrated circuit failures, and disk failures.
...
1
2
3
4
5
...