## Algorithmic problems in power management (2005)

Venue: | SIGACT News |

Citations: | 55 - 3 self |

### BibTeX

@ARTICLE{Irani05algorithmicproblems,

author = {Sandy Irani and Kirk R. Pruhs},

title = {Algorithmic problems in power management},

journal = {SIGACT News},

year = {2005},

volume = {36},

pages = {2005}

}

### Years of Citing Articles

### OpenURL

### Abstract

We survey recent research that has appeared in the theoretical computer science literature on algorithmic

### Citations

3667 |
L.: Convex Optimization
- BOYD, VANDENBERGHE
- 2004
(Show Context)
Citation Context ...t λi, i = 1, . . ., n be a variable (Lagrangian multiplier) associated with the function fi(x). Then a necessary and sufficient KKT conditions for solutions x and λ to be primal and dual feasible are =-=[10]-=-: ∇f0(x) + fi(x) ≤ 0 i = 1, . . ., n (1) λi ≥ 0 i = 1, . . ., n (2) λifi(x) = 0 (3) n� λi∇fi(x) = 0 (4) i=1 To state the energy minimization problem as a convex program, we break time into intervals t... |

1029 | An Energy-Efficient MAC Protocol for Wireless Sensor Networks
- Ye, Heidmann, et al.
- 2002
(Show Context)
Citation Context ...ng down devices will also be an important mechanism in minimizing energy consumption in sensor networks. Researchers have observed that a large fraction of wasted energy happens when sensors are idle =-=[38, 15, 13]-=- and have turned their attention to research issues related to powering down sensors when not receiving or transmitting. Different researchers assume different models as to how this can be achieved. S... |

482 | Next century challenges: mobile networking for smart dust
- Kahn, Katz, et al.
- 1999
(Show Context)
Citation Context ...-aware computation is absolutely critical. The explosive interest in sensor networks is the result of the development of low-cost, low-power mutifunctional sensor devices, such as the Smart Dust Mote =-=[1, 22]-=-, that are small in size and communicate untethered at short distance. There is an inherent conflict between power reduction and performance; in general, the more power that is available, the better t... |

393 | A scheduling model for reduced CPU energy
- Yao, Demers, et al.
- 1995
(Show Context)
Citation Context ...ormally, speed scaling problems involve determining the speed of the processor at each point in time. Theoretical investigations of speed scaling algorithms were initiated by Yao, Demers, and Shankar =-=[37]-=-. Yao et al. [37] propose formulating speed scaling problems as scheduling problems. The setting is a collection of tasks, where each task i has a release time ri when it arrives into the system, and ... |

317 | Temperature-aware microarchitecture: Modeling and implementation
- Skadron, Stan, et al.
- 2004
(Show Context)
Citation Context ...ractice. 1.1 Motivation The power consumption rate of computing devices has been increasing exponentially. Since the early 1970s, the power densities in microprocessors have doubled every three years =-=[34]-=-. This increased power usage poses two types of difficulties: • Energy Consumption: As energy is power integrated over time, supplying the required energy may become prohibitively expensive, or even t... |

297 |
Design challenges of technology scaling
- Borkar
- 1999
(Show Context)
Citation Context ... is the only way to combat leakage loss, which is the energy lost independent of the device’s state. It is predicted that leakage loss will become comparable to switching loss in the near term future =-=[9]-=-. Thus, power-gating strategies will become increasingly important at the architectural level. At a higher level, the powered-down component might be a disk drive or even the whole system (e.g., a lap... |

179 | Speed is as powerful as clairvoyance
- Kalyanasundaram, Pruhs
(Show Context)
Citation Context ...l schedule can change radically with an infinitesimal change in E, making the changes difficult or potentially impossible to track. It is possible to get a positive result using resource augmentation =-=[23, 29]-=-. That is, one can get an (1 + ɛ)-speed O(1)-approximate polynomial time algorithm [5]. Online Algorithms for Flow Time: In most situations, the system does not know of a task until its release time. ... |

135 | Power: A first-class architectural design constraint
- Mudge
- 2001
(Show Context)
Citation Context ...r, see http : //www.cs.pitt.edu/ ∼ kirk/cool.avi.(You will need a DivX codec installed.) These two factors have resulted in power becoming a first-class design constraint for modern computing devices =-=[28]-=-. There is an extensive literature on power management in computing devices. Overviews can be found in [11, 28, 36]. All of these techniques that have been investigated are similar in that they reduce... |

134 |
Power-Aware Microarchitecture: Design and Modeling Challenges for NextGeneration Microprocessors
- Brooks
(Show Context)
Citation Context ...have resulted in power becoming a first-class design constraint for modern computing devices [28]. There is an extensive literature on power management in computing devices. Overviews can be found in =-=[11, 28, 36]-=-. All of these techniques that have been investigated are similar in that they reduce or eliminate power to some or all components of the device. Sensor networks have emerged as an important new ∗ Sup... |

110 | Speed scaling to manage energy and temperature
- Bansal, Kimbrel, et al.
(Show Context)
Citation Context ...line feasibility; that is, each task must finish by its deadline. They study the problem of minimizing the total energy used subject to the deadline feasibility constraints. Bansal, Kimbrel and Pruhs =-=[7, 8]-=- study the problem of minimizing the maximum temperature attained subject to the deadline feasibility constraints. 2.2 Energy and Temperature Before proceeding further, we need to explain how speed, p... |

106 |
The case for higher-level power management
- Ellis
- 1999
(Show Context)
Citation Context ...mine how essential performance is at any given time and how to apply a particular power reduction technique. Current tools and mechanisms for power management are inadequate and require more research =-=[14]-=-. Furthermore, there is a growing consensus that these policies must incorporate information provided by applications and high levels of the operating system in order to achieve necessary advances [14... |

75 | Reducing power in high-performance microprocessors
- Tiwari, Singh, et al.
- 1998
(Show Context)
Citation Context ...have resulted in power becoming a first-class design constraint for modern computing devices [28]. There is an extensive literature on power management in computing devices. Overviews can be found in =-=[11, 28, 36]-=-. All of these techniques that have been investigated are similar in that they reduce or eliminate power to some or all components of the device. Sensor networks have emerged as an important new ∗ Sup... |

74 |
Real-Time Computing Systems
- Hard
- 2005
(Show Context)
Citation Context ...ay be natural deadlines for the various tasks imposed by the application. In other settings, the system may impose deadlines to better manage tasks or insure a certain quality of service to each task =-=[12]-=-. Yao et al. [37] assume that the system’s performance measure is deadline feasibility; that is, each task must finish by its deadline. They study the problem of minimizing the total energy used subje... |

64 | Low power distributed MAC for ad hoc sensor radio networks
- Zhong, Guo, et al.
- 2001
(Show Context)
Citation Context ...ng down devices will also be an important mechanism in minimizing energy consumption in sensor networks. Researchers have observed that a large fraction of wasted energy happens when sensors are idle =-=[38, 15, 13]-=- and have turned their attention to research issues related to powering down sensors when not receiving or transmitting. Different researchers assume different models as to how this can be achieved. S... |

58 | Optimal voltage allocation techniques for dynamically variable voltage processors
- Kwon, Kim
(Show Context)
Citation Context ...r jobs with a fixed priority, Yun and Kim [39] show that it is NP-hard to compute an minimum energy schedule. They also give a fully polynomial time approximation scheme for the problem. Kwon and Kim =-=[25]-=- give a polynomial time algorithm for the case of a processor with discrete speeds. In the online version of the problem, an algorithm only learns about a task at its release time, at which time it is... |

49 | Getting the best response for your erg
- Pruhs, Uthaisombut, et al.
- 2008
(Show Context)
Citation Context ...and one needs a schedule that doesn’t deplete the battery. 6sThe one paper that we are aware of that considers a power management problem for a scheduling objective other than deadline feasibility is =-=[30]-=-. In [30], Pruhs, Uthaisombut and Woeginger consider the problem of minimizing the total flow time subject to a bound E on the total energy used. The flow time of a task i is Fi = Ci − ri, which is th... |

45 | Competitive analysis of dynamic power management strategies for systems with multiple power saving states
- Irani, Shukla, et al.
- 2002
(Show Context)
Citation Context ...Power Management (DPM) in the systems literature 8srefers to the problem of developing policies for powering down based on the dynamically changing system state, functionality and timing requirements =-=[17, 19]-=-. We will use the more descriptive term Power-Down (PD) for this problem. There are numerous examples of systems that can be run at multiple speeds, have a sleep state, and receive tasks with deadline... |

34 |
Online strategies for dynamic power management in systems with multiple power-saving states
- irani, Shukla, et al.
(Show Context)
Citation Context ...Power Management (DPM) in the systems literature 8srefers to the problem of developing policies for powering down based on the dynamically changing system state, functionality and timing requirements =-=[17, 19]-=-. We will use the more descriptive term Power-Down (PD) for this problem. There are numerous examples of systems that can be run at multiple speeds, have a sleep state, and receive tasks with deadline... |

33 | Optimal powerdown strategies
- augustine, Irani, et al.
(Show Context)
Citation Context ... to skip some sleep states. In this scenario, an algorithm must identify an optimal sequence of states as well as the time thresholds at which these states will be reached. Augustine, Irani and Swamy =-=[6]-=- develop a deterministic algorithm that achieves a competitive ratio of 8 for any system with arbitrary transition costs. The bound improves to 3 + 2 √ 2 ≈ 5.8284 under the reasonable assumption that ... |

28 |
Speed scaling to manage temperature
- Bansal, Pruhs
- 2005
(Show Context)
Citation Context ...a while, and then cools to the final temperature along an Euler-Lagrange curve. After some work, this gives sufficient information to efficiently compute a separating hyperplane Bansal and Pruhs show =-=[8]-=- that while the YDS schedule may not be optimal for temperature, YDS is cooling oblivious (recall, this means O(1)-approximate for temperature). Thus, this constructively shows that there are schedule... |

24 | Online computation
- Irani, Karlin
- 1997
(Show Context)
Citation Context ...ks only to minimize the expected cost. We call algorithms in this latter category probability-based. The power-down problem for two states is a continuous version of the well-known Ski Rental Problem =-=[18]-=-, and is well understood. There is a simple 2-competitive algorithm, which stays in the active state until the total energy consumed is equal to the transition energy. It is known that this algorithm ... |

24 |
Min-energy voltage allocation for treestructured tasks
- Li, Liu, et al.
- 2006
(Show Context)
Citation Context ...uced bt the YDS algorithm clearly has these properties and hence is optimal. A naive implementation of YDS runs in time O(n 3 ). This can be improved to O(n 2 ) if the intervals have a tree structure =-=[26]-=-. It would be interesting to see if the cubic running time of YDS for arbitrary instances can be improved. For jobs with a fixed priority, Yun and Kim [39] show that it is NP-hard to compute an minimu... |

16 | Latency Effects of System Level Power Management Algorithms
- Ramanathan, Irani, et al.
- 2000
(Show Context)
Citation Context ...o the active state will serve to delay the execution of tasks which will in turn shorten future idle periods. The interplay between power-down policies and latency has been examined experimentally in =-=[31, 32]-=- in the context of disk-drives. We would like to understand more formally, how the choice of power-down policy effects latency over a series of idle periods and develop algorithms which allow a user o... |

13 |
Randomized competitive algorithms for non-uniform problems
- Karlin, Manasse, et al.
- 1990
(Show Context)
Citation Context ...at this algorithm is optimally competitive. Furthermore, if the idle period is generated by a known probability distribution, then there is a probability-based algorithm that is (e/e − 1)-competitive =-=[24]-=-, and this is optimally competitive. For some systems, like disk drives, the energy needed and time spent to go from a higher power state to a lower power state is negligible. Irani et al. show in [17... |

12 | Procrastination scheduling in fixed priority real-time systems
- Jejurikar, Gupta
- 2004
(Show Context)
Citation Context ..., until waiting any longer would require running faster than the critical speed in order to complete all jobs by their deadlines. This idea of ‘procrastination scheduling’ has been studied further in =-=[20]-=- who examine the combined SS-PD problem under different assumptions about how jobs can be scheduled. 3.2 Open Problems There are still important research questions that remain open in this area. Study... |

11 |
Jiri Sgall, and Eric Torng. On-line scheduling
- Pruhs
- 2004
(Show Context)
Citation Context ...l schedule can change radically with an infinitesimal change in E, making the changes difficult or potentially impossible to track. It is possible to get a positive result using resource augmentation =-=[23, 29]-=-. That is, one can get an (1 + ɛ)-speed O(1)-approximate polynomial time algorithm [5]. Online Algorithms for Flow Time: In most situations, the system does not know of a task until its release time. ... |

10 |
On energy-optimal voltage scheduling for fixed priority hard real-time systems
- Yun, Kim
(Show Context)
Citation Context ...n 2 ) if the intervals have a tree structure [26]. It would be interesting to see if the cubic running time of YDS for arbitrary instances can be improved. For jobs with a fixed priority, Yun and Kim =-=[39]-=- show that it is NP-hard to compute an minimum energy schedule. They also give a fully polynomial time approximation scheme for the problem. Kwon and Kim [25] give a polynomial time algorithm for the ... |

6 |
A.: Thermal Management Handbook
- Seargeant, Krum
- 1998
(Show Context)
Citation Context ... is the speed of the device at time We now turn our attention to temperature. Cooling, and hence temperature, is a complex phenomenon that can not be modeled completely accurately by any simple model =-=[33]-=-. In [7], Bansal Kimbrel and Pruhs propose a model in which the environmental temperature is assumed to be constant. While this assumption certainly is not strictly true, the hope is that it is suffic... |

5 | An analysis of system level power management algorithms and their effects on latency
- RAMANATHAN, IRANI, et al.
(Show Context)
Citation Context ...o the active state will serve to delay the execution of tasks which will in turn shorten future idle periods. The interplay between power-down policies and latency has been examined experimentally in =-=[31, 32]-=- in the context of disk-drives. We would like to understand more formally, how the choice of power-down policy effects latency over a series of idle periods and develop algorithms which allow a user o... |

2 |
Algorithms for Power Savings
- Gupta, Shukla
- 2003
(Show Context)
Citation Context ...at takes as input a probability distribution over the length of the next idle time and a description of the system and outputs a power-down strategy that minimizes the expected energy consumption. In =-=[16]-=- we (joint with Rajesh Gupta and Sandeep Shukla) initiated a theoretical investigation into Speed Scaling with Power-Down (SS-PD). This work assumes two states, an active state and a sleep state. The ... |

1 |
aware wireless microsensor systems
- Power
- 2002
(Show Context)
Citation Context ...ng down devices will also be an important mechanism in minimizing energy consumption in sensor networks. Researchers have observed that a large fraction of wasted energy happens when sensors are idle =-=[38, 15, 13]-=- and have turned their attention to research issues related to powering down sensors when not receiving or transmitting. Different researchers assume different models as to how this can be achieved. S... |

1 |
Leakage aware energy efficient task scheduling in embedded real-time systems
- Jejurikar, Pereira, et al.
- 2004
(Show Context)
Citation Context ...ate an idle period in which the device can sleep. Work in the systems literature has further explored this idea in order to understand where this critical speed lies based on different leakage models =-=[21]-=-. All of the algorithms developed for SS-PD operate on the principle that if a device is shut down, it will continue to remain in that state, even if there are pending jobs, until waiting any longer w... |

1 |
unpublished manuscript. 11
- Augustine, Irani, et al.
- 2004
(Show Context)
Citation Context ...cult or potentially impossible to track. It is possible to get a positive result using resource augmentation [23, 29]. That is, one can get an (1 + ɛ)-speed O(1)-approximate polynomial time algorithm =-=[5]-=-. Online Algorithms for Flow Time: In most situations, the system does not know of a task until its release time. Thus we would like to develop and analyze online algorithms for this problem. The firs... |