#### DMCA

## Practical PACE for Embedded Systems (2004)

Venue: | In Proceedings of ACM International Conference on Embedded Software (EMSOFT |

Citations: | 33 - 5 self |

### Citations

10435 | Introduction to Algorithms
- Cormen, Leiserson, et al.
- 1990
(Show Context)
Citation Context ...teed to be found. However, the running time of the algorithm still has no polynomial time bound guarantee. Inspired by the fully polynomial time approximation scheme (FPTAS) of the subset-sum problem =-=[5]-=-, we obtain a FPTAS for our problem, presented next. The FPTAS, further reduces the size of LABEL(i). 5.3 Eliminations Affecting Optimality The intuition for the FPTAS is that we need to further trim ... |

1475 |
and Combinatorial Programming
- Linear
- 1976
(Show Context)
Citation Context ... ey = f α y , where α ≥ 2. Then, we have the following mathematical programming problem: Minimize Subject to ∑ 1≤y≤W C ∑ fy 1≤y≤W C (1 − cdf(y − 1)) · f α y 1 ≤ D whichcanbesolvedbyLagrangiantechnique=-=[13]-=-orJensen’s inequality [9]. 2.2 Piecewise constant speed schedules Obviously, computing speed for each cycle is too overwhelming considering that a task usually takes millions of cycles. Furthermore, t... |

542 | A scheduling model for reduced CPU energy
- Yao, Demers, et al.
- 1995
(Show Context)
Citation Context ...ry life. Dynamic voltage scaling (DVS) allows a processor to change the clock frequency and supply voltage on-the-fly to trade performance for energy conservation. Using DVS, quadratic energy savings =-=[7, 19]-=- can be achieved at the expense of just linear performance loss. Various chip makers, such as Transmeta, Intel, and AMD, manufacture processors with this feature. When applied to real-time embedded sy... |

390 | Voltage scheduling problem for dynamically variable voltage processors
- Ishihara, Yasuura
- 1998
(Show Context)
Citation Context ...2], assuming continuous speeds. If a CPU only offers a fixed, limited set of valid speeds, using the two speeds which are immediate neighbors to the optimal speed will minimize the energy consumption =-=[8]-=-. Note that this requires intra-task speed change, which can be realized through a timeout mechanism, or by inserting system calls into the task at compile time. However, more often than not, a task’s... |

202 | Dynamic and aggressive scheduling techniques for power-aware real-time systems
- Aydin, Melhem, et al.
- 2001
(Show Context)
Citation Context ...er of cycles it will execute) is known deterministically, it has been shown that one can safely commit to a constant optimal CPU speed during the execution of atask to minimize the energy consumption =-=[2]-=-, assuming continuous speeds. If a CPU only offers a fixed, limited set of valid speeds, using the two speeds which are immediate neighbors to the optimal speed will minimize the energy consumption [8... |

173 | Improving dynamic voltage scaling algorithms with PACE
- Lorch, Smith
- 2001
(Show Context)
Citation Context ...onal requirement is only known probabilistically, there is no constant optimal speed for the task and the expected energy consumption is minimized by gradually increasing speed as the task progresses =-=[11]-=-. It is possible to find the optimal speed schedule if we assume continuous speed and a welldefined power function, which are assumptions that do not hold in practice. In this paper, we study the prob... |

171 |
Power-Aware Microarchitecture: Design and Modeling Challenges for Next-Generation Microprocessors
- Brooks, Bose, et al.
(Show Context)
Citation Context ...e also tackled this problem, but with a completely different solution [6], which is geared towards soft real-time tasks. Both PACE and GRACE apply the well-known cubicroot rule of the power functions =-=[3]-=- and hence use e(f)= αf 2 + β (α and β are constants) to approximate the actual energy/cycle function. This is based on the fact that in CMOS circuits, the dominant component of power consumption is p... |

144 | Energy-Efficient Soft Real-Time CPU Scheduling for Mobile Multimedia Systems
- Yuan, Nahrstedt
- 2003
(Show Context)
Citation Context ...speeds and well-defined power functions, such as perfectly cubic power versus frequency function. These assumptions typically do not hold in practice. A number of schemes, such as PACE [11] and GRACE =-=[20]-=-, have proposed to use well-defined functions to approximate the actual power function and solve the continuous version of the problem before rounding the speeds to the available 54discrete speeds. T... |

135 | Design issues for dynamic voltage scaling
- Burd, Brodersen
(Show Context)
Citation Context ...ke into consideration the speed change overheads with a slight modification. Since the time penalty and energy penalty for a speed change depend on the speed before changing and the speedafterchanging=-=[4]-=-, weonlyneedtomodifyLine7ofthe PPACE algorithm such that if the chosen frequency for the current phase is different from that for the previous phase, we add the energy overhead to the energy component... |

125 | Synthesis techniques for low-power hard real-time systems on variable voltage processors
- Hong, Qu, et al.
- 1998
(Show Context)
Citation Context ...ry life. Dynamic voltage scaling (DVS) allows a processor to change the clock frequency and supply voltage on-the-fly to trade performance for energy conservation. Using DVS, quadratic energy savings =-=[7, 19]-=- can be achieved at the expense of just linear performance loss. Various chip makers, such as Transmeta, Intel, and AMD, manufacture processors with this feature. When applied to real-time embedded sy... |

87 |
real-time scheduling for low-energy using stochastic data and DVS processors
- Hard
- 2001
(Show Context)
Citation Context ...WN SCHEMES In this section, we give a brief review and discussion of the schemes provided by PACE and GRACE. Other researchers have also tackled this problem, but with a completely different solution =-=[6]-=-, which is geared towards soft real-time tasks. Both PACE and GRACE apply the well-known cubicroot rule of the power functions [3] and hence use e(f)= αf 2 + β (α and β are constants) to approximate t... |

57 | Practical voltage-scaling for fixed-priority RT-systems
- Saewong, Rajkumar
- 2003
(Show Context)
Citation Context ...xpected energy consumed by tasks (both PACE and GRACE ignore the idle power). Not only this assumption does not hold in practice 1 , but also could lead to an abnormality called inefficient frequency =-=[14]-=-. For example, for the IBM PPC405LP (Table 1), the energy consumed by each cycle running at frequency 266MHz, 600 ≈ 2.2556, is higher than the en266 ergy consumed by each cycle running at frequency 33... |

57 | Compile-time dynamic voltage scaling settings: Opportunities and limits
- Xie, Martonosi, et al.
- 2003
(Show Context)
Citation Context ...is beyond the scope of this paper. The interested reader is referred to [1], which assumed the sequential form of program execution, where a program can be divided into nsegments of equal length, and =-=[16]-=-, which made use of compiler and placed the transition points on the edges of the control flow graph of the program. 552.3 The Impact of Idle Power In general, if the CPU has no task to execute, it w... |

33 | Collaborative operating system and compiler power management for real-time applications
- AbouGhazaleh, Mosse, et al.
- 2003
(Show Context)
Citation Context ...a task usually takes millions of cycles. Furthermore, the ability to change speed in any cycle is unreasonable since real-world operating systems have some granularity requirement for changing speeds =-=[1, 12]-=-. Hence, in practice, we should probably change the speed for a task no more than some reasonable number of times. Thus, we need a schedule with a limited number of transition points at which speed ma... |

30 |
Operating System Modifications for Task-based Speed and Voltage Scheduling
- Lorch, Smith
- 2003
(Show Context)
Citation Context ...s M−2 done by computing a cumulative cdf, defined as ccdf(k) = ∑k i=1 cdf(i), before the brute force search. This algorithm is easy to implement and good for small M and r. For example, the system in =-=[12]-=- has only 3available discrete frequencies; if we set r = 100, the number of possibilities is just 103. But for large M and r, this algorithm be((r+M−1) comes impractical since Θ M−2 ) (( ) ) M−2 r+M−1... |

15 | Operating systems techniques for reducing processor energy consumption
- Lorch
- 2001
(Show Context)
Citation Context ...E proposes to linearly scan all the phases to adjust the speeds. But this still does not guarantee that the deadline will not be missed. If this happens, PACE will simply use a default speed schedule =-=[10]-=-. 56900 800 Actual descrete power function Least square fitting with ax 3 +b 1800 1600 Actual descrete power function Least square fitting with ax 3 +b 700 1400 600 1200 Power (mW) 500 400 Power (mW)... |