Q. Qiu, Q. Wu and M. Pedram, "Stochastic modeling of a power-managed system: Construction and optimization," Proc. Symp. on Low Power Electronics and Design, pages 194-199, August 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....a voltage scaling scheme for system level communication pipelines [38, 39, 40] it can reduce energy consumption compared to shutting down idle component while satisfying all QoS requirement. Qiu, Wu and Pedram proposed generalized stochastic Petri nets (GSPN) as a model for communication systems [35, 34, 36, 37]. They derived power consumption from a set of QoS constraints such as delay and jitter, and proposed a linear programming solution to finding the optimal power management policy. Although these techniques manage power for the entire communication system, they have a few limitations that cannot ....

Q. Qiu, Q. Wu, and M. Pedram. Stochastic modeling of a power-managed system: construction and optimization. In Proc. International Symposium on Low Power Electronics and Design, pages 194--199, August 1999.

.... has fueled research in 192 hardware and software power consumption estimation [176] 180] Reducing embedded software power consumption through compiler optimizations [181] source level transformations [178] 182] customized memory management schemes [183] power management schemes [168] [184], device driver and operating system policies [185] and variable voltage processors [186] 190] has been investigated. Researchers have also investigated sophisticated methods of using operating systems to dynamically disable peripherals in order to save power [191] 193] Others have advocated ....

Q. Qiu, Q. Wu, and M. Pedram, "Stochastic modeling of a power-managed system: Construction and optimization," in Proc. Int. Symp. Low Power Electronics & Design, pp. 194--199, Aug. 1999.

....aimed to achieve power reduction [10] by predicting the system idle time or event distribution and shutting down resources when idle. The simplest power management policy is time out based on a fixed or predicted amount of time before the system s shutdown or powerup [11, 12] Stochastic models [13, 14] are used to address the uncertainty in system behaviors. Simunic et al. [15] combine stochastic modeled power management techniques together with dynamic voltage scaling techniques and achieve significant power reduction in portable systems. Although the above techniques are effective to reduce ....

....scenario that has resource dependency. a) Initial schedule and power profile; b) greedy voltage clock scaling results in a power spike that violate maximum power constraint; c) a feasible solution meets both power and timing constraints, and saves energy as well. DPM LPS MS [11, 12] [13, 14] [5,3,6,7] 8, 9] Timing as constraint N N Y Y Y Power as constraint N N N N Y Timing overhead Y Y N N Y Power overhead Y Y N N Y Multiple resources N Y N Y Y Figure 3: Comparison of dynamic power management (DPM) low power scheduling (LPS) and mode selection (MS) 3 Modeling Resource ....

Q. Qiu, Q. Wu, and M. Pedram. Stochastic modeling of a power-managed system construction and optimization. In Proc. 1999.

.... can be based on fixed idle times, adaptive timeout, or predictive based on a mix of profile and runtime history [44, 41, 10, 2] The simplest power management policy is time out based on a fixed or predicted amount of time before the system s shutdown or power up [45, 16] Stochastic models [3, 34] are used to address the uncertainty in system behaviors. Simunic et al. [40] combines stochastic modeled power management with dynamic voltage scaling to achieve significant power reduction in portable systems. DPM techniques can be effective for minimizing energy and time penalties on average, ....

....especially important as we increase the dynamic range of power by increasing parallelism. Second, they have not considered inter component dependency in a system, with the exception of Qiu, Qu and Pedram in [35] which models multiple service providers and their Generalized DPM DVS MS [45, 16] [3, 34] [13, 38, 39, 36] 26, 28] Timing as constraint N N Y Y Y Power as constraint N N N N Y Timing overhead Y Y N N Y Power overhead Y Y N N Y Multiple resources N Y N Y Y Figure 3: Comparison of dynamic power management (DPM) dynamic voltage scaling (DVS) and mode selection (MS) Stochastic ....

Q. Qiu, Q. Wu, and M. Pedram. Stochastic modeling of a power-managed system construction and optimization. In Proc. 1999.

....while power modes and power consumption on peripheral devices are not considered. The cost of mode changes on the processor is often reasonably neglected [6, 9, 10] In dynamic power management techniques, researchers concentrated on systems of a single device without strong timing guarantees [2, 7, 11]. Tradeoffs are made between the power consumption and system performance. In [8] the authors did model multiple servers and relationships in the system. However, the modeled servers have identical behaviors (handling incoming requests) with the only difference in server parameters (handling ....

Q. Qiu, Q. Wu, and M. Pedram. Stochastic modeling of a power-managed system construction and optimization. In Proc. 1999.

....The advantage of the discrete time approach is that decisions are re evaluated periodically so the decision can be reconsidered thus adapting better to arrivals that are not truly geometrically distributed. An extension to the DTMDP model is a continuous time Markov decision process (CTMDP) model [10, 11]. In CTMDP power manager (PM) issues commands upon event occurrences instead of at discrete time settings. System transitions are assumed to follow exponential distribution. We show that models that are exclusively based on the exponential arrival times in the idle state can have high energy costs ....

Q. Qiu, Q. Wu and M. Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimization", Proceedings of ISLPED, pp. 194-199, 1999.

....the power consumption at the instruction level for different CPU and DSP architectures and derived specific power optimizing compilation strategies. Other approaches focus on a whole system in order to optimize for low power. System power management approaches have been explored by Qiu et al.[11], among others. 3 Code Compression This section is organized as follows: we first introduce the basics of code compression and define terms we will use subsequently. We then proceed to an outline of our algorithm, and we explain its advantages over other methods. We also give implementation ....

Q. Qiu, Q. Wu, M Pedram, Stochastic Modeling of a Power--Managed System: Construction and Optimization, IEEE/ACM Proc. of International Symposium on Low Power Electronics and Design (ISLPED'99), pp. 194-199, 1999.

....between off and active states are best fit with uniform distribution. Our model has two non exponential transitions occurring at the same time when the card transitions from doze mode into off state. Thus we could not apply policy optimization algorithms based on exponential models, such as [9, 10, 13]. Large errors result if exponential distribution is used for all transitions, as was shown in [11] Another approach to handle non exponential transitions is to use adaptive method, such as in [12] This method requres policy interpolation at very short time increments, regardless of the device ....

....savings with low performance penalty. The competitive algorithm [15] guarantees to be within a factor of two of oracle policy. Although its power consumption is low, it has a performance penalty that is an order of magnitude larger than for our policy. A policy that assumes Poisson arrivals only [13] has a very large performance penalty because it makes the decision as soon as the system enters doze state. Table 2 shows the measurement results for a 2hr telnet trace. Again our policy performs best, with a factor of five in power savings and a small performance penalty. Telnet application ....

Q. Qiu, Q. Wu and M. Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimization", Proceedings of ISLPED, 1999.

.... research in co estimation of hardware and software power consumption [16, 17, 18] Optimization of embedded software system power consumption through compiler optimizations [19] source level transformations [16, 20] customized memory management schemes [21] use of power management schemes [10, 22], and through variable voltage processors [23, 24, 25] has been investigated. Our work draws on state of the art research in the areas of processor, application specific hardware, and memory power estimation in order to provide embedded software designers new perspectives on system power ....

Q. Qiu, Q. Wu, and M. Pedram, "Stochastic modeling of a power-managed system: Construction and Optimization," in Proc. Int. Symp. Low Power Electronics & Design, Aug. 1999.

No context found.

Q. Qiu, Q. Wu and M. Pedram, "Stochastic modeling of a power-managed system: Construction and optimization," Proc. Symp. on Low Power Electronics and Design, pages 194-199, August 1999.

No context found.

Q. Qiu, Q. Wu, M. Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimization", Proceedings of the International Symposium on Low Power Electronics and Design, 1999.

No context found.

Q. Qiu, Q. Wu and M. Pedram, "Stochastic modeling of a powermanaged system-construction and optimization," IEEE Transactions on Computer-Aided Design, pp. 1200-1217, Oct. 2001.

No context found.

Qinru Qiu, Qing Wu and Massoud Pedram, "Stochastic modeling of a power-managed system-construction and optimization," Computer-Aided Design of Integrated Circuits and Systems, IEEE Transactions on, pp. 1200-1217, Oct. 2001.

No context found.

Q. Qiu, Q. Wu, M. Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimization", Proceedings of the International Symposium on Low Power Electronics and Design, 1999.

....are determined, an optimal power management policy for achieving the best power delay trade off in the system can be generated. In [9] the authors improved [8] by modeling the power managed system using a continuous time Markov decision process (CTMDP) Further research results can be found in [10] [13] The DPM approaches based on Markov decision processes offer significant improvements over heuristic power management policies in terms of the theoretical framework and ability to apply strong mathematical optimization techniques [14] 20] However, previous works based on Markov decision ....

....behaviors that are present. For example, we need to consider the synchronization of LSQs and SPs, the synchronization of the SR and LSQs, the dispatch behavior of the RD, and so on. In this situation when complex system behaviors is a major part of the system model, the modeling techniques in [8] [10] become powerless because they only offer stochastic models for individual components and require that global system behaviors be captured manually. Obviously, we need new DPM modeling techniques for large systems with complex behaviors. In this work, we first present a methodology based on ....

[Article contains additional citation context not shown here]

Q. Qiu, Q. Wu, M. Pedram, "Stochastic Modeling of a PowerManaged System: Construction and Optimization", Proceedings of the International Symposium on Low Power Electronics and Design, pp. 194-199, Aug. 1999.

....are determined, an optimal power management policy for achieving the best power delay trade off in the system can be generated. In [9] the authors extend [8] by modeling the power managed system using a continuous time Markov decision process (CTMDP) Further research results can be found in [10] [13] In situations where complex system behaviors, such as concurrency, synchronization, mutual exclusion, and conflict, are present, the modeling techniques in [8] 10] become inadequate because they are effective only when constructing stochastic models of simple systems consisting of ....

.... the power managed system using a continuous time Markov decision process (CTMDP) Further research results can be found in [10] 13] In situations where complex system behaviors, such as concurrency, synchronization, mutual exclusion, and conflict, are present, the modeling techniques in [8] [10] become inadequate because they are effective only when constructing stochastic models of simple systems consisting of non interacting components. In [14] a technique based on controllable generalized stochastic Petri nets (GSPN) with cost is proposed that is powerful enough to compactly model a ....

[Article contains additional citation context not shown here]

Q. Qiu, Q. Wu, and M. Pedram, "Stochastic Modeling of a PowerManaged System: Construction and Optimization," Intl. Symp. on Low Power Electronics and Design, 1999.

No context found.

Qinru Qiu, Stochastic Modeling of a Power-Managed System: Construction and Optimization, Dissertation, Dec. 2000.

No context found.

Q. Qiu, Q. Wu, M. Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimization", Proceedings of the International Symposium on Low Power Electronics and Design, 1999.

No context found.

Q. Qiu, Q.Wu, M.Pedram, "Stochastic Modeling of a Power-Managed System: Construction and Optimisation", IEEE Transactions on Computer Aided Design ,Vol.20, no.9, Oct2001, pp. 1200-1217.

No context found.

Q. Qiu and M. Pedram, "Stochastic Modeling of a Power Managed System-Construction and optimization", IEEE Transactions on CAD, pp. 1200-- 1217, vol. 20, Oct 2001.

No context found.

Q. Qiu, Q. Wu, and M. Pedram, "Stochastic Modeling of a PowerManaged System: Construction and Optimization," Proc. Int'l Symp. Low Power Electronic Devices, pp. 194-199, 1999.

No context found.

Q. Qiu, Q. Wu, and M. Pedram. Stochastic modeling of a power-managed system: construction and optimization. IEEE Transactions on Computer Aided Design, 20(9):1200-1217, 2001.

No context found.

Q. Qiu, Q. Wu and M. Pedram. Stochastic Modeling of a Power-Managed System: Construction and Optimization. In Proceedings of the International Symposium on Low Power Electronics and Design, pages 194-199, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC