Dynamic power management

1
Dynamic power management

• Introduction
• Implementation, levels of operation
• Modeling
• Power and performance issues regarding power
  management
• Policies
• Conclusions

2
Introduction

• To provide the requested services and performance
  levels with a minimum number of active components
  or a minimum load on such components.
• Assume non-uniform workload.
• Assume predictability of workload.
• Low overhead of caused by power manager
  performance and power.

3
Introduction

• The power manager (PM) implements a control
  procedure based on observations and assumptions
  about the workload.
• The control procedure is called a policy.
• Oracle power manager

4
Implementation

• Hardware
• Frequency reduction
• Supply voltage
• Power shutdown
• Software
• Mostly used
• Most flexible
• Operative system power manager (OSPM)
• Microsofts OnNow
• ACPI

5
Modeling

• View the system as a set of interacting
  power-manageable components (PMCs), controlled by
  the power manager (PM).

6
Modeling

• Independent PMCs.
• Model PMCs as FSMs PSMs
• Transition between states have a cost.
• The cost is associated with delay, performance
  and power loss.
• Service providers and service requesters.

7
Modeling

• Ex. StrongArm SA-1100 processor (Intel)

8
Power and performance issues..

• Power management degrades performance.

9
Power and performance issues..

• Break-even time Tbe - minimum length of an idle
  period to save power. Move to sleep state if
  Tidle gt Tbe
• T0 Transition delay (shutdown and wakeup)
• E0 Transition energy
• Ps , Pw Power in sleeping and working states

10
Policies

• Different categories
• Predictive
• Adaptive
• Stochastic
• Application dependent
• Statistical properties
• Resource requirements

11
Policies - Predictive

• Fixed time-out
• Static
• Assume that if a device is idle for ?, it will
  remain idle for at least Tbe.
• If device idle for ?, change state to sleep.
• Time-out ? is computed and set off-line.
• Very simple to implement. Requires a timer.
• Power is wasted in waiting for time-out.
• Can cause many under-predictions.
• Adaptive version where ? is adjusted online.

12
Policies - Predictive

• Predictive shut-down Golding 1996
• Take decisions based observations of past idle
  and busy times. Take decision as soon as an idle
  time starts.
• The equation f yields a predicted idle time Tpred
• Shut down if
• Sample data and fit data to a non-linear
  regression equation f (off-line).
• Computation and memory requirements.

13
Policies - Predictive

• Predictive shut-down Srivastava 1996
• Take decision based on observing the last busy
  time. Take decision as soon as an idle time
  starts.
• If change state.
• Suitable for devices where short busy periods are
  followed by long idle periods. L-shape plot
  diagrams (idle period vs busy periods).
• FSMs similar to multibit branch prediction in
  processors.
• Predictive wake-up

14
Policies - Adaptive

• Static policies are ineffective when the workload
  is nonstationary or not known in advance.
• Time-out revisited
• 1. Adapt the time-out ?.
• 2. Keep a pool of time-outs and choose the one
  that will perform best in this
  context.
• 3. As above, but assign a weight to each time-out
  according to how well it will perform relative
  to an optimum strategy for the last requests.

15
Policies - Adaptive

• Low pass filter Wu1997

16
Policies - Stochastic

• Predictive and adaptive policies lack some
  properties
• They are based on a two state system model.
• Parameter tuning can be hard.
• Stochastic policies provide a more general and
  optimal strategies.
• Modeled by Markov chains, Pareto.

17
Policies - Stochastic (Markov)

• Stationary (or WSS). Statistical properties do
  not depend on the time shift, k.

• A set of states. Probability associated with the
  transitions.
• The solution of the LP produces stationary,
  randomized (nondeterministic) policy.
• Finding the minimum power policy that meets a
  given performance constraint can be cast as a
  linear program (LP, solved in polynomial time).

18
Policies - Stochastic (Markov)
19
Policies - Stochastic (Markov)

• The policy computed by LP is globally optimum
  Puterman 1994.
• However, requires knowledge of the system and its
  workload statistics in advance.
• An adaptive extension Chung 1999
• Policy precharacterization (PC)
• Parameter learning (PL)
• Policy interpolation (PI)

20
Policies - Stochastic (Markov)

• An adaptive(cont.)
• Two-parameters Markov. Parameters describe the
  current workload.
• PC constructs a 2-dim table, addressed by the
  values of the two parameters.
• The table elements contain the optimal policy,
  identified by the pair.
• Parameter learning is performed during operation.
• PI is performed to find a policy as a combination
  of the nearby policies given by the table and the
  parameters.

21
Conclusions

• The policies are application dependent and have
  to be adopted to devices.
• Policies based on stochastic control and
  specially Markov allows a flexible and general
  design, where all requirements can be
  incorporated.
• Current models are based on observing requests
  arrivals. A trend in power management is to
  include higher-level information, particularly
  software-based information from compilers and
  OSs.