|
Improving
the power consumption attributes of embedded and real-time systems has
become an important area of interest as processors and other system
components have become increasingly powerful and demanding in their
energy consumption. The
focus of this thesis is on examining the characteristics of real-time
hard-disk scheduling algorithms, as hard-disk power usage can amount to
a substantial portion of the total system power consumption.
Conventional disk scheduling algorithms typically either disregard power
consumption completely, or at best apply a fairly naïve power
management policy. Given
the potential implications for improving the efficiency and longevity of
real-world disk systems, we investigate
the problem of scheduling a set of independent real-time disk requests
such that the total power consumption is minimized, while the efficacy
of the disk system is not compromised.
We define a power consumption model that can reasonably
approximate the performance characteristics of real-world disks.
Next, we
discuss two power-aware power management
policies, I/O Burstiness for Energy Conservation (IBEC)
and Speed-Aware Real-time Disk Scheduling for energy conservation (SARDS),
which integrate differing power management policies into the disk
scheduling algorithms for real-time I/O-intensive applications. Furthermore, to evaluate the performance of the proposed
algorithms against existing solutions and ensure that the efficacy of
the system is not compromised, we incorporate the earliest deadline
first (EDF) and least
laxity first (LLF) scheduling policies into SARDS and IBEC to implement
power-aware real-time scheduling algorithms.
Experimental results from real-world traces and synthetically
generated workloads show that the dynamic algorithms have
the potential to substantially reduce power consumption over existing
scheduling algorithms and power management policies without compromising
the overall performance of disk systems.
Keywords:
SARDS;
IBEC; power-aware scheduling; real-time; guarantee ratio; power
consumption; hard-disk; power modeling; simulation
|