Power QoS

Design Scenarios

Why do we need Quality of Service power management (QoSPM)? QoSPM focuses on the power management problem from a usability perspective, and it enables the power/performance decisions to be made by the lowest levels where most of the specific power/performance hardware knowledge tends to reside. Usability can be expressed in terms latencies (wake up and operational) and throughputs.

Why consider QoS for PM? Design scenarios Here are some scenarios of how QoSPM parameters could be used to define a power management policy.

Note: Not all of these scenarios are feasible right now. But they are examples of how you can define power-performance policy from latency (wakeup and operational), throughput, and idle time out parameters.

• Improved CPU Idle QoSPM-requested wake up latencies constrain how deep of sleep state the CPU is allowed to enter when idle (that is, QoSPM-requested operational latencies constrain how deep of a C-state to attempt) (already exists)
• WiFi QoSPM enables the transmitter power state to be defined by operational latency and bandwidth expectations, such as VoIP, games, and web browsing (practical)
• Network Latency network packets and interrupt events can be buffered up to reduce power use at the expense of some latency. If the QoS needs don't call for very low latency, the network driver can automatically set these behaviors
• Audio
  • You can use an idle time out to turn off the CODEC/amplifier subject to acceptable wake up latency (practical)
  • You can derive power savings at the audio hardware level from a throughput QoS request
  • The audio subsystem may have a latency constraint that makes certain C-States un-desirable.
• Platform Suspend
  • you can put the system into S3 or S4, as a function of acceptable wake up latency
  • you can specify the delay in dropping into a suspend state as an idle time out
• Hard Disk subject to the acceptable wake up latency, you can request idle time out to specify the delay before the disk spins down. If acceptable wake up latency is too short then spin down is disabled.
• USB you can use an idle time out parameter to define when to start turning off parts of the USB stack. The acceptable wake up latencies constrains how much of the USB stack you can shut off.
• Graphics
  • You could specify performance in terms of bits or pixels/sec. If user doesn't need high pixel rates, turn off hardware acceleration.
  • Turning on the hardware acceleration has a latency that could impact usability. Defining the graphics subsystem's wake up latency and throughput information could allow that subsystem to save power without effecting usability.

QoSPM Applications QoSPM applications are expected to be identified as this infrastructure gets put into place. For now, the initial applications of this concept include: Deep C-state enabling, Gigabit NIC power/latency trade offs, and WiFi power performance tradeoffs.

Get involved If you have additional ideas, scenarios, or applications for the QoSPM concept, join our mailing list and post them.