Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
  • G06F1/26—Power supply means, e.g. regulation thereof
  • G06F1/32—Means for saving power
  • G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
  • G06F1/3206—Monitoring of events, devices or parameters that trigger a change in power modality
  • G06F1/3228—Monitoring task completion, e.g. by use of idle timers, stop commands or wait commands
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
  • G06F1/26—Power supply means, e.g. regulation thereof
  • G06F1/32—Means for saving power
  • G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
  • G06F1/3234—Power saving characterised by the action undertaken
  • G06F1/324—Power saving characterised by the action undertaken by lowering clock frequency
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
  • G06F1/26—Power supply means, e.g. regulation thereof
  • G06F1/32—Means for saving power
  • G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
  • G06F1/3234—Power saving characterised by the action undertaken
  • G06F1/3296—Power saving characterised by the action undertaken by lowering the supply or operating voltage
• • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
  • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
  • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

• Portable computing devices are ubiquitous. These devices may include cellular telephones, portable digital assistants (PDAs), portable game consoles, palmtop computers, and other portable electronic devices. In addition to the primary function of these devices, many include peripheral functions.
• a cellular telephone may include the primary function of making cellular telephone calls and the peripheral functions of a still camera, a video camera, global positioning system (GPS) navigation, web browsing, sending and receiving emails, sending and receiving text messages, push-to-talk capabilities, etc.
• GPS global positioning system
• the computing or processing power required to support such functionality also increases. Further, as the computing power increases, there exists a greater need to effectively manage the processor, or processors, that provide the computing power.
• FIG. 1 is a front plan view of a first aspect of a portable computing device (PCD) in a closed position;
• PCD portable computing device
• FIG. 2 is a front plan view of the first aspect of a PCD in an open position
• FIG. 3 is a block diagram of a second aspect of a PCD;
• FIG. 4 is a block diagram of a processing system;
• FIG. 5 is a flowchart illustrating a first aspect of a method of dynamically controlling power within a CPU
• FIG. 6 is a flowchart illustrating a first portion of a second aspect of a method of dynamically controlling power within a CPU
• FIG. 7 is a flowchart illustrating a second portion of the second aspect of a method of dynamically controlling power within a multicore CPU
• FIG. 8 is an exemplary graph showing the dynamic clock and voltage scaling (DCVS) controlled CPU frequency plotted over time.
• FIG. 9 is an exemplary graph showing effective transient response times for various performance levels.
• FIG. 10 is a block diagram illustrating logical components and information flows in computing device implementing a dynamic clock frequency/voltage scaling (DCVS) solution that enforces performance guarantees in accordance with the various aspects.
• DCVS dynamic clock frequency/voltage scaling
• FIGs. 11 A-B are process flow diagrams illustrating an aspect method of generating a performance guarantee.
• FIGs. 12-13 are process flow diagrams illustrating various aspect methods of enforcing a performance guarantee to ensure a processing core does not remain in a busy state for more than a predetermined amount of time above that which is required for that processing core to complete its pre-computed, predicted, and/or actual steady state workload.
• FIG. 14 is a component block diagram of a mobile device suitable for use in an aspect.
• FIG. 15 is a component block diagram of a server device suitable for use in an aspect.
• FIG. 16 is a component block diagram of a laptop computer device suitable for use in an aspect.
• FIG. 17 is an illustration of an example DCVS solution that sets per core performance guarantees for a thread or workload executing in a multicore system.
• FIG. 18 is an illustration of an example DCVS solution that sets group based performance guarantees for a thread or workload executing in a multicore system.
• FIG. 19 is a process flow diagram illustrating an aspect method of enforcing a performance guarantee to ensure that processors in a multiprocessor system do not remain in a busy state for more than required to complete a pre- computed, predicted, and/or actual steady state workload.
• an “application” may also include files having executable content, such as: object code, scripts, byte code, markup language files, and patches.
• an "application” referred to herein may also include files that are not executable in nature, such as documents that may need to be opened or other data files that need to be accessed.
• content may also include files having executable content, such as: object code, scripts, byte code, markup language files, and patches.
• content referred to herein, may also include files that are not executable in nature, such as documents that may need to be opened or other data files that need to be accessed.
• module “module,” “system,” and the like are intended to refer to a computer-related entity, either hardware, firmware, a combination of hardware and software, software, or software in execution.
• a component may be, but is not limited to being, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and/or a computer.
• an application running on a computing device and the computing device may be a component.
• One or more components may reside within a process and/or thread of execution, and a component may be localized on one computer and/or distributed between two or more computers.
• these components may execute from various computer readable media having various data structures stored thereon.
• the components may communicate by way of local and/or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network such as the Internet with other systems by way of the signal).
• a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network such as the Internet with other systems by way of the signal).
• an exemplary portable computing device is shown and is generally designated 100.
• the PCD 100 may include a housing 102.
• the housing 102 may include an upper housing portion 104 and a lower housing portion 106.
• FIG. 1 shows that the upper housing portion 104 may include a display 108.
• the display 108 may be a touch screen display.
• the upper housing portion 104 may also include a trackball input device 110.
• the upper housing portion 104 may include a power on button 112 and a power off button 114.
• the upper housing portion 104 of the PCD 100 may include a plurality of indicator lights 116 and a speaker 118. Each indicator light 116 may be a light emitting diode (LED).
• LED light emitting diode
• the upper housing portion 104 is movable relative to the lower housing portion 106. Specifically, the upper housing portion 104 may be slidable relative to the lower housing portion 106.
• the lower housing portion 106 may include a multi-button keyboard 120.
• the multi-button keyboard 120 may be a standard QWERTY keyboard. The multi-button keyboard 120 may be revealed when the upper housing portion 104 is moved relative to the lower housing portion 106.
• FIG. 2 further illustrates that the PCD 100 may include a reset button 122 on the lower housing portion 106.
• the PCD 320 includes an on-chip system 322 that includes a multicore CPU 324.
• the multicore CPU 324 may include a zeroth core 325, a first core 326, and an Nth core 327.
• a display controller 328 and a touch screen controller 330 are coupled to the multicore CPU 324.
• a display/touch screen 332 external to the on-chip system 322 is coupled to the display controller 328 and the touch screen controller 330.
• FIG. 3 further indicates that a video encoder 334, e.g., a phase alternating line (PAL) encoder, a sequential 07 a memoire (SECAM) encoder, or a national television system(s) committee (NTSC) encoder, is coupled to the multicore CPU 324.
• a video amplifier 336 is coupled to the video encoder 334 and the display/ touch screen 332.
• a video port 338 is coupled to the video amplifier 336.
• a universal serial bus (USB) controller 340 is coupled to the multicore CPU 324.
• a USB port 342 is coupled to the USB controller 340.
• USB universal serial bus
• a memory 344 and a subscriber identity module (SIM) card 346 may also be coupled to the multicore CPU 324.
• SIM subscriber identity module
• a digital camera 348 may be coupled to the multicore CPU 324.
• the digital camera 348 is a charge-coupled device (CCD) camera or a complementary metal-oxide semiconductor (CMOS) camera.
• a stereo audio CODEC 350 may be coupled to the multicore CPU 324.
• an audio amplifier 352 may coupled to the stereo audio CODEC 350.
• a first stereo speaker 354 and a second stereo speaker 356 are coupled to the audio amplifier 352.
• FIG. 3 shows that a microphone amplifier 358 may be also coupled to the stereo audio CODEC 350.
• a microphone 360 may be coupled to the microphone amplifier 358.
• a frequency modulation (FM) radio tuner 362 may be coupled to the stereo audio CODEC 350.
• an FM antenna 364 is coupled to the FM radio tuner 362.
• headphones 366 may be coupled to the stereo audio CODEC 350.
• FIG. 3 further indicates that a radio frequency (RF) transceiver 368 may be coupled to the multicore CPU 324.
• An RF switch 370 may be coupled to the RF transceiver 368 and an RF antenna 372.
• a keypad 374 may be coupled to the multicore CPU 324.
• a mono headset with a microphone 376 may be coupled to the multicore CPU 324.
• a vibrator device 378 may be coupled to the multicore CPU 324.
• FIG. 3 also shows that a power supply 380 may be coupled to the on-chip system 322.
• the power supply 380 is a direct current (DC) power supply that provides power to the various components of the PCD 320 that require power.
• DC direct current
• the power supply is a rechargeable DC battery or a DC power supply that is derived from an alternating current (AC) to DC transformer that is connected to an AC power source.
• FIG. 3 further indicates that the PCD 320 may also include a network card 388 that may be used to access a data network, e.g., a local area network, a personal area network, or any other network.
• the network card 388 may be a Bluetooth network card, a WiFi network card, a personal area network (PAN) card, a personal area network ultra- low-power technology (PeANUT) network card, or any other network card well known in the art.
• the network card 388 may be incorporated into a chip, i.e., the network card 388 may be a full solution in a chip, and may not be a separate network card 388.
• the display/touch screen 332, the video port 338, the USB port 342, the camera 348, the first stereo speaker 354, the second stereo speaker 356, the microphone 360, the FM antenna 364, the stereo headphones 366, the RF switch 370, the RF antenna 372, the keypad 374, the mono headset 376, the vibrator 378, and the power supply 380 are external to the on-chip system 322.
• one or more of the method steps described herein may be stored in the memory 344 as computer program instructions. These instructions may be executed by the multicore CPU 324 in order to perform the methods described herein. Further, the multicore CPU 324, the memory 344, or a combination thereof may serve as a means for executing one or more of the method steps described herein in order to a dynamically control the power of each CPU, or core, within the multicore CPU 324.
• a processing system is shown and is generally designated 400.
• the processing system 400 may be incorporated into the PCD 320 described above in conjunction with FIG. 3.
• the processing system 400 may include a multicore central processing unit (CPU) 402 and a memory 404 connected to the multicore CPU 402.
• the multicore CPU 402 may include a zeroth core 410, a first core 412, and an Nth core 414.
• the zeroth core 410 may include a zeroth dynamic clock and voltage scaling (DCVS) algorithm 416 executing thereon.
• the first core 412 may include a first DCVS algorithm 417 executing thereon.
• the Nth core 414 may include an Nth DCVS algorithm 418 executing thereon.
• each DCVS algorithm 416, 417, 418 may be independently executed on a respective core 410, 412, 414.
• the memory 404 may include an operating system 420 stored thereon.
• the operating system 420 may include a scheduler 422 and the scheduler 422 may include a first run queue 424, a second run queue 426, and an Nth run queue 428.
• the memory 404 may also include a first application 430, a second application 432, and an Nth application 434 stored thereon.
• the applications 430, 432, 434 may send one or more tasks 436 to the operating system 420 to be processed at the cores 410, 412, 414 within the multicore CPU 402.
• the tasks 436 may be processed, or executed, as single tasks, threads, or a combination thereof.
• the scheduler 422 may schedule the tasks, threads, or a combination thereof for execution within the multicore CPU 402. Additionally, the scheduler 422 may place the tasks, threads, or a combination thereof in the run queues 424, 426, 428.
• the cores 410, 412, 414 may retrieve the tasks, threads, or a combination thereof from the run queues 424, 426, 428 as instructed, e.g., by the operating system 420 for processing, or execution, of those task and threads at the cores 410, 412, 414.
• FIG. 4 also shows that the memory 404 may include a parallelism monitor 440 stored thereon.
• the parallelism monitor 440 may be connected to the operating system 420 and the multicore CPU 402. Specifically, the parallelism monitor 440 may be connected to the scheduler 422 within the operating system 420.
• a first aspect of a method of dynamically controlling the power of a central processing unit is shown and is generally designated 500.
• the method 500 may commence at block 502 with a do loop in which when device is powered on, the following steps may be performed.
• a power controller e.g., a dynamic clock and voltage scaling (DCVS) algorithm, may monitor one or more CPUs.
• the power controller may determine whether a transient performance deadline for a CPU has expired. If not, the method 500 may end. Otherwise, if the transient performance deadline has expired, the method 500 may proceed to block 508 and the power controller may move the CPU to a higher performance level, i.e., a next higher operating frequency. In one aspect, the controller may move the CPU to a maximum performance level, i.e., a maximum CPU frequency.
• DCVS dynamic clock and voltage scaling
• the CPU may not jump to a maximum performance level.
• the CPU may jump to an intermediate level and then, jump again, either to the maximum level or another higher performance level.
• the number of intermediate jumps, and the amount of time between jumps, may be used to determine the frequency value of the jump.
• the CPU may enter an idle condition. Further, at block 512, the transient performance deadline may be reset. At block 514, the CPU may exit the idle condition. Moving to decision 516, the power controller may determine whether the upcoming CPU frequency is at a maximum CPU frequency. If so, the method 500 may end. Otherwise, if the CPU frequency is not at the maximum CPU frequency, the method may proceed to block 518 and the timer may be rescheduled. Then, the method 500 may end.
• a second aspect of a method of dynamically controlling the power of a central processing unit is shown and is generally designated 600.
• a central processing unit CPU
• a power controller e.g., a dynamic clock and voltage scaling (DCVS) algorithm
• DCVS dynamic clock and voltage scaling
• the power controller may determine a busy time (BusyTime) by subtracting a start idle time
• the CPU may enter a software wait for interrupt (SWFI) condition.
• SWFI software wait for interrupt
• the CPU may exit the SWFI condition.
• the power controller may set an end idle time (EndldleTime) equal to a current time (CurrentTime). Further, at block 614, the power controller may determine an idle time (IdleTime) by subtracting the start idle time
• the power controller may determine an upcoming CPU frequency (CPUFreq) from an updated steady state filter (UpdateSteadyStateFilter) a busy time (BusyTime) and an idle time (IdleTime). Thereafter, the method 600 may continue to block 702 of FIG. 7.
• CPUFreq upcoming CPU frequency
• UpdateSteadyStateFilter updated steady state filter
• BusyTime busy time
• IdleTime idle time
• the power controller may determine an effective transient budget (EffectiveTransientBudget) using the following formula:
• EffectiveTransientBudget (TransientResponseDeadline*NextCPUFreq)/(
• TransientResponseDeadline A transient response deadline, i.e., slack budget
• NextCPUFreq A next CPU frequency that is one frequency step higher than an upcoming CPU frequency
• CPUFreq An upcoming CPU frequency (CPUFreq).
• the power controller may set a deadline to jump to a higher frequency (SetJumpToFrequency) equal to the end idle time (EndldleTime) plus the effective transient budget (EffectiveTransientBudget).
• the deadline to jump may be the current time plus the transient budget.
• the method 600 described in conjunction with FIG. 6 and FIG. 7 may be used to calculate the amount of time that the CPU may remain at the frequency determined by the DCVS before the transient deadline is exhausted and schedule a jump to the higher CPU frequency by that amount of time in the future. If idle is reentered prior to the jump to the higher frequency, the scheduled jump may be cancelled.
• the method 600 may delay the jump to the higher frequency by the amount of time determined as the
• PCD portable computing device
• the PCD may be a mobile telephone device, a portable digital assistant device, a smartbook computing device, a netbook computing device, a laptop computing device, a desktop computing device, or a combination thereof.
• a DCVS algorithm is a mechanism which measures CPU load/idle time and dynamically adjusts the CPU clock frequency to track the workload in an effort to reduce power consumption while still providing satisfactory system performance.
• the change in CPU throughput may track, but also necessarily lag, the changes in the workload.
• QoS Quality of Service
• the transient performance guarantee may be defined as the maximum amount of time that a continuously busy pulse may be delayed, as compared to running at the higher performance level. This may be accomplished by getting to the higher performance level prior to the transient performance deadline expiring and resetting the deadline whenever we go idle, since if the CPU is idle, it is by definition not in an oversubscribed state. As disclosed herein, the timer may be rescheduled to preserve the QoS guarantee whenever the system comes out of idle and the system CPU is not running at the maximum frequency.
• the present system and methods minimize the likely hood that an incoming pulse may require a frequency increase in order to meet the deadline. This may be accomplished by delaying the frequency, i.e., performance level, change until the effective transient budget has been exhausted and then, jumping straight to the higher performance level and staying there until the pulse is complete as shown in FIG. 8.
• the effective transient budget is calculated as the transient response deadline scaled to the current performance level. For example, if the CPU is running a 75% of the maximum clock rate and the transient response deadline is 16 ms, the effective transient budget is 64 ms, i.e., 16 ms/(l- 0.75). The effective transient budget represents how long the CPU may run at the current performance level prior to exhausting the budget. If the CPU is idle, the effective transient budget may be the same as the transient response deadline. If we are at the maximum performance level, the effective transient budget is infinite as shown in FIG. 9.
• the system may provide a strict bound on the maximum amount of time a task might run at some level other than the maximum level, and therefore implicitly provide a calculable bound on completion for tasks that require QoS guarantees, while still allowing dynamic CPU clock scaling.
• the bound may be set based on what tasks are currently running, a global system property, DCVS algorithm design or other properties, and may be entirely disabled if the system is not running any tasks that have QoS requirements or if the CPU is running at max clock.
• the present methods may be extended by, instead of jumping to the maximum frequency when the deadline has expired, setting shorter internal effective deadlines and jumping to one, or more, intermediate frequencies, while still ensuring that the CPU is at the maximum frequency before the maximum QoS delay has been exhausted. Further, the present methods may substantially ensure that a well defined transient QoS is
• the system and methods described herein may utilize opportunistic sampling.
• the system and methods may check for timer expiration on a periodic basis.
• the system and methods may not utilize opportunistic sampling.
• the various aspects provide a strict and calculable bound (e.g., a performance guarantee) for the completion for tasks.
• performance guarantees may be implemented as part of a dynamic clock and voltage/frequency scaling (DCVS) solution to improve processor performance and/or reduce power consumption on portable computing devices (PCDs), including mobile devices such as cellular telephones, smartphones, personal or mobile multi-media players, personal data assistants (PDA's), laptop computers, tablet computers, smartbooks, ultrabooks, palm-top computers, wireless electronic mail receivers, multimedia Internet enabled cellular telephones, wireless gaming controllers, and similar personal electronic devices that include a memory, a programmable processor or core (herein collectively "processing core”) and operate under battery power such that power conservation methods are of benefit.
• processing core programmable processor or core
• the aspects are particularly useful for portable and mobile computing devices that run on battery power, the aspects are generally useful in any computing device (e.g., general-purpose computers, desktop computers, servers, etc.) that includes a processor and for which reduced power consumption is beneficial.
• the dynamic power (switching power) dissipated by a chip is C*V 2 * , where C is the capacitance being switched per clock cycle, V is voltage, and/is the switching frequency.
• C the capacitance being switched per clock cycle
• V voltage
• Dynamic power may account for approximately two-thirds of the total power consumed by a processor chip.
• Voltage scaling may be accomplished in conjunction with frequency scaling, because the frequency at which a chip runs may be related to its operating voltage.
• the efficiency of some electrical components, such as voltage regulators may decrease with increasing temperature so that the power consumption increases with temperature. Since increasing power use may increase the temperature, increases in voltage or frequency may increase system power demands even further.
• the battery life of a computing device may be improved by reducing the frequency and/or voltage applied to its processors when they are idle or lightly loaded. Such reductions in frequency and/or voltage may be accomplished in real time or "on the fly” via a dynamic clock and voltage/frequency scaling (DCVS) solution.
• DCVS dynamic clock and voltage/frequency scaling
• a DCVS solution monitors the proportion of the time that a processor is idle (compared to the time it is busy), and determines how much the frequency/voltage of the processors should be adjusted based on the proportion of the time the processors is idle and/or busy.
• Monitoring the proportion of the time that a processor is idle may include computing and/or measuring a value (e.g., amount of time, number of CPU cycles, etc.) indicative of the duration that the processor executes an idle process or thread (e.g., a system idle process, etc.).
• An operating system may execute an idle software application, process, or thread (herein collectively "thread") on the processor when it determines that there are no other threads ready to be scheduled on that processor.
• the idle thread may perform various tasks (e.g., wait for interrupt task, sleep task, etc.), and each task may include a number of processor operations.
• a processor executes an idle thread that processor may be said to be "idle” in an "idle state,” and/or in an "idle condition.”
• an operating system may maintain one or more idle threads for each processor. Idle threads remain ready for execution so that each processor always has a thread ready to execute. In this manner, whenever a thread relinquishes a processor (e.g., due to the thread completing its scheduled tasks or workloads), the operating system has a thread ready for execution on that processor (e.g., via the availability of the idle thread), even when all other threads are complete, waiting for resources, or otherwise not currently ready to execute.
• a DCVS solution may adjust a processor's frequency and/or voltage based on that processor's workloads, which may include a steady state workload.
• a steady state workload may be determined in advance of execution time, that is in advance of the processing core entering a busy or active state to perform operations in furtherance of the workload.
• the steady state workload may be determined in advance by computing, estimating, or predicting the number of CPU clock cycles, number of operations, number of instructions, and/or amount of time required to complete the tasks scheduled on that processing core.
• Each processor may have more than one workload (e.g., a steady state workload and a transient workload), and each processor may be required to remain in a busy, running, or active state (herein collectively “busy state”) until all the tasks in all its workloads are complete.
• a steady state workload e.g., a steady state workload and a transient workload
• bus state e.g., a busy, running, or active state
• a DCVS solution may reduce a processor's frequency and/or voltage (i.e., the processor's speed) to achieve power savings without impacting the processor's performance. For example, when the processor's workload includes a task whose execution time is dominated by memory access times, a reduction in frequency may not have a significant impact on that processor's performance or that task's execution time. However, more often, the DCVS solution must balance tradeoffs between the processor's performance (e.g., time required to complete a given set of tasks, etc.) and power consumption (e.g., amount of battery power consumed in accomplishing the given set of tasks) characteristics. Typically, the faster the tasks are
• a DCVS solution may be configured to balance the performance and power consumption based on a processor' s steady state workload and steady state performance requirements.
• the steady state performance requirements may be determined by computing or measuring a value (e.g., amount of time, number of CPU cycles, etc.) indicative of a duration in which the processor is busy and/or idle, averaging the results of the computed/measured values, and determining the amount of time/processing that is required to complete that processor' s steady state workloads.
• the DCVS solution may compute an upper frequency threshold and a lower frequency threshold within which the processor may operate to meet the computed steady state requirements while achieving reduced power consumption and acceptable levels of responsiveness (e.g., such that a mobile device user does not notice a difference, etc.).
• transient workloads may be any task or unit of work that is not known to the system in advance, including any unit of work that is dynamic, temporary, or which causes an unexpected spike in the processor's workload.
• a transient workload may include any or all of tasks performed by a processor in response to user inputs, system events, detected environmental conditions, remote procedure calls, etc.
• a transient workload may be generated when a user touches a touchscreen of a portable computing device (PCD) to initiate a user action, to which the PCD must respond immediately (e.g., with an interface update, by displaying a new image, launching a new action, etc.).
• PCD portable computing device
• transient workloads are not continuous steady state workloads for which the DCVS solution can properly account for in advance (e.g., as part of determining the upper and lower thresholds).
• a transient workload may cause a processor to remain in a busy state for longer than expected and/or otherwise result in uncertainties in processor execution times.
• uncertainties may cause the computing device to allocate processing and system resources inefficiently or improperly, and may have a significant impact on the overall performance and/or responsiveness of a computing device, especially when the computing device includes multiple processing cores.
• Modern computing devices are often multiprocessor systems that include system-on-chips (SoCs) and/or multiple processing cores (e.g., processors, cores, etc.).
• SoCs system-on-chips
• processing cores e.g., processors, cores, etc.
• results of one thread in a first processing core it is also common for the results of one thread in a first processing core to trigger operations in another thread executing in a second processing core.
• one or more processing cores may be dependent on the results generated by a currently active processor, and may be required to remain in an idle or wait state until the currently active processor completes its workloads and/or finishes processing one or more tasks.
• each processing core may alternatively enter an idle/wait state while it awaits the results of processing from the currently active processor. While these processing cores wait for the results generated by the currently active processor, their respective DCVS solutions may reduce their operating speeds (i.e., via a reduction in frequency/voltage), causing the computing device to appear non- responsive or slow. That is, a DCVS solution implemented on multiprocessor computing devices may incorrectly conclude that the some of the processing cores should be operated at lower frequency or voltage than is optimal for running the currently active threads, and cause the computing device to appear non-responsive or slow.
• the various aspects overcome the above mentioned limitations by computing and enforcing performance guarantees that ensure a processing core does not remain in a busy state (e.g., due to transient workloads) for more than a predetermined amount of time above which is required for that processing core to complete its pre-computed steady state workload.
• Such performance guarantees may be used by the operating system, resources, DCVS solutions, and/or other processing cores to better estimate, schedule, and/or plan future operations, such as allocating resources and scheduling threads for execution. In this manner, the performance guarantees enable the computing device to meet its responsiveness requirements, and thus improve the user experience.
• Performance guarantees allow a DCVS solution to adjust the frequency and/or voltage of a processor based on a variable delay, which ensures that the processing core only falls behind its steady state workload by, at most, a defined maximum amount of work, irrespective of the current or previous operating frequencies/voltages of the processor.
• FIG. 10 illustrates logical components and information flows in an aspect computing device 1000 implementing a dynamic clock frequency/voltage scaling (DCVS) solution that enforces performance guarantees.
• the computing device 1000 may include a hardware unit 1002, a kernel space software unit 1004, and a user space software unit 1006.
• the kernel space software unit 1004 and the user space software unit 1006 may be included in an operating system or kernel of the computing device 1000.
• the computing device may include a kernel that is organized into user space (where non-privileged code runs) and kernel space (where privileged code runs). This separation is of particular importance in Android and other general public license (GPL) environments where code that is part of the kernel space must be GPL licensed, while code running in user-space doesn't need to be GPL licensed.
• GPL general public license
• the hardware unit 1002 may include a number of processing cores (e.g., CPU 0, CPU 1, 2D-GPU 0, 2D-GPU 1, 3D-GPU 0, etc.) and a resources module 1020 that includes various hardware resources (e.g., clocks, power management integrated circuits or "PMICs,” scratchpad memories or "SPMs,” etc.) shared by the processing cores.
• processing cores e.g., CPU 0, CPU 1, 2D-GPU 0, 2D-GPU 1, 3D-GPU 0, etc.
• resources module 1020 that includes various hardware resources (e.g., clocks, power management integrated circuits or "PMICs,” scratchpad memories or "SPMs,” etc.) shared by the processing cores.
• PMICs power management integrated circuits
• SPMs scratchpad memories
• the kernel space software unit 1004 may include processor modules (CPU_0 Idle stats, CPU_1 idle stats, 2D-GPU_0 driver, 2D-GPU_1 driver, 3D- GPU_0 driver, etc.) that correspond to at least one of the processing cores in the hardware unit 1002, each of which may communicate with one or more idle stats device modules 1008.
• the kernel space software unit 1004 may further include a timer driver module 1014, input event modules 1010, and a CPU request stats module 1012. In an aspect, the timer driver module 1014 may drive (or maintain) a timer for each processing core.
• the user space software unit 1006 may include a DCVS control module 416 configured to receive inputs from each of the idle stats device modules 1008, input event modules 1010, timer driver module 1014, and CPU request stats module 1012, and/or to send outputs to a CPU frequency hot-plug module 1018.
• the CPU frequency hot-plug module 1018 may be configured to send
• the CPU frequency hot- plug module 1018 may be further configured to apply voltage/frequency changes to each core, either individually (e.g., one at a time, sequentially, etc.) or simultaneously (e.g., at approximately the same point in time).
• the DCVS control module 1016 may include threads suitable for execution on any or all of the processing cores (e.g., CPU 0, CPU 1, 2D-GPU 0, 2D-GPU 1, 3D-GPU 0, etc.) and/or suitable for implementing a DCVS solution on the computing device 1000.
• the DCVS control module 1016 may include a thread that monitors a port or a socket for an occurrence of an event (e.g., filling of a data buffer, expiration of a timer, state transition, etc.) that causes the DCVS control module 1016 to collect information from one or more processing cores and perform DCVS operations on a processing core.
• the DCVS control module 1016 may include a single-threaded DCVS solution that monitors two or more processing cores.
• the DCVS control module 1016 may include a DCVS solution thread for each processing core.
• the DCVS control module 1016 may be configured to generate pulse trains.
• the DCVS control module 1016 may generate the pulse trains by monitoring or sampling the busy and/or idle states (or the transitions between states) of the processing cores.
• the DCVS control module 1016 may also generate the pulse trains based on information obtained from monitoring the depth of one or more processor run-queues.
• a run-queue may include a running thread as well as a collection of one or more threads that are capable of running on a processing core, but not yet able to do so (e.g., due to another active thread that is currently running, etc.).
• Each processing core may have its own run- queue, or a single run-queue may be shared my multiple processing cores.
• Threads may be removed from the run queue when they request to enter a sleep state, are waiting on a resource to become available, or have been terminated.
• the number of threads in the run queue i.e., the run queue depth
• the DCVS control module 1016 may be configured to compute steady state workloads, steady state requirements, and/or upper and lower frequency/voltage thresholds based on the generated pulse trains.
• the upper and lower frequency/voltage thresholds may define a frequency/voltage rage within which a processing core may operate to meet the its steady state performance requirements while simultaneously achieving reduced power consumption and meeting the responsiveness requirements of the computing device 1000.
• Meeting the responsiveness requirements may include performing all the tasks in the workloads so that a user of the computing device 1000 does not notice a reduction in the computing device's performance or speed.
• the DCVS control module 1016 may be configured to monitor the overall computing device 1000 performance and/or insure that one or more of the processing cores operate between established upper and lower frequency thresholds.
• the DCVS control module 1016 may adjust processing resources and/or the operating frequencies of the processing cores so that they are commensurate with the thresholds.
• the DCVS control module 1016 may generate pulse trains.
• pulse trains generated for two or more of the processing cores may be synchronized in time and cross-correlated to generate correlation models that include information suitable for determining whether the processing cores are performing operations that are co-operative and/or dependent on one another.
• the DCVS control module 1016 may use the correlation models to determine the upper and lower frequency thresholds, initial operating frequency, steady state requirements, and processor workloads so that these value account for the interdependencies between the processing cores.
• the DCVS control module 1016 may be configured to compute and/or enforce performance guarantees.
• processing cores may be required to process/execute transient workloads that the DCVS solution cannot properly account for in advance.
• transient workloads may cause the DCVS control module 1016 to operate one or more of the processing cores at a sub-optimal frequency level or within a sub-optimal frequency range.
• the DCVS control module 1016 cannot account for the transient workloads in advance, it may improperly conclude that the processing core may be operated at a lower frequency level than that which is required to complete both the steady state workload and the transient workload in a time period that suitable for meeting the responsiveness requirements of the computing device 1000.
• a performance guarantee provides the computing device 1000 with a strict and calculable bound that may be used by the DCVS control module 1016 to ensure a processing core does not remain in a busy state (e.g., due to transient workloads) for more than a predetermined amount of time/work above that which is required for that processing core to complete its steady state workload requirements.
• a performance guarantee allows the DCVS control module 1016 to ensure that a processing core completes both its steady state workload and its transient workloads in a time period that suitable for meeting the responsiveness requirements of the computing device 1000.
• a performance guarantee may be computed in, defined in, and/or include any unit of measurement suitable for measuring processor performance or a duration, such as an amount of time, an amount of work, a number of tasks, a number of instructions, a number of CPU cycles, etc.
• a performance guarantee may be associated with a frequency and/or may be a function of frequency.
• a performance guarantee may include one or more performance guarantee values.
• the performance guarantee values (e.g., a deadline value, a budget value, a jump-to-max value, etc.) may be expressed in any unit of measurement suitable for measuring processor performance or a duration, such as an amount of time, an amount of work, a number of tasks, a number of instructions, a number of CPU cycles, etc.
• the performance guarantee values may include a budget value (e.g., slack budget, transient budget, etc.), a deadline value (e.g., transient deadline, transient response deadline, performance deadline, etc.), and/or a jump-to-max value.
• the deadline value may be a value indicative of a relative time before which the processing core is to complete its workload processing and/or a value indicative of a relative time after which the frequency of a processing core is to be increased.
• the budget value may be a value indicative of an amount of time remaining before which the processing core is to complete its workload processing and/or after which the frequency of a processing core is to be increased.
• the jump-to-max value may be a value indicative of a relative time before which the processing core is to complete its workload processing and/or after which the frequency of a processing core is to be increased to the maximum frequency.
• the performance guarantee values may be a related to, associated with, and/or function of frequency or voltage.
• each of the budget, deadline, and/or jump-to values may be a time value that is computed as a function of an operating frequency of a corresponding processing core.
• each of these values may belO milliseconds when the processing core operates at a frequency of 100 MHz, 20 milliseconds at a frequency of 200 MHz, 40 milliseconds at a frequency of 400 MHz, etc.
• the performance guarantee values may used by the DCVS solution to implement a variable delay for increasing the frequency of a processing core.
• the DCVS solution may implement variable delays. Such variable delays ensure that the processing core only falls behind its steady state workload by, at most, a defined maximum amount of work, irrespective of the actual operating frequencies of the processing core.
• the DCVS solution may set the defined maximum amount of work (i.e., by which the processing core may fall behind its steady state workload) to be equal to the deadline value multiplied by the maximum frequency/voltage of the processing core. In this manner, the performance guarantee is not affected by the DCVS solution adjusting the frequency/voltage of the processing core based on the steady state requirements or dynamically or "on the fly.”
• the DCVS control module 1016 may be configured to set the deadline value to be equal to the budget value each time the corresponding processing core transitions from idle to busy, enters a busy state (e.g., beings processing the workload, etc.) and/or exits the idle state (e.g., when an idle thread relinquishes the processing core, etc.).
• the DCVS control module 1016 may be configured to set or reset an existing deadline value each time the corresponding processing core transitions from busy to idle, enters an idle state (e.g., beings executing an idle thread, etc.) and/or exits the busy state (e.g., completes all tasks in the workload, etc.).
• FIGs. 11A-B illustrate an aspect DCVS solution method 1100 of generating/computing a performance guarantee that ensures a processing core does not remain in a busy state for more than a predetermined amount of time above that which is required for that processing core to complete its pre- computed, predicted, and/or actual steady state workload (e.g., due to transient workloads, etc.).
• the operations of the DCVS solution may be performed by a thread executing on the processing core or on another processing core.
• one or more of the operations of the DCVS solution may be performed by an idle thread executing on the processing core.
• a DCVS solution may cause a processing core to transition from an idle state to a busy state.
• the DCVS solution may set the value of an end idle time parameter (EndldleTime) to be equal to a current time value (CurrentTime).
• the end idle time parameter (EndldleTime) may store a value that is indicative of the time at which the processing core last exited the idle state.
• the operations of blocks 1102 and 1104 may be performed sequentially, in parallel, and/or in any order.
• the DCVS solution may set the value of an end idle time parameter (EndldleTime) before transitioning the processing core from idle to busy.
• the DCVS solution may set the value of an end idle time parameter (EndldleTime) after transitioning the processing core from idle to busy.
• the DCVS solution may monitor the operating frequency or voltage of the processing core and make adjustments as necessary.
• the DCVS solution may cause the processing core to transition from the busy state to the idle state.
• the DCVS solution may transition the processing core to the idle state by commencing execution of an idle thread on the processing core.
• the DCVS solution may transition the processing core to the idle state after the processing core completes all the tasks associated with all its workloads.
• the DCVS solution may set or reset an existing deadline value.
• a deadline value may be a performance guarantee value that is included in, or associated with, a performance guarantee. Additional details regarding the operations of setting, reset, and/or computing the deadline value are provided further below.
• the DCVS solution may set the value of a start idle time parameter (StartldleTime) to be equal to a current time value (CurrentTime).
• the DCVS solution may set the value of a busy time parameter (BusyTime) to be equal to the difference between the time when the processing core last exited a previous idle state (which may be represented by the end idle time parameter "EndldleTime”) and the time when the processing core entered the current idle state (which may be represented by the start idle time parameter "StartldleTime”)-
• the busy time parameter (BusyTime) may store a value that is indicative of a duration in which the processing core most recently remained in the busy state.
• the DCVS solution may cause the processing core to perform various idle state operations, such as sleep operations, deep sleep operations, or software wait for interrupt operations. Accordingly, in block 1116, the DCVS solution (e.g., via the idle thread, operating system, etc.) may cause the processing core to enter a sleep state, a deep sleep state, a wait for interrupt state, etc.
• the DCVS solution and/or idle thread may receive an interrupt request and/or otherwise determine that the processing core is to be transitioned from its current state to a busy state. This may be achieved by DCVS solution receiving a notification (e.g., from an operating system scheduler, controller, etc.) that tasks have been scheduled for execution on the processing core and/or scheduled tasks are ready for execution.
• a notification e.g., from an operating system scheduler, controller, etc.
• the DCVS solution may set the value of the end idle time parameter (EndldleTime) to be equal to a current time value (CurrentTime).
• the DCVS solution may set the value of an idle time parameter (IdleTime) to be equal to a difference between the value of the StartldleTime parameter and the value of the EndldleTime parameter.
• the idle time parameter (IdleTime) may store a value indicative of the duration in which the processing core last remained the idle state.
• the DCVS solution may compute an operating frequency, frequency range, and/or frequency thresholds within which the processing core is to operate.
• the DCVS solution may compute the frequency or frequency range based on the duration of time the processing core last remained in a busy state (e.g., BusyTime) and/or the duration of time the processing core last remained in the idle state (e.g., IdleTime).
• the DCVS solution may compute the operating frequency, frequency range, and/or frequency thresholds based on historical information, such as an average (or moving average) of the durations in which the processor previously remained in the busy and/or idle states (e.g., over a predetermined time period or time window).
• the DCVS solution may compute the operating frequency, frequency range, and/or frequency thresholds based on pulse trains.
• pulse trains may be generated based on a sampling of the busy and/or idle states, transitions between states, depth of run-queues, etc.
• the DCVS solution may compute or select a deadline value.
• the deadline value may be a value that is indicative of a relative time after which the frequency of the processing core is to be set to be increased, either to a next higher frequency step or to the maximum frequency.
• the deadline value may computed based on configuration settings, driver inputs, the quantity and/or types of scheduled tasks, the predicted steady state workload, and/or the responsiveness requirements of the computing device.
• the deadline value may be determined based on static and/or dynamic values. For example, the deadline value may be determined based on a static configuration value, or based on the types of tasks (e.g., streaming 1080p video vs. streaming 720p video, etc.) scheduled to be performed on the processing core.
• the deadline value may be inversely proportional to the responsiveness requirements of the computing device (i.e., the higher the responsiveness requirements, the shorter the deadline).
• the deadline value may be a time value that is a function of a current operating frequency of the processing core (e.g., 10 milliseconds at a frequency of 100 MHz, 20 milliseconds at a frequency of 200 MHz, 40 milliseconds at a frequency of 400 MHz, etc.).
• the DCVS solution may compute or select a budget value.
• the budget value may be a value that is indicative of an amount of time in which the processing core may remain in the active or busy state without exceeding the sum of the deadline value and the time determined to be required for the processing core to complete its steady state workload requirements.
• the budget value may be a time value that is a function of a current operating frequency of the processing core (e.g., 10 milliseconds at a frequency of 100 MHz, 20 milliseconds at a frequency of 200 MHz, 40 milliseconds at a frequency of 400 MHz, etc.).
• the budget value may computed based on the deadline value, a plurality of frequency levels or steps, the maximum processor frequency, a steady state processor frequency, etc.
• the budget value may an effective transient budget and/or computed via any of the formulas discussed above.
• the DCVS solution may compute a jump to max value.
• the jump-to-max value may be a value that is indicative of a relative time after which the frequency of the processing core is to be set the maximum processing frequency.
• the jump-to-max value may be computed by summing the value of the EndldleTime parameter and the budge value.
• the DCVS solution may transition the processing core from the idle state to a busy state.
• the DCVS solution may set the deadline value to be equal to the budget value.
• the DCVS solution may be configured to set the deadline value to be equal to the budget value each time the processing core transitions from idle to busy, enters an active or busy state (e.g., beings processing the workload, etc.) and/or exits an idle state (e.g., when an idle thread relinquishes the processing core, etc.).
• FIG. 12 illustrates an aspect DCVS solution method 1200 of enforcing a performance guarantee to ensure a processing core does not remain in a busy state for more than a predetermined amount of time above that which is required for that processing core to complete its pre-computed, predicted, and/or actual steady state workload (e.g., due to the presence of transient workloads, etc.).
• a DCVS solution may compute the predicted steady state workload based on the scheduled tasks.
• the DCVS solution may compute various performance requirements for the processing core, such as frequency thresholds to meet the power consumption and/or responsiveness requirements of the computing device.
• the performance requirements e.g., frequency
• thresholds may be determined based on the steady state workload, historical information (e.g., amount of time previously spent in a busy state, etc.), processor characteristics, responsiveness requirements, etc.
• the DCVS solution may compute and set the initial operating frequency and/or various performance guarantee values (e.g., deadline value, budget value, jump-to-max value, etc.).
• the DCVS solution may compute the amount of time or work (e.g., CPU cycles, instructions, etc.) required for the processing core to complete all the tasks in the steady state workload while meeting the various performance requirements.
• the DCVS solution may transition the processing core from an idle state to a busy state so that the processing core runs at the computed initial operating frequency/voltage (or within the computed thresholds) and/or meets the various device or system requirements.
• the DCVS solution may monitor the actual workload and/or operating frequency of the processing core and adjust the frequency/voltage as necessary (e.g., in
• the DCVS solution may update the performance guarantee values based on the current operating frequency/voltage of the processing core.
• the DCVS solution may determine whether the processing core has remained in the busy state longer than the computed time/work (i.e., the amount of time/work determined to be required for the processing core to complete all the tasks in the predicted steady state workload).
• the DCVS solution may continue to monitor the actual workload/frequency and make adjustments when necessary.
• the DCVS solution may determine whether the budget has been exhausted.
• the DCVS solution may determine that the budget has been exhausted when the budget value equals zero and/or when the processing core has remained in the busy state for a duration (measured in either time or work) that is greater than or equal to the deadline value plus the computed time/work.
• the DCVS solution may continue to monitor the actual workload/frequency and make adjustments as necessary.
• the DCVS solution may increase the operating frequency/voltage of the processing core.
• the DCVS solution may increase operating frequency/voltage of the processing core to the maximum processor frequency in block 1220.
• the DCVS solution may increase operating frequency/voltage thresholds in block 1220.
• the DCVS solution may increase operating frequency/voltage of the processing core in steps.
• FIG. 13 illustrates another aspect DCVS solution method 1300 of enforcing a performance guarantee.
• the DCVS solution may perform operations that are the same as, or similar to, the operations discussed above with respect to blocks 1202-1214 of FIG. 12.
• the DCVS solution may determine whether there is a high probably that the processing core will complete its current workload before the computed time (i.e., the amount of time determined to be required for the processing core to complete all the tasks in the predicted steady state workload) plus the deadline value (Computed Time + Deadline).
• the DCVS solution may continue to monitor the actual workload/frequency and make adjustments to the operating frequency/voltage as necessary.
• the DCVS solution may increase the operating frequency/voltage of the processing core.
• the operating frequency/voltage of the processing core may be increased to the maximum processor frequency or in steps.
• Various aspects include methods of improving performance on a computing device having a processor, which may include determining a steady state workload of the processor, determining an amount of work required to perform the determined steady state workload on the processor, computing a performance guarantee value for the processor, transitioning the processor from an idle state to a busy state, performing dynamic clock and voltage scaling operations to scale a frequency of the processor based on an actual workload of the processor, updating the performance guarantee value based on the scaled frequency, determining whether the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the processor to a maximum processor frequency.
• increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the processor in steps.
• the method may further include repeatedly performing the operations of updating the performance guarantee value based on the scaled frequency, determining whether the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• computing a performance guarantee value for the processor may include computing a deadline value.
• computing a performance guarantee value for the processor may include computing a budget value.
• determining a steady state workload of the processor may include determining requirements of tasks scheduled to execute on the processor.
• the method may include generating pulse trains by sampling transitions between the busy and idle states.
• the operations of determining a steady state workload of the processor, determining an amount of work required to perform steady state workload on the processor, and computing a performance guarantee value for the processor are performed by a single thread.
• the single thread executes on the processor.
• the single thread executes on a second processor of the computing device.
• means for increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include means for increasing the frequency of the processor to a maximum processor frequency.
• means for increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include means for increasing the frequency of the processor in steps.
• the computing device may further include means for repeatedly performing the operations of updating the performance guarantee value based on the scaled frequency, determining whether the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the processor when it is determined that the processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• means for computing a performance guarantee value for the processor may include means for computing a deadline value.
• means for computing a performance guarantee value for the processor may include means for computing a budget value.
• means for determining a steady state workload of the processor may include means for determining requirements of tasks scheduled to execute on the processor.
• the computing device may include means for generating pulse trains by sampling transitions between the busy and idle states.
• the computing device may include means for accomplishing the operations of determining a steady state workload of the processor, determining an amount of work required to perform steady state workload on the processor, and computing a performance guarantee value for the processor via a single thread.
• the computing device may include means for executing the single thread on the processor.
• the computing device may include means for executing the single thread on a second processor of the computing device.
• FIG. 1 Further aspects include a computing device that may include a processor configured with processor-executable instructions to perform operations that may include a first processor configured with processor-executable instructions to perform operations including determining a steady state workload of a second processor, determining an amount of work required to perform the determined steady state workload on the second processor, computing a performance guarantee value for the second processor, transitioning the second processor from an idle state to a busy state, performing dynamic clock and voltage scaling operations to scale a frequency of the second processor based on an actual workload of the second processor, updating the performance guarantee value based on the scaled frequency, determining whether the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• the first processor may be configured with processor- executable instructions such that increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the second processor to a maximum processor frequency.
• the first processor may be configured with processor- executable instructions such that increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the second processor in steps.
• the first processor may be further configured with processor-executable instructions to repeatedly perform the operations of updating the performance guarantee value based on the scaled frequency, determining whether the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• the first processor may be configured with processor-executable instructions such that computing a performance guarantee value for the second processor may include computing a deadline value.
• the first processor may be configured with processor- executable instructions such that computing a performance guarantee value for the second processor may include computing a budget value.
• the first processor may be configured with processor-executable instructions such that determining a steady state workload of the second processor may include determining requirements of tasks scheduled to execute on the second processor.
• the first processor may be configured with processor- executable instructions to perform operations further including generating pulse trains by sampling transitions between the busy and idle states.
• the first processor may be configured with processor- executable instructions such that the operations of determining a steady state workload of the second processor, determining an amount of work required to perform steady state workload on the second processor, and computing a performance guarantee value for the second processor are performed by a single thread.
• the first processor may be configured with processor- executable instructions such that the single thread may be executed on the first processor.
• the first processor may be configured with processor-executable instructions such that accomplishing the operations of determining a steady state workload of the second processor, determining an amount of work required to perform steady state workload on the second processor, and computing a performance guarantee value for the second processor may include executing the single tread on the second processor.
• FIG. 1 Further aspects include a non-transitory server-readable storage medium having stored thereon processor-executable instructions configured cause a computing device to perform operations that may include determining a steady state workload of a second processor, determining an amount of work required to perform the determined steady state workload on the second processor, computing a performance guarantee value for the second processor, transitioning the second processor from an idle state to a busy state, performing dynamic clock and voltage scaling operations to scale a frequency of the second processor based on an actual workload of the second processor, updating the performance guarantee value based on the scaled frequency, determining whether the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the second processor to a maximum processor frequency.
• the stored processor-executable software instructions may be configured to cause a second processor to perform operations such that increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value may include increasing the frequency of the second processor in steps, and repeatedly performing the operations of updating the performance guarantee value based on the scaled frequency, determining whether the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of the second processor when it is determined that the second processor has remained in the busy state for a period greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that computing a performance guarantee value for the second processor may include computing a deadline value.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that computing a performance guarantee value for the second processor may include computing a budget value.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that determining a steady state workload of the second processor may include determining requirements of tasks scheduled to execute on the second processor.
• the stored processor- executable software instructions may be configured to cause a processor to perform operations including generating pulse trains by sampling transitions between the busy and idle states.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that the operations of determining a steady state workload of the second processor, determining an amount of work required to perform steady state workload on the second processor, and computing a performance guarantee value for the second processor are performed by a single thread.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that the single thread executes on the processor.
• the stored processor-executable software instructions may be configured to cause a processor to perform operations such that the single thread executes on the second processor.
• the various aspects may be implemented in laptops and other mobile devices in which performance, power consumption, and/or responsiveness are important.
• the various aspects may be implemented in server and personal computers to decrease energy and cooling costs for lightly loaded machines. Reducing the heat output allows the system cooling fans to be throttled down or turned off, reducing noise levels, and further decreasing power consumption.
• the various aspects may also be used for reducing heat in insufficiently cooled systems when the temperature reaches a certain threshold.
• the various aspects may be implemented in a variety of portable or mobile computing devices, an example of which is illustrated in FIG. 14.
• the portable computing device 1400 may include a processing core 1401 coupled to memory 1402 and to a transceiver 1405.
• the transceiver 1405 may be coupled to an antenna 1404 for sending and receiving electromagnetic radiation.
• the portable computing device 1400 may also include a display 1403 (e.g., touch screen display) and menu selection buttons or rocker switches 1406 for receiving user inputs.
• multiple processors 1401 may be provided, such as one processor dedicated to wireless communication functions and one processor dedicated to running other applications.
• server 1500 typically includes a processing core 1501, and may include multiple processor systems 1511, 1521, 1531, one or more of which may be or include multi-core processors.
• the processing core 1501 may be coupled to volatile memory 1502 and a large capacity nonvolatile memory, such as a disk drive 1503.
• the server 1500 may also include a floppy disc drive, compact disc (CD) or DVD disc drive 1506 coupled to the processing core 1501.
• the server 1500 may also include a floppy disc drive, compact disc (CD) or DVD disc drive 1506 coupled to the processing core 1501.
• CD compact disc
• 1500 may also include network access ports 1504 coupled to the processing core
• a network 1501 for establishing data connections with a network 1505, such as a local area network coupled to other broadcast system computers and servers.
• a laptop computer 1600 may include a processing core 1601 coupled to volatile memory 1602 and a large capacity nonvolatile memory, such as a disk drive 1604 of Flash memory.
• the computer 1600 may also include a floppy disc drive 1606 and a compact disc (CD) drive 1608 coupled to the processing core 1601.
• the computer device 1600 may also include a number of connector ports coupled to the processing core 1601 for establishing data connections or receiving external memory devices, such as a USB, Fire Wire®, or Lightning ® connector sockets, or other network connection circuits for coupling the processing core 1601 to a network or computer.
• the computer housing includes the touchpad 1616, the keyboard 1618, and the display 1620 all coupled to the processing core 1601.
• Other configurations of computing device may include a computer mouse or trackball coupled to the processor (e.g., via a USB input) as are well known.
• the processing core 1401, 1501, 1601 may be any programmable processor, microprocessor, microcomputer, multi-core processor, or multiple- processor chip that can be configured by software instructions (applications) to perform a variety of functions, including the functions and operations of the various aspects described herein.
• software applications may be stored in the internal memory 1402, 1502, 1602 before they are accessed and loaded into the processing core 1401, 1501, 1601.
• Each processing core 1401, 1501, 1601 may include internal memory sufficient to store the application software instructions.
• additional memory chips e.g., a Secure Data (SD) card
• SD Secure Data
• the internal memory 1402, 1502, 1602 may be a volatile or nonvolatile memory, such as flash memory, or a mixture of both.
• a general reference to memory refers to all memory accessible by the processing core 1401, 1501, 1601, including internal memory 1402, removable memory plugged into the mobile device, and memory within the processing core 1401.
• the processing core 1401, 1501, 1610 may include internal memory sufficient to store the application software instructions.
• the internal memory may be a volatile or nonvolatile memory, such as flash memory, or a mixture of both.
• a general reference to memory refers to memory accessible by the processor 1401, 1501, 1610 including internal memory or removable memory plugged into the device and memory within the processing core 1401, 1501, 1610 itself.
• Performance guarantees designed for a single processing CPU typically do not account for thread migration across multiple CPUs. As such, when the operating system scheduler migrates the thread from one CPU to another, the transient timer may be restarted on the next CPU, which may cause an
• the DCVS solution may fail to account for the operating system scheduler moving threads from one core to another, in which case a transient deadline value may not be reached/hit when intended. This may cause the system to not elevate the performance level (e.g., CPU frequencies, frequency thresholds, etc.) properly or sufficiently, which may cause the system to fail to adequately meet its performance requirements (e.g., power consumption and/or responsiveness requirements of the computing device, etc.).
• the performance level e.g., CPU frequencies, frequency thresholds, etc.
• the various aspects may include systems, devices, and methods of using transient deadlines within groups of cores so that when the scheduler moves threads among cores within the group, the triggering of the transient deadline is not affected.
• the transient deadline is still exceeded and the performance level of all the CPUs in the group will be elevated in order to meet the performance requirements despite the workload being moved from a first processing core (e.g., CPU0) to a second processing core (e.g., CPUl).
• a first processing core e.g., CPU0
• a second processing core e.g., CPUl
• FIG. 17 illustrates an example DCVS solution method 1700 of setting per core performance guarantees (e.g., transient deadlines) for a thread or workload 1702 executing in a multicore system.
• the workload 1702 is assigned a 20 ms transient deadline 1704 on a first processing core (CPUO) when it begins executing in the first processing core (CPUO). This may be accomplished by setting a transient timer associated with the first processing core (CPUO) to expire 20 ms after the workload 1702 starts execution, or after 20ms of total execution time.
• the operating system scheduler migrates the workload 1702 from the first processing core (CPUO) to the second processing core (CPUl), and the workload 1702 is assigned a new 20 ms transient deadline 1706 on the second processing core (CPUl). This may be accomplished by setting a transient timer associated with the second processing core (CPUl) to expire 20 ms after the workload 1702 is transferred to the second processing core (CPUl), or after 30ms of total execution time.
• the DCVS solution illustrated in FIG. 17 sets performance guarantees (e.g., transient deadlines) individually for each core (i.e., the solution uses per core performance guarantees), the workload 1702 does not reach or hit the transient deadline after 20ms of total execution time as may have been originally intended for the workload 1702. As a result, the DCVS solution may not elevate the performance level (e.g., CPU frequency) of the processing cores (e.g., CPUl) properly or sufficiently, which may cause the system to fail to adequately meet its performance or power consumption requirements.
• performance guarantees e.g., transient deadlines
• the processing cores e.g., CPUl
• FIG. 18 illustrates an aspect DCVS solution method 1800 of setting group-based performance guarantees for a thread or workload 1702 executing in a multicore system.
• the workload 1702 is assigned a 20 ms transient deadline 1804 for a processing group that includes the first and second processing cores (CPUO and CPUl) when the workload 1702 begins executing in the first processing core (CPUO). This may be accomplished by setting a transient timer associated with both the first and second processing cores (CPUO and CPUl) to expire 20 ms after the workload 1702 starts execution, or after 20ms of total execution time.
• the workload 1702 is transferred by the operating system scheduler from the first processing core (CPUO) to the second processing core (CPUl), and the workload 1702 is still bound by 20 ms transient deadline 1804 set for the group.
• FIG. 19 illustrates an aspect DCVS solution method 1900 of setting and enforcing a group-based performance guarantee in a multiprocessor system.
• a DCVS solution may compute the predicted steady state workload based on the scheduled tasks.
• the DCVS solution may compute various performance requirements for the processing core, such as frequency thresholds to meet the power consumption and/or responsiveness requirements of the computing device.
• the performance requirements e.g., frequency
• thresholds may be determined based on the steady state workload, historical information (e.g., amount of time previously spent in a busy state, etc.), processor characteristics, responsiveness requirements, etc.
• the DCVS solution may compute a performance guarantee value (e.g., deadline value, budget value, jump-to-max value, etc.) for a processing group that includes a first and a second processing core.
• a performance guarantee value e.g., deadline value, budget value, jump-to-max value, etc.
• the DCVS solution may compute performance guarantee values for the first processing core, and assign the computed values to a group that includes both the first and second processing core.
• the DCVS solution may compute performance guarantee values for a combination of the first and second processing cores, and assign the computed values to a group that includes both the first and second processing core.
• the DCVS solution may compute the amount of time or work (e.g., CPU cycles, instructions, etc.) required for the first processing core to complete all the tasks in the steady state workload while meeting the various performance requirements.
• the DCVS solution may transition the first processing core from an idle state to a busy state so that the first processing core runs at a computed initial operating frequency/voltage (or within the computed thresholds) and/or meets the various device or system requirements.
• the DCVS solution may monitor the actual workload and/or operating frequency of the first processing core and adjust the frequency/voltage as necessary (e.g., in accordance with a default clock and voltage scaling algorithm).
• the DCVS solution may update the performance guarantee values for the group based on the current operating frequency/voltage of the first processing core.
• the DCVS solution may determining whether the first and second processors have remained in the busy state, with respect to the predicted steady state workload or thread, for a combined period that is greater than or equal to a sum of the determined amount of work (i.e., the amount of time/work determined to be required for the processing core to complete all the tasks in the predicted steady state workload) and the
• the DCVS solution may continue to monitor the actual workload/frequency and make adjustments when necessary.
• the DCVS solution may increase the operating frequency/voltage of the first or second processing cores.
• the operating frequency/voltage of the first or second processing core may be increased to the maximum processor frequency, or increased in steps.
• Various aspects include methods of improving performance on a computing device having multiple processors, the method including determining a steady state workload of a first processor, determining an amount of work required to perform the determined steady state workload on the first processor, computing a performance guarantee value for a processing group that includes the first processor and a second processor, transitioning the first processor from an idle state to a busy state, performing dynamic clock and voltage scaling operations to scale a frequency of the first processor based on an actual workload of the first processor, determining whether the first and second processors have remained in the busy state for a combined period greater than or equal to a sum of the determined amount of work and the performance guarantee value, and increasing the frequency of one of the first and second processors when it is determined that the first and second processors have remained in the busy state for a combined period that is greater than or equal to a sum of the determined amount of work and the performance guarantee value.
• Further aspects may include a computing device having one or more processors configured with processor-executable instructions to perform various operations corresponding to the methods discussed above.
• FIG. 1 Further aspects may include a computing device having various means for performing functions corresponding to the method operations discussed above.
• FIG. 1 Further aspects may include a non-transitory processor-readable storage medium having stored thereon processor-executable instructions configured to cause a processor to perform various operations corresponding to the method operations discussed above.
• DSP digital signal processor
• ASIC application specific integrated circuit
• FPGA field programmable gate array
• a general-purpose processor may be a multiprocessor, but, in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine.
• a processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a multiprocessor, a plurality of multiprocessors, one or more multiprocessors in conjunction with a DSP core, or any other such configuration. Alternatively, some steps or methods may be performed by circuitry that is specific to a given function. [0161] In one or more exemplary aspects, the functions described may be implemented in hardware, software, firmware, or any combination thereof.
• Non-transitory computer-readable storage media may be any available storage media that may be accessed by a computer.
• such computer-readable media may comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that may be used to carry or store desired program code in the form of instructions or data structures and that may be accessed by a computer.
• Disk and disc includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

Landscapes

• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Power Sources (AREA)
• Debugging And Monitoring (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=49382574

Family Applications (1)

Country Status (5)

Families Citing this family (12)

Family Cites Families (7)

• 2013
  • 2013-09-25 WO PCT/US2013/061649 patent/WO2014123587A1/en active Application Filing
  • 2013-09-25 CN CN201380072130.XA patent/CN104969142B/zh active Active
  • 2013-09-25 JP JP2015555992A patent/JP5982588B2/ja active Active
  • 2013-09-25 EP EP13777365.1A patent/EP2954385A1/de not_active Withdrawn
  • 2013-09-25 KR KR1020157022977A patent/KR101659705B1/ko active IP Right Grant

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events