KR100211789B1 - Disk drive for portable computer with adaptive demand-driven power management - Google Patents

Abstract

Data recording disk drives used in battery powered portable computers have several power saving modes of operation. The power saving modes are entered after a time elapsed since the previous data read or write command. The time to enter power saving mode is calculated based on the real-time workload of the computer user, and thus changes continuously during the operation of the disk drive. The disk drive detects the user's current workload by calculating the frequency of the disk drive accesses, and based on the history of the workload, the disk drive determines which of the multiple power saving modes is appropriate and when to enter the power saving mode. Decide

A read or write access to each disk drive is detected and used to calculate the current access frequency. The current access frequency is compared to a threshold frequency previously calculated and continuously updated. The threshold frequency represents a uniform or sporadic access pattern and is calculated from a formula that includes adjustable gain coefficients.

Appropriate power saving modes are entered when the current access frequency is less than the threshold frequency. Intermediate power saving modes can be omitted based on the detected access pattern. Disk drives can also dynamically adapt to changing workloads, thereby storing more energy without compromising performance. This is accomplished by adjusting the gain factors which govern the actual performance of the system, which changes the threshold frequency. The disk drive also determines when to exit sleep mode without having to wait indefinitely for the user's access. Power saving modes can also be customized for different disk drive products. In this case, the power saving modes do not have to follow a standard for setting a fixed time or the number of power saving modes. New user-selected parameters such as ON / OFF and performance and energy targets can be used in place of fixed mode-entry time. These parameters adjust the threshold frequency by adjusting the gain factors. This changes the formula used to determine when to enter the power saving mode.

Description

Disk Drives for Portable Computers with Adaptive Demand-Based Power Management

1 is a block diagram showing a disk drive and a computer system showing various elements that consume energy in the power supply and the disk drive.

2 is a flow chart illustrating the process of counting accesses for one or more power-consuming disk drive components within a predetermined time window.

3 is a block diagram illustrating a microprocessor coupled with a ring buffer in which access densities are accumulated.

4 is a flowchart illustrating the process of calculating an access frequency when the disk drive enters a power saving mode and comparing it with a threshold frequency to determine the access frequency.

5 is a flowchart illustrating a process of calculating a threshold frequency using a previous access frequency stored in a ring buffer.

6 is a flowchart illustrating a process of calculating a threshold frequency using a previous access frequency stored in a ring buffer.

7 is a graph showing the time sequence for entering and exiting a power saving mode.

8 is a block diagram illustrated in FIG. 3 with the addition of a counter for counting time windows between entering and exiting a power saving mode.

9 is a graph showing cumulative energy / response penalty as a function of time to illustrate the adjustment of gain coefficients when the trip level is exceeded.

10 is a flow chart illustrating the process of calculating energy / response penalties and lost opportunity penalties when power saving mode ends.

11 is a flowchart illustrating a process of adjusting gain coefficients and calculating a cumulative lost opportunity penalty when the trip level is exceeded.

* Explanation of symbols for main parts of the drawings

1: Spindle Driver 2: VCM Driver

3: preamplifier / record driver 4: data recording channel

5: Spindle Control Electronic Circuit Part 6: Servo Control Electronic Circuit Part

7 disk controller electronic circuit 9 microprocessor

10: buffer memory 11: interface module

12: power control module 13: computer interface controller

19: control electronic circuit 20: power supply

30: rotary voice coil motor 31: actuator

32: spindle motor 33: data head

40: spindle driver 41: computer

70, 71: Bus

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates generally to disk drivers used in portable computers that are battery powered, such as laptop computers or notebook computers, and more particularly to techniques for minimizing energy consumption. It relates to disk drivers comprising a.

Portable computers can only run for a few hours before they run out of battery power, and the main power consumption factor is the hard disk. Important power management techniques for hard disk drives used in portable computers include several reduced-power or power-save modes of operation, which are used after a disk drive read or write operation. Each of the modes is entered after a time has elapsed for a set fixed time period. For example, at the end of a fixed time period after the user last writes data to the hard disk, the read / write head moves to the parking position and the spindle motor of the disk drive stops. Then when the user accesses the disk drive again, the spindle motor rotates and the heads move across the disk with the appropriate data tracks to read or write data. The biggest disadvantage of such a power saving mode is that it takes time to exit the mode, during which time the user is forced to wait. This has a significant impact on computer performance. In general, the length of a fixed time period is set by the computer user using software.

This conventional power saving technique has a problem that the user does not have the data necessary to select a good value as a fixed time period. In other words, the user has only limited knowledge of the type of access and no information about the energy and performance parameters of the disk drive. The user is isolated from the information in the form of internal access of the disk drive by the hardware and software in the system. It is not easy to compromise properly between energy and performance because the fixed time to enter sleep mode is not authorized for the user workload. The user must change the settling time in anticipation of the workload, and choosing the settling time too short or too long can have the opposite impact on performance and energy consumption.

When the access type is in the form of a long period of inactivity following an abruptly active state, entry time in a short mode saves energy. However, excess energy is used when the period of inactivity is close to the entry time of the mode. In addition, entering the mode after a short period of inactivity can affect performance, and in general, mode recovery time results in more frequent access delays.

If longer periods of inactivity are less common than shorter periods of inactivity, long mode entry times relatively reduce the impact on computer performance and the tendency to use excessive energy.

During certain user workloads, it is likely that the optimal time for fixed time will not be constant. Moreover, the workload can change without the user's knowledge, due to the action of the application software that the user runs.

In fact, users want to choose between energy consumption and computer performance, not between fixed mode entry times. The user selects fixed mode entry times only by guesses in achieving energy and / or performance goals. It is certainly desirable to receive energy and performance targets from disk drive users. With these goals, the disk drive can take any that is a suitable way to meet these goals. This also allows the disk drive to use more power saving modes. This is because the user does not need to know the specific power saving modes operating in the drive unless the user chooses a further fixed time for each mode entry.

As discussed above, disk drives can detect changes in workload, actively adapt to changes in workload, and use energy and performance targets instead of fixed time to determine when to enter and exit sleep mode. What is needed is a method and system for achieving power management.

It is therefore an object of the present invention to provide a disk drive that performs power management to determine the entry and exit times of a power saving mode by predicting a subsequent user's request from a past access history of the disk drive.

These disk drives do not know how much performance and energy consumption costs are associated with entering and exiting sleep mode, so that in terms of current user-selected predetermined or mode-entry time or fixed mode-entry time, Has an advantage. The disk drive has information on energy break-even time and recovery time related to the power saving mode. The energy break-even time is the time required to keep the disk drive in a specific power saving mode so that the extra energy consumed while returning from the power saving mode is balanced with the reduced energy consumption while in the power saving mode. The recovery time is the time it takes for the disk drive to return from the sleep mode to the active state. The disk drive maintains a history of the trajectories of the type of access, i.e., the various requests for reading / writing data or moving the actuator. As a result, the disk drive detects the current user's workload and determines which of the multiple power saving modes is appropriate and when to enter the power saving mode.

In the preferred embodiment, if an access of each disk drive is detected, it is used to calculate the current access frequency. The current access frequency is compared with the one previously calculated and continues to update the threshold frequency. The critical frequency represents a uniform or sporadic access form and is calculated from a formula that includes adjustable gain factors. While the disk drive is operating, a suitable power saving mode is entered when the current access frequency is less than the threshold frequency. The intermediate power saving mode based on the detected access type can be omitted. Disk drives can also be actively applied to changing workloads, which saves a lot of energy without compromising performance. This is accomplished by adjusting the gain factors related to the actual performance of the system. The disk drive also determines when to exit sleep mode without having to wait for user access.

In the present invention, many power saving modes are disclosed in which the user does not have to remember the trajectory of the situation change. Power saving modes can also be applied appropriately for different disk drives without having to follow any standard for setting fixed time or number of power saving modes. New user-selected parameters such as ON / OFF and performance and energy targets can be used in place of fixed mode-entry time. These parameters adjust the threshold frequency by adjusting the gain factors. This changes the formula used to determine when to enter the power saving mode.

Hereinafter, the characteristics and advantages of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram illustrating various components used in power management for a disk drive of the present invention. Generally, the disk drive 40 is a hard disk drive contained within a housing of a laptop computer such as the computer 41. The disc drive 40 includes one or more discs, such as a conventional disc 34, which are attached to the spindle motor 32 and rotated. The data head 33 is connected to an actuator 31 which is generally operated by a ratary voice coil motor (hereinafter referred to as VCM) 30. Spindle motor 32 is driven by spindle driver 1 and spindle control electronics 5. The servo control electronics 6 is used to position the head 33 over the respective data tracks of the disc 33 and is connected to the VCM driver 2 which supplies current to the VCM 30. Also shown in FIG. 1 is a crash stop 37 and a Load / Unload ramp 38 for the actuator 31. If the disk drive 40 is of the Contact Start / Stop type in which the head 33 remains in the disk 34 when the spindle motor is not rotating, the actuator is driven by the current supplied to the VCM 30. Driven towards the crash stop 37, the head 33 then stops on a nondata landing zone close to the inner diameter (ID) of the disc. If the disk drive 40 is of the L / UL type, the actuator 31 is driven when the spindle motor is paused to support the head 33 so that the suspension over the lamp 38 and in contact with the disk 33 Unload). ID landing areas on CSS disk drives and ramps on L / UL disk drives are often referred to as head-parking lications.

The data head 33 reads / writes user data on the disk 34 and is generally a thin film inductive (TFI) read / write head or magneto-resistive (MR) TFI recording head having a read head. The data head 33 is connected to the preamplifier / write driver 3, the data write channel 4 and the disk controller electronic circuitry 7. The data recording channel 4 can be in any form, such as a peak detector or Partial-Response Maximum Likelihood (PRML), and includes functions such as data detection, encoding and decoding. The disk controller 7, connected to the microprocessor 9, handles the process of reading and writing data, coordinates communication with the computer 41, manages the buffer memory 10, and controls the servo control electronics 6. Commands the spindle control electronics circuitry 5.

The disk drive 40 also includes a microprocessor 9, an associated memory 8, a buffer memory 10, an interface module 11 and a power control module 12. Microprocessor memory 8 is used to store code and data for microprocessor 9. The buffer memory is used to store data transferred from the computer 41 to the disk drive 40 and is generally cache memory. The interface module 11 controls the transfer of information on the interface from the computer interface controller 13. Integrated Drive Electronics (IDE) interfaces and Small Computer System Interfaces (SCSI) are the most common types of interfaces.

All elements of the disk drive 40 shown in FIG. 1 require operating power. The power control module 12 controls power management of the disk drive 40, which receives power from the power source 20 via the bus 71. All components are not shown in FIG. 1 with power connections, each of which is powered directly or through other components rather than via bus 71. The power control module 12 may be physically included as a logic circuit in the controller 7 or stored in the memory 8 in microcode form executed by the microprocessor 9. The power line from the power source 20 to each energy consuming element is directly controlled by the output of the module 12 or the microprocessor 9 or the elements are turned off or turned on. commands to turn on and to change the power state of each element.

Although the connection between the various elements of the disk drive 40 is schematically shown in FIG. 1, the above functions can be obtained by arranging the elements differently. For example, microprocessor memory 8 may be included within microprocessor 9 or may be combined with buffer memory 10. Moreover, it is common to have one or more microprocessors in a disk drive, one of which is primarily used for interface and drive management functions and the other for servo functions. In this case the servo control electronics 6 may comprise a separate separate servo microprocessor.

The computer 41 includes an interface controller 13, a processor 14, a memory 15, a display 16, a keyboard 17 or other device, a peripheral device 18, and a power source 20. And a control electronics unit 19. The interface controller 13 controls communication with the disk drive 40. The power source 20 is a power source for the computer 41 and the disk drive 40. The power source used in a portable device may be another type of power source such as an AC power supply, but is generally a rechargeable battery. The power source 20 supplies power through the bus 70 to various elements in the computer 41 and the disk drive 40 supplies power through the other bus 71. The power supply 20 also communicates with the control electronics 19. For example, batteries used in portable computers include internal control electronics that monitor battery status, such as discharge status.

Usually disk drive 40 does not continue to consume energy at a constant rate. In other words, an intermittent current pulse to the VCM 30 causes the current to flow through the VCM 30 rather than when track following, such as keeping the head 33 on a single data track. More energy is consumed when seeking, such as causing it to move across the data tracks of the disc 34. In addition, during the write operation, additional energy is consumed due to the flow of the write current through the coil of the TFI head. With such disk drive accesses, the active power mode is roughly divided into two cases: That is, seek / read / write power mode and idle power mode. The seek / read / write power mode is generally estimated because it is based on the assumption of user workload. Idle power mode is a normal track tracking operation when no data is recorded / read, so no search occurs. The term active state is used to indicate the disk drive state when in either seek / read / write or idle power modes.

Two common power saving modes are referred to as idle2 and standby. In idle 2 mode, the actuator 31 is parked (i.e. moved towards the crash stop 30 or unloaded onto the ramp 37), the servo control electronics 6 and the preamplifier 3 and the channel 4 The write / read electronic circuit portion including the is turned off. The idle 2 mode thus substantially reduces or eliminates the power supplied to the VCM driver 2, the servo control electronics 6, the preamplifier 3 and the channel 4. In idle 2 power saving mode, it is also possible to reduce the power supplied to the disk controller 7 and the microprocessor 9 since the servo operation and the write / read operation are not active.

In the standby mode, the actuator 31 is moved to the parking position and the spindle motor 32 is turned off. The standby power saving mode has all the power saving effects of the idle 2 mode, but also reduces the power supplied to the spindle control electronics 5 and the spindle driver 1. In some embodiments, buffer 10 may be turned off in either or both of idle2 mode and standby mode. Additional power saving modes are also possible. For example, the sleep mode includes the power saving feature of the standby mode, and also supplies power only to those parts necessary for responding to the sleep recovery command from the part of the interface controller 11 and the computer interface controller 13. And almost all remaining electronics are turned off.

Table 1 shows power values for a typical 2.5-inch disk drive in the two active and two power saving modes. It is clear from this table that the power saving mode, i.e., idle 2 and standby mode, significantly reduces energy consumption.

Table 1 also shows the recovery time (T) for the two sleep modes, which is the time it takes for the disk drive to return from sleep mode to the active state. The average recovery power P is also shown, where the energy break-even time T can be calculated from this information. This is the time the drive needs to stay in a particular power saving mode so that the extra energy consumed while recovering from that power saving mode is in balance with the reduced energy consumption while in this mode. Assuming that the normal active mode is idle and the recovery time T is only applied in the power saving mode, T can be estimated by the following equation.

Table 2 shows the energy break-even times T _BE for the example of Table 1.

In the present invention, an actual user workload is used to determine which power mode is most appropriate and when to enter that mode. Thus, entering the power saving mode is preferably in accordance with the needs of the user's workload, rather than being predetermined by a fixed time of user-selection.

On the other hand, in order to optimize energy consumption, it is very important to know the type of access to the disk drive and what type of operation each element exhibits. In the power save mode, an accessor is defined as an operation that requires a return from the power save mode. For example, assuming that the power saving mode is idle 2, the disk 34 is rotating, the servo control electronics 6 and the VCM driver 2 are turned off, and the buffer 10 includes the disk cache. If is active, the disk drive access to move actuator 31 to read / write data is counted as one access. Cache hits are not access to this power saving mode, but are not certain other interface commands such as status queries from computer 41 because they do not require read or write to disk 34. In a simple design, the effect of cache hits on counting accesses does not need to be considered. In this case, the read or write command received through the interface is counted as one access, even though all the data remains in the cache. These assumptions may reduce energy savings, but they can be cost effective because they reduce the complexity of designing power management technologies.

The recent access pattern must be taken into account when making a decision to enter a specific power saving mode, for example standby (reducing the rotational speed of disk 34). The access pattern contains information about the software process that is driving the access. Access patterns can be characterized by frequency, ie the rate at which disk drive access occurs, and the distribution of frequencies can be determined from the access history. It is possible to determine when the observed access frequency does not belong to the frequency distribution of the recent access history. This determination is made statistically by estimating the probability that the access frequency is not part of a recent access pattern during the day, thus indicating that the processing of the access pattern and the software associated with it is stopped. Many decision coefficients are involved in making this judgment. In the present invention, these coefficients are also dynamically adjusted based on the performance of the disk drive to provide adaptive power management.

It is also possible to detect periodic accesses from the access frequency and to enter and exit the power saving mode by predicting the start and end of the periodic access. An example of periodic access is the case of autosave in which a document storage interval time is specified in a word processing program. That is, the software user can use the automatic processing of the word processing software to automatically write the file to disk every five minutes. If you can exit sleep mode just before entering periodic access, you will improve performance because you don't know the response delay. Entering the power saving mode immediately after the end of the periodic access can save energy because the delay time can be entered into the power saving mode.

In the preferred embodiment, the disk access pattern is assumed to belong to one of two categories: uniform access and sporadic access. For example, the mean and standard deviation of the access frequency can be calculated, where the access frequency can be considered well defined if the standard deviation is a few minutes of the mean. Otherwise, the access pattern is considered sporadic, i.e. poorly defined by the mean and standard deviation.

In the case of uniform access, the access pattern is considered to be terminated if the observed access frequency is less than or equal to the mean minus several times the standard deviation. This is equivalent to selecting the probability that the observed access frequency belongs to the observed access pattern. Alternatively, it is also possible to estimate the end of an access pattern using a few minutes of the minimum observed access frequency. The basic principle is to use a recent access frequency to characterize an access pattern to determine a threshold frequency from it. Thus, when the access frequency exceeds this threshold frequency, it is assumed that the probability of the access pattern stopping is very high.

The access frequency is measured by selecting a time window, counting the number of accesses occurring within the window, and converting this number to frequency. Different time windows can be selected for each power saving mode. The number of accesses occurring within one time window is called the access density.

2 is a flowchart showing the process of measuring the access density in detail. The timer is first checked in step 403 to determine if the window time has expired. If the window has not yet exited, then check in step 401 if an access has occurred, and if so, in step 402, increase the density counter.

2 shows the behavior when all the time windows are a plurality of shortest windows. In this case, the densities for all power saving modes are increased in step 402. The reset of the density counter is not shown, but before the access to the next window begins, the density value for the current window is transferred to another storage device, such as a ring buffer, and then initialized. In a power saving mode with time windows in the minimum time window of a certain multiple, the density count is initialized only at the end of the longer window. 2 illustrates the operation of a polling loop design. The equivalent interrupter-based design may be easily derived from FIG.

The process shown in FIG. 2 is part of the functionality of power module 12 in FIG. 1 and may be implemented in hardware and / or in software. 3 shows a suitable hardware configuration for implementing the process of FIG. The timer 201, the counter 202 and the ring buffer 203 are shown as part of the controller 7. The ring buffer 203 is addressed by the microprocessor 9. 4th The same is true of the process of FIG. 6, but a set of program instructions for performing each of the steps of the process shown in FIG. 2 are stored in microcode in the memory 8 addressed by the microprocessor 9. The access signal 220 is input to the density count 202 in the disk controller 7. For example, signal 220 may be sent to controller 7 at the same time as responding to a read request from interface 11 and interface controller 13 or when controller 7 sends a read command to channel 4. Is caused by. The access signal 220 is sent to the counter 202 every time the disk drive is accessed, and the counter 202 counts the accesses 220. The timer 201 continues to operate to output a signal 221 at the end of each time window. The density value at the counter is output as signal 222 and passed to a density storage buffer 203, such as a ring buffer. The end-of-window signal 221 from the timer 201 causes the ring buffer 203 to write the density value 222 to the next memory location and to write the buffer's write pointer value. To increase. The signal 221 then initializes the counter 202. The density value is read from the ring buffer 203 by the microprocessor 9. Microprocessor 9 transmits clear 223 and read 225 signals to read buffer 203 and receives output signals 224 indicating the density value. The microprocessor 9 uses the density value to issue instructions to change the power state of each component of the disk drive while executing the microcode stored in the memory 8. Changing the power state here means entering or exiting the appropriate power saving mode.

In general, the time window for the power saving mode is selected to allow a range of access frequencies that the disk drive can appropriately respond to the power saving mode in order to obtain a good response. The frequencies of interest here are cases where the period has values similar to the latency and energy break-even times of the power saving mode. This allows for a time window with a value close to the energy break-even time, which can be further optimized by examining the behavior of the disk drive or through simulation. Other factors, such as performance goals, should also be taken into account when selecting the time window. In one example of the disk drive of Table 1 above, it is preferable to select a time window of 400 ms for idle 2 and 1.6 s for standby.

The access density value obtained in the above description is the number of accesses occurring in a particular time window. The density value is then converted to a frequency value by a scale, which is a scaling method. The dynamic range in frequency is extended by implementing such that a density of zero is not equal to a frequency of zero. When the density is zero, the frequency is calculated from the number of consecutive zero densities. One such conversion formula is:

Here, the number of zero density (Number_zero_density) means the number of consecutive time windows having a density of zero. Scale refers to a scale factor for converting density into frequency. From the above formulas (2) and (3), a good approximation of the access frequency can be obtained when the access frequency has a value less than 1 / time-window. To facilitate processing, it is common to use frequency units with integer values defined by magnification factors. In other words, the density with a value of 1 corresponds to the frequency of the scale. Magnification factors are chosen for simplicity of calculation and provide a desirable dynamic range of frequencies. Typically, a value of 256 is used as the scale, which is suitable for 16-bit processing, and also makes scaling simple when the density is not zero, thereby reducing the number of bit-shift operations. Because it reduces.

Table 3 below is an example of the density-frequency conversion using the above formulas (2) and (3) and a time window of 400 s, that is, the idle 2 mode. The first line shows the start time for each time window, the second line shows an example of the access density value for each window, and the third line shows the access frequency derived from the access density in the second line. For example, 19 accesses occur in a time window between 0.4 s and 0.8 s, and the magnification factor has a value of 256, corresponding to frequency 4864. Also, since there is no access in the time window between 1.2s and 2.0s, two zero density values are obtained. These three density values are converted to a single frequency by the formula 256/3 = 85.

If you want to reduce the delay even a little during the disk drive's access cycle, it is advisable to minimize energy management calculations. This approach to density has the advantage that it does not require much computation when access occurs in the disk drive. It is rather desirable to omit this density measurement or density-frequency conversion when the disk drive is accessed too frequently. By doing so, the response time is faster, and the access density for the interval in this case can be set to a fixed value applied in the case of a large access case. In this case, the access density or frequency value for the interval is updated when the disk drive becomes idle. For example, a density of 256 corresponds to a frequency value of 65,536 having a magnification factor of 256 and thus may be a busy period.

The value of the access frequency obtained in the density-frequency conversion described above is an estimate because the exact timing of the accesses is not obtained from the density. However, if the window time is properly selected, the estimate is accurate enough in the frequency range of interest. Measuring or estimating the access frequency is also possible with other techniques such as Fourier transform technique instead of the above method.

The microprocessor 9 determines to enter the power saving mode when the current access frequency becomes less than the threshold frequency determined in the access history. 4 is a flowchart showing details of entering a power saving mode. In this flowchart, it is assumed that various power saving modes are assigned to the maxmode in order from the value of 1 in order of increasing the effect of power saving. The term tf [mode] refers to the threshold frequency of a particular power saving mode. The term lf [mode] refers to a low-frequency flag for a particular mode. This flag helps to control the calculation of the threshold frequency. In step 301, the last mode is set to 1, implying that all power saving modes have been checked. And for all modes, the threshold frequency is cleared (tf [mode] = 0) and the low-frequency flag is also cleared (If [mode] = 0).

In step 302, the process then waits for the minimum window time to elapse. This process has already been shown and described in FIG. Returning to FIG. 4 again, when one or more window times, i.e., a unique time window is used for each power saving mode, assume that the longer window time is an integer multiple of the minimum window time. If window time has elapsed once and the access density has been measured, step 303 is performed. In step 303, the power saving mode to be checked is set to the max mode in which energy saving is maximum. In step 304, the process checks whether the current time is window time for the mode to be checked. If no, perform step 308.

In step 308, if the current mode is the previous mode, the flow returns to step 302 where the density is measured for the next time window. At this point, you must determine which of the active power saving modes is time to check. If the current mode is not the previous mode in step 308, the next lower mode (mode-l) is selected in order of decreasing power saving effect in step 309. Thereafter, step 304 is performed again. Note that the state of previous mode = 1 in the check of step 304 is always true since it is the same as the state of ending step 302. If the check is true in step 304, a new density value is made available for the mode, and step 305 is reached. In step 305, the access density measured in step 302 is converted to the frequency freq using the conversion formulas (2) and (3). In step 306, the access frequency freq calculated in step 305 is compared with the threshold frequency, tf [mode], for the current mode. If the access frequency is greater than the threshold frequency, step 307 is reached. In step 307, the threshold frequency tf [mode] for the current mode is obtained. The threshold frequency may or may not change depending on the condition. Details of adjusting the threshold frequency will be described later with reference to FIG. Then in step 308, the process continues for the next mode as described above.

On the other hand, if the access frequency is less than or equal to the threshold frequency in step 306, the flow advances to step 310, and the current power saving mode is entered into the mode. This means that the microprocessor 9 sends a signal to the appropriate element of the disk drive that is trying to reduce power. Thereafter, if the current mode is the max mode in step 311, since there is no power saving mode to perform any more, the process proceeds to step 313 and the process of this flowchart is stopped. However, if there is still a power saving mode to be checked, go to step 312 to update the previous mode to [mode + 1]. This is to reflect that all power saving modes less than or equal to the entered mode are no longer considered. Thereafter, the process returns to step 302 to continue the next step.

In step 304, the order of checking the power saving modes starts in order from the most energy saving mode first, and then from the energy saving mode in order to save as much energy as possible. By doing so, it is possible to enter the most advantageous power saving mode for the disk drive in step 310. For example, by applying the present invention, a disk drive in idle mode can enter standby mode without first entering idle 2 mode. If a specific power saving mode other than the maximum mode is entered, the disk drive will operate in that specific mode until it is suitable for entering a better mode or until it exits that particular mode.

5 is a flowchart showing the details of the preferred embodiment of calculating and adjusting the threshold frequency for the power saving mode. The most recent access frequency is stored in a ring buffer (not shown) addressed by the microprocessor 9. The buffer may be different for each power saving mode. The ring buffer is a set of registers or memory locations where each frequency measurement is stacked. The ring buffer holds several frequencies corresponding to multiple registers or memory locations for a set. When several frequencies are loaded, adding an additional frequency value causes the oldest value to be lost, which provides the sliding view of the recent access history.

In step 501, a check is made to determine if there are enough values in the ring buffer to calculate the threshold frequency for the current power saving mode. The minimum number of frequency values is typically two. The larger the number, the higher the statistical accuracy, while the values must be selected before calculating the threshold frequency. If the number of values in the ring buffer is not sufficient, step 503 is reached. In step 503, it is checked whether the current frequency freq is greater than scale / 2. The scale here is defined in formulas (2) and (3) above. If the result of this check is 'true', control proceeds to step 504 where the current frequency value is entered into the ring buffer. If it is 'false', the number of consecutive time windows with zero density continues to increase, so the previous frequency value in the ring buffer must be changed. This is done in step 509. Note that this is not an input operation to the ring buffer, so the number of values in the ring buffer does not change. Step 510 enters from steps 504 and 509 and returns the process to step 501.

In step 501, if the number of values in the ring buffer is sufficient, step 502 is reached. In step 502, the current access frequency freq is compared to the active frequency threshold freq_act. This value speeds up the process and provides a safety threshold. Scale / 2 is the first selectable value for freq_act. If the check at step 502 is true, the disk drive access frequency is greater than the active threshold, so no further calculation of the threshold frequency is necessary. Thereafter, the process proceeds to step 505, after clearing If [mode], the process proceeds to step 503, after which the process proceeds as described above.

On the other hand, if the check in step 502 is 'false', then the process needs to proceed further as the access frequency is less than the active threshold. In step 506, the low-frequency flag If [mode] is checked. If the flag has been set, go back to step 503 and the current low-frequency threshold is maintained. Suppose the end of an access pattern is checked when the low-frequency flag is set. As a result, the current frequency is assumed to be not part of this pattern and cannot be used to change the threshold frequency. However, this frequency is still placed in the ring buffer because the low-frequency flag has cleared as the activity threshold freq_act increases. In this case, the assumption that the access pattern has been suspended is wrong. Thus, the frequencies occurring during the test are now considered part of the access pattern.

On the other hand, if the low-frequency flag is cleared at step 506, then a new frequency needs to be calculated, and the flow goes to step 507. This calculation is performed in step 507, and then passed back to step 503 with the low-frequency mode flag set in step 508.

The process shown in FIG. 5 uses the most recent access history to calculate the threshold frequency for each power saving mode. The calculation is performed only when the access frequency is less than the active threshold. This has the advantage that performance is less affected because the access frequency is calculated only when the disk drive is inactive.

In step 507, the threshold frequency is calculated based on the access pattern in the ring buffer associated with the previous power save mode. That is, the mean and standard deviation for the values in the ring buffer are calculated. These calculations include complex number calculations. Instead, the average frequency meanf for the uniform access pattern is calculated from the maximum and minimum frequencies, maxf and minf in the ring buffer by the following equation.

This assumes that the distribution is well specified and gives a good estimate for the uniform access pattern. The standard deviation sdevf is calculated from the range of frequencies as

However, since it is not necessary to accurately calculate the mean value and standard deviation, they are more preferably calculated using simplified formulas (4) and (5).

In sporadic access patterns, only a few fractions of the minimum frequency is sufficient.

Formula (6) shown below (8) This is a formula used in step 507.

The t1 [mode] is a value corresponding to the uniform access pattern, while the t2 [mode] is a value corresponding to the sporadic access pattern. The larger of these two values is used as the threshold frequency for the mode.

There are two gain factors, g1 [mode] and g2 [mode], which are for uniform access patterns and sporadic access patterns, respectively. These gain coefficients have different values for each mode. For the disk drive of Table 1, the values of g1 = 1 and g2 = 4 are suitable for both idle 2 and standby modes. Formula (6) As can be seen from (8), increasing the value of the gain coefficient has the effect of decreasing the threshold frequency, and decreasing the value of the gain coefficient has the effect of increasing the threshold frequency. Therefore, the entry action of the power saving mode is suitably harmonized with selecting and adjusting these gain factors. These gain factors can be selected and adjusted by the user through the appropriate system or application software.

Since a separate ring buffer is used for each power saving mode, the size of the buffer must be selected independently. Increasing the size of the buffer increases the time required for the access history to be written, while decreasing the size decreases the time. When choosing a buffer size, the length of the history must also be taken into consideration in order to realize an improved response to recent events and to reduce the amount of memory required. For the disk drive in Table 1, the buffer size was properly chosen to be 16 for the two power saving modes. More sophisticated ways of maintaining access histories are possible. For example, frequencies in a ring buffer have weighting factors related to the length of time in the buffer. A histogram of access frequencies can be used in place of the ring buffer as a mechanism for deleting old data, such as renormalization. However, ring buffers have the advantage of simplicity of design.

The ring buffer can be constructed using density instead of frequency, as shown in FIG. In this case, the density is converted to frequency using a critical frequency calculation. The process of calculating the threshold frequency using the density buffer in FIG. 6 is shown. This flowchart corresponds very closely to the flowchart in FIG. 5 with some exceptions. That is, step 601 corresponds to step 501. Then, step 603 is added to convert the density density to the frequency freq using formulas (2) and (3). Steps 604, 605, and 606 correspond to steps 502, 505, and 506, respectively. Next, step 607 was added to convert each density value in the ring buffer to a frequency value using formulas (2) and (3). Note that this conversion step results in fewer frequency values in the ring buffer than the number of density values since successive zero densities are converted to single frequency values. Steps 608 and 609 correspond to steps 507 and 508, respectively. Finally, step 602 enters the current density value into the ring buffer for the selected power saving mode.

Both configurations for the ring buffer described above are suitable for calculating the threshold frequency. The frequency buffer has the advantage that the hysteresis length increases with time when there are low frequencies. Density buffers, on the other hand, have the advantage of having a fixed hysteresis length. The density buffer also reduces computational overhead during the interval between activity disk accesses because the density-frequency conversion is delayed until the active frequency threshold is reached.

As described above, once a disk drive enters a power saving mode, it enters another power saving mode or terminates in either of which the disk drive returns to an active state. The former occurs when the estimated access frequency continues to decrease to reach a threshold frequency in another mode. The latter occurs when a disk drive access occurs or a periodic access pattern is detected that results in an active drive-initiated entering. When the disk drive returns to the active state, it is desirable to clear the ring buffers because it is assumed that a new access pattern is measured. This further reduces the influence of the old pattern in the ring buffer. In another configuration, it may be desirable to keep ring buffer data if it is possible to change the weighting coefficient for the data from the previous pattern.

Still other coefficients can be used to adjust the gain factor, which in turn affects the determination of the threshold frequency. For example, attempting to enter a power saving mode in a standby state where the speed of the disk where the head cannot immediately read / write data is reduced can be aggravated by the current cache hit rate. High cache hit rates are used to adjust the threshold frequency to a higher value, for example by reducing the gain factor. Although the access occurs for a short period of time after reaching the threshold frequency, this access hits the cache. As a result, the possibility of using extra energy is still reduced. Similarly, a lower cache hit rate would result in a higher probability of successive accesses requiring reads or writes from the disk, thus lowering the threshold frequency accordingly. In such a case, an additional power saving mode for controlling the power required for the cache, such as buffer 10 in FIG. 1, is desirable. By doing so, the cache buffer remains active even in standby mode. The gain coefficients for this mode are of course affected by the cache hit rate. In cases where the cache hit rate is high, it would be desirable for performance to decrease the threshold frequency while allowing more cache hits. It is also possible to treat read and write operations separately or to characterize cache accesses using the locality in which access occurs. All of these are used to adjust the gain factor and eventually the threshold frequency.

As discussed above, there are several adjustable parameters in the above-described design, which include gain coefficients g1 and g2, time window size, ring buffer size, density-frequency conversion scale factor scale, and active threshold frequency. do. However, in the preferred embodiment, the most suitable parameter for adjusting the entry action of the power saving mode is the gain factor in formulas (6) and (7). The adjustable parameters have fixed values at the time of manufacture of the disk drive, but they can be adjusted by the user through specific tasks or by dynamic power management while taking performance penalties.

As described above, the computer 41 may issue a command to set the gain coefficients directly through the interface controller 13 of FIG. However, it is rather advantageous that the computer 41 is not directly involved in the internal detailed process for energy management. So we use a separate coefficient of performance pf. It is controlled by commands from the computer 41 and is not affected by the specific energy management design applied to the disk drive.

In general, energy management in sleep mode on disk drives requires a good tradeoff between access performance and energy savings. This is due to the direct effect of the recovery latencies of the power save mode described above. Thus, a single weighted coefficient of performance is required which expresses the importance of compromise between energy and performance. A single weighted coefficient of performance at one extreme maximizes energy savings without considering performance, while a coefficient at the other extreme maximizes performance without considering energy savings. This method is completely different from the conventional disk drive power management technique, which adjusts the fixed time to enter the power saving mode by the energy management command. These commands are not directly related to performance or energy savings, and furthermore, different drives with the same fixed time setpoint will produce different energy savings and performance effects. However, the scale of energy vs. performance factor has a fixed value so that all drives operate similarly.

In a preferred embodiment, the coefficient of performance command is defined as in Table 4 below.

Here, a value of 0 corresponds to maximum energy saving, and a value of 255 indicates that maximum performance, that is, energy management is not required. This user-selected performance factor command may also be emulated using standard fixed time commands available in common disk drive interfaces such as SCSI or IDE. In the case of the IDE, the standby mode time value value ranges from 0 to 255 with no power saving mode, and actually ranges from 5 seconds to about 20 minutes because time = 5s * value. In this case, in converting the command to a coefficient of performance, 0 represents the maximum performance, so the linear scale from Table 4 is 0. 1 instead of 254 It will have a range of 255 values.

In a preferred embodiment, the input from the coefficient of performance command is changed to the operation of the energy management system by adjusting the gain factors g1 and g2. The resulting value of the gain factor varies within a range between two limits that represent the extremes of energy and performance. For example, in the case of uniform access of formula (6), g1 corresponds to a multiplier in estimating the standard deviation sdev. This is a practical range for g1 g1 5 is presented. Similarly, g2 is only a few minutes of the minimum frequency minf, so as a practical range 1 g2 10 is presented. Statistically, the lower limit of g1 corresponds to about 20% probability for normally distributed data where the threshold frequency is actually part of the distribution. If the lower limit is too small, there is an inadequate chance of entering a power saving mode, resulting in increased energy use. The upper limit of g1 corresponds to a probability of about 10 ^{−6 which is} slightly larger than sufficient. Although not statistically well defined, g2 exhibits a similar pattern to the above change of the limit value. To make the selection of the limit value more effective, it is desirable to consider empirical data on performance and energy action. Moreover, in practice many of the setpoints have values that are clearly distinct from one another, so that the precise resolution shown in Table 4 is not required.

Equations (9) and (10) below show a conversion formula of the coefficient of performance pf for the commands of Table 4 for the disk drive having the power values of Table 1.

The limit used here is approximately 0.75 g1 4.75 and 1.5 g2 9.5. Formulas (9) and (10) are designed to have integer values for ease of implementation. As a final step for calculating the formulas (6) and (7), it is preferable to carry out with a scale factor of 16.

Other parameters, such as the time window size, can also be adjusted using the performance factor. In general, larger values give improved performance in power consumption costs. Moreover, a subset of possible power saving modes is also selected. For example, if there are several power saving modes that affect performance more than others, and you think performance is important, you should not use them. The above mentioned coefficient of performance pf has the advantage of allowing the disk drive designer to determine the parameters that need to be adjusted to achieve the performance goal. The end user or system assembler does not need to know the details of a particular implementation.

In general, entering power saving modes is based on disk drive access. Satisfactory power management behavior can be obtained using equations (6) and (7), where the parameters have a fixed value until new values are selected. Selecting a set of parameters, such as gain coefficients, determines the suitability of the selection according to the actual access pattern. If there is a change in the access pattern affecting the behavior of the power saving mode entry, then new parameters must be selected. However, the parameters are dynamically adjusted as the access pattern changes. This is accomplished by using an adaptive system that determines how to approach performance targets and dynamically adjusts the gain factors accordingly.

Achieving dynamic adaptability requires the ability to judge the actual action compared to the action required. The best performance is obtained when the magnitude and direction of deflection from the performance target is measured. These measurements are treated as penalties and then used to adjust the gain factors. For power management, it is convenient to define two penalties: Energy / Response Penalties (hereinafter referred to as ERP) and Missed Opportunity Penalties (hereinafter referred to as MOP). The former occurs when the disk drive uses excess energy or affects performance, implying that the threshold frequency tf is chosen too high. The latter occurs when the disk drive enters power saving mode when deemed appropriate, implying that the threshold frequency tf is selected too low. Since the two forms of penalty have opposite effects on determining the critical frequency, they are used to balance system behavior.

Penalties are used to adjust the parameters in the formula of the threshold frequency, so they can be calculated in any unit at your convenience. Calculating the penalty on a time basis has the advantage of being simple to calculate. On the other hand, calculating the penalty in frequency units has the advantage of having several input values available from the demand-driven calculation.

7 is a graph showing a time sequence for entering and exiting a power saving mode. Where the horizontal axis represents time and the vertical axis represents power. Each short tick mark on the horizontal axis represents a time window for power saving mode. Three power levels are used in the graph: search / read / write power PO, idle power P1 and mode power P2. To make the distinction easier, all accesses to disk drives are marked at short intervals in the power level PO. The time TO is the beginning of the time window in which access occurs. The device enters the power saving mode at the time T1 slightly passed and continues until the time T2. The drive then returns to the active state when the next disk access occurs. The next disk access occurs in the time window starting at time T4.

In performance penalty, the duration of the power saving mode is very important.

A counter 204 is added in the hardware configuration of FIG. 3 shown and described above to illustrate it in FIG. Counter 204 counts the number of time windows 221 between all entry 226 and mode exits shown as access 220. Output 227 represents the duration of the mode in time window units. This value may be used directly for time units or may be converted to frequency units via formulas (2) and (3).

In the energy / response penalty erp, both the impact on energy and the response time are evaluated. The energy penalty, ep, is a measure of how much extra energy was used and when inadequately entering sleep mode. The response time penalty rp is a measure of how much impact the actual data throughput will have on entering the power saving mode improperly. In the example of FIG. 7, two mode entry is shown, one from T1 to T2 and the other from T5 to T6. The energy / response penalty is not generated for the mode entered at T1 but at the mode entered at T5. This is indicated by the ERP level in FIG.

In the energy penalty, there is an energy break-even time T _BE for the power saving mode that causes the energy saved for the duration of the mode to be rs shaped by the return energy. For reference, referring to the first mode entry of FIG. 7, T0 is the access density. Represents a previous time window of 0, T1 represents the time to enter the mode, and T2 represents the time to exit the mode. The value of (T2-T1) is the mode duration counted in the counter 9 of the implemented hardware. (T2-T1) If T _BE , an energy penalty ep occurs. The energy penalty ep is calculated as a function of (T2-T1) and T _BE . That is, in a preferred embodiment,

((T2-T1) If T _BE ) is true, then ep = 16-16 * ((T2-T1) / T _BE ),

In formula (II), the energy penalty ep is calculated to range linearly from the value of 16. At this time, the duration of the power saving mode is essentially zero for the value of zero when the response time exceeds the limit. This is because access occurs immediately after entering the mode. Another energy penalty formula is possible which gives different weights to the penalties. However, formula (II) is quite simple and gives a good estimate of the energy impact. In practice, the values of the parameters in formula (II) do not have to be exact, and the calculation is performed as an integer calculation. Integer calculation has 16 penalty levels that provide sufficient resolution.

The response time impact of the power saving mode is calculated based on the additional latency incurred upon return. This is an estimate of the impact on throughput, a measure of disk drive performance. As a result, the upper limit tub of throughput effect can be derived from the gain factor. The throughput impact from the power save mode is measured as rl / (T2-T0). As a result, the response timeout trl is determined according to the upper limit tub. Trl = rl / tub. (T2-T0) In the case of trl, there is a response penalty at any time. The value of (T2-T0) is calculated from the threshold frequency and the mode duration T2-T1.

The response penalty rp is calculated as a function of (T2-T0) and tr1.

In a preferred embodiment,

((T2-T0) If Tr1) is true, rp = 16-16 * ((T2-T0) / tr1),

In equation (14), the response time penalty rp is calculated to range linearly from a value of sixteen. At this time, the duration of the power saving mode is essentially zero for the value of zero when the response timeout exceeds the limit. Another energy penalty formula is possible which gives different weights to the penalties. However, equation (14) is quite simple and gives a good estimate of the energy impact. In practice, the values of the parameters in formula (14) need not be exact, and the formula is performed with integer calculation. Integer calculation has 16 penalty levels that provide sufficient resolution.

The energy penalty ep and the response time penalty rp are combined to produce an energy / response penalty erp. In a preferred embodiment, only the larger of the two penalties is used as the energy / response penalty erp. This simplifies the calculation because only penalties with a larger time range are calculated.

However, it is desirable to calculate the energy / response penalty using both energy and response penalties, for example by using a weighted average.

On the other hand, the energy and response penalties can be calculated as frequency units. In this case, the frequency is calculated from the time values T _BE and trl according to formulas (2) and (3). Here, the energy break-even frequency fbe and the response limit frequency frl are obtained. The duration of the mode is converted into a frequency value using formulas (2) and (3). The penalty formula is:

(fmd feb) is true, then ep = 16-16 * feb / fme,

And,

(1 / fm-1 / tf If 1 / fr1) is true,

rp = 16-16 * fr1 * tf * fmd / (feb / fmd),

In a preferred embodiment, if only the larger of the two penalties is used again, only the penalty with the lower frequency limit is calculated.

Energy and response penalties are calculated when the mode ends to return to the active state. Assuming entry into a deeper mode, no penalty occurs because the return penalty is associated with the selected mode.

On the other hand, we look at another form of penalty, the lost opportunity penalty mop. Compare power saving mode with this penalty compared to the ideal behavior where no energy or response penalty is incurred. An opportunity loss penalty occurs when the power saving mode is not used (type 1), or when the threshold frequency entering the mode is too low (type 2). In opportunity, the two types are only considered if it is possible to use a time interval without incurring an energy or response penalty. In FIG. 7, the time interval MOP1 from T3 to T4 is an example of a type 1 opportunity, while the time interval MOP2 from T0 to T1 and the time interval from T5 to T4 are examples of type 2 opportunities. Like the energy / response penalty, the opportunity loss penalty can be calculated in terms of frequency or time that you think is most convenient.

First, the opportunity loss penalty is calculated in time units. For the power save mode, a loss penalty occurs when the time interval is lost. For Type 1 opportunities, the relevant time is T3, the previous time window with an access density greater than zero, and T4, the next disk access time when it has not yet entered sleep mode within the time interval. The (T4-T3) value represents the length of the time interval and is calculated from the access frequency using the following formula.

Where frequency freq is the measured access frequency. Of course, only time breaks greater than the energy break-even time and the response limits should be considered.

The formula for a Type 1 lost opportunity penalty is:

If the opportunity (T4-T3) is penalized when implemented with integer calculation, a penalty of zero is given by the above formula. The magnitude of the penalty is determined by how big the opportunity is, rather than how big the energy break-even time and response limits are. Therefore, the greater the chance, the greater the penalty, and the smaller the chance, the smaller the penalty.

For type 2 opportunities entering all 1s, the (T1-T0) value is calculated from the threshold frequency using the following formula.

Type 2 penalties are only used when a power save mode with no energy or response penalty is possible, and given an inappropriate amount of energy that is exhausted before the threshold frequency enters the mode. This is obtained by the following formula (22).

As in the case of Type 1, if the opportunity results in an energy or response penalty, no penalty occurs. In the case of Type 1, this is an absolute limit. In the case of type 2, this is an approximation since the energy penalty or response penalty depends on both type 2 opportunity and mode duration. However, formula 22 has the advantage that it is very easy to implement.

The opportunity loss penalty is also calculated in frequency units. For type 1 opportunities, a penalty occurs when the measured frequency freq is less than the frequency penalty lower limit mfl. In this case, the following formula is a very good penalty formula.

For Type 2, the formula is:

The advantage of calculating the penalties using frequency units is that the current frequency freq and the threshold frequency tf can be easily obtained. However, time units are preferred because they can be easily implemented as counters in hardware implementations, and also use time units to maintain consistency with energy and response penalties.

Of course, other penalty formulas are possible. In the above description it is assumed that no opportunity penalty is incurred if an energy or response penalty is incurred. In some cases, it is rather desirable for this opportunity penalty to occur. The penalty then has an additional weighting factor depending on the magnitude of the energy or response penalty.

When in multiple power saving modes, penalty values are preferably determined for each mode. This is because the modes have different time windows, which correspond to different parts of the access frequency spectrum. For one mode, entering other modes is used to determine the penalty. For example, penalties are checked for all modes at all time intervals or only for the most optimal mode. Moreover, it is convenient to maintain multiple penalties for each power saving mode. For example, formula (6) (8) shows the case of an even and sporadic access pattern. The set of penalties is maintained with independent dynamic adaptability to these two cases. This is implemented by knowing which type of pattern has been activated during a given time interval and calculating the penalty for that pattern.

The time history of the penalties is used to adjust parameters such as gain coefficients g1 and g2, etc. in the calculations (6) and (7) of the threshold frequency. In a preferred embodiment, as each penalty is generated, they are added to the corresponding cumulative penalty value, cumulative energy / response penalties cerp and cumulative lost opportunity penalties cmop. When one of these values reaches a predetermined trip level, the selected parameters of the threshold frequency equation are changed accordingly, and the cumulative value is reduced.

If a penalty is entered without entering a power saving mode, the corresponding cumulative penalty value is reduced by a significant amount. This has the effect of reducing the penalty over time, whereby previous penalties have a relatively small effect compared to new penalties. This process is shown in the graph in FIG. The horizontal axis is the mode interval, and each tick corresponds to a mode entry or opportunity. The vertical axis represents the magnitude cerp of the cumulative penalty. From the first penalty interval 0, the cumulative value continues to increase. Since no penalty occurred in interval 1, the cumulative penalty value decreases. The cumulative penalty value is continuously increased for new penalties, and if there is no penalty, the cumulative penalty value is decreased at a fixed rate. The new penalty size, combined with the cumulative penalty at interval 7, is sufficient to exceed the penalty trip level as shown in the graph. At this point, the gain coefficients in formulas (6) and (7) change, and the cumulative penalty value decreases to zero. The cumulative penalty is cleared when the action of adjusting the parameters is assumed to have changed. This is the moment when new measurements are required. This result can be seen at time intervals after interval 8 with few penalties, with penalties less than at the preceding type intervals.

In a preferred embodiment, the energy / response penalty selected as the larger of ep and rp is added to the cumulative penalty cerp when this penalty occurs.

The cumulative penalty is then compared to the preset trip level crpt, and the gain coefficients g1 and g2 in the threshold frequency calculation are changed. In a preferred embodiment, the penalties are detected separately for uniform and sporadic accesses, respectively, and the appropriate gain regime g1 and g2 are increased. A good value for erpt is 16. As described above, there are practical limiting values in the gain coefficients, which are used to limit the upper range in gain.

If there is no penalty, the cumulative penalty cerp decreases by a predetermined amount cerd at the end of each mode to the lower limit of zero.

Where cerd acts as the decay rate for the cumulative penalty. Typically, 2 is appropriate for the value of cerd. Larger values cause penalties at a faster rate, while smaller values cause penalties at a slower rate.

Loss of opportunity cumulative penalty cmop is calculated similar to the energy / response cumulative penalty. It is possible to treat Type 1 and Type 2 penalties as separate cumulative penalties, or apply the same cumulative penalty to them for simplicity. When an opportunity loss penalty occurs, the appropriate value of mop1 or mop2 is added to the cumulative penalty cmop.

In a preferred embodiment, the appropriate gain factors g1 and g2 are reduced when the cumulative penalty exceeds the trip level mopt. The good value in mopt is 16. As detailed above, there are practical limits to the gain coefficients, which are used to limit the lower range in gain. If there is no penalty, the cumulative penalty cmop decreases by the preset amount cmod to the lower limit of zero at the end of each mode. As in the above embodiment, 2 is appropriate for the value of cmod.

The energy / response penalty erp and the type 2 missed penalty mop2 are calculated when the mode ends. Figure 10 shows a detailed process for this. Step 350 enters from step 313 of FIG. 4 when entering the deepest mode. Step 352 enters from step 401 of FIG. 2 when access occurs and the drive is currently in power save mode. Wait for access in step 351 when entering from step 350. In step 352, a mode duration is obtained. In step 353, the energy penalty ep and the response penalty rp are calculated as described above. In step 354, it is checked whether there are penalties for the two penalties. If no penalty, go to step 356. And if there is a penalty, go to step 355. In step 355, the penalty is added to the cumulative penalty cerp. In step 358, the cumulative penalty is checked against the penalty trip level erpt. If no trip level has been exceeded, control proceeds to step 367 and the control returns to step 300 in FIG.

If the trip level is exceeded, in step 359 the gain factor is changed to an appropriate value. In step 360, the cumulative penalty is set to zero and step 367 is reached. In step 356, there is no penalty, so the cumulative penalty value is reduced. In step 357, a type 2 opportunity loss penalty is calculated using formula (24). In step 361, it is checked whether a penalty has occurred. If no penalty has occurred, at step 362 the cumulative penalty value is reduced and step 367 is reached. If a penalty has been incurred, step 363 adds the penalty incurred to the cumulative penalty. In step 364, it is checked whether the cumulative penalty has exceeded the trip level. If not, go to Step 367. If the cumulative penalty exceeds the trip level, the gain is changed to an appropriate value at step 365, and at step 366, the cumulative penalty is set to zero and step 367 is reached.

Type 1 missed penalty mopl is calculated when the access frequency is less than the frequency penalty limit. Referring to FIG. 4, comparison and calculation of the values take place as part of step 307. This process is illustrated in more detail in FIG. Step 325 is entered from step 307 of FIG. In step 326, the penalty value mopl is calculated as formula (23). In step 327, a check is made for penalties. If no penalty is found, step 328 returns to step 307. If a penalty has been incurred, step 329 adds to the cumulative penalty cmop. In step 330, the cumulative penalty is compared to the lost opportunity trip level mopt. If no trip level has been exceeded, step 328 is reached. And if the trip level is exceeded, the process proceeds to step 311 where the gain factor is changed to an appropriate value. At step 332d, the cumulative penalty is cleared, and step 328 is reached.

It is also possible to adjust the gain factors without reliably using the lost opportunity penalty. In this case, the frequency of occurrence of the energy / response penalties is used instead of the lost opportunity penalty. This applies when a certain number assumes that optimal action occurs when energy / response penalties are incurred. Too low an occurrence rate indicates that the system did not save enough energy. Thus, if the incidence rate drops below a certain level, the gain coefficients can be changed, just as a loss of opportunity penalty is incurred.

It is also possible to design another way of feeding back penalty information into the threshold frequency calculation. However, this embodiment has the advantage of being very simple to design and provides an immediate response to the behavior of the actual system. Other methods include maintaining penalty statistics, such as by histogram, and thereby providing time-weighting to penalties such as renormalizing. Moreover, it is desirable to adjust the penalty trip level, the penalty attenuation rate, and other parameters such as the penalty magnitude transform coefficients and limit values in the gain component as used in formula (14). These may be adjusted dynamically via commands from the computer 41 or when the cumulative penalty trip level is exceeded. It is also desirable to allow the various coefficients to be adjusted based on a performance / energy target command.

For disk drives, it is desirable to inform the computer how well it is meeting its specified energy / performance goals. For example, this information can be conveyed in a status command. This information allows the computer to change the energy / performance targets on a case-by-case basis and inform the user that the specified target has not yet been achieved.

On the other hand, in the case of periodic access, it is also possible to improve energy savings without affecting performance. Periodic access is quite common. For example, most word processing programs include an autosave feature that saves the currently working document on a disk drive at a user-specified time interval, such as every ten minutes. If this type of access pattern can be detected, it can enter the power saving mode more quickly and exit the mode before the foreseen access. This hides the power save mode return time from the user and actually saves more energy by entering the mode earlier. The access frequency method by itself makes it possible to naturally detect periodic accesses. In this case, the action on Very Low Frequency (vlf) access is a concern. The vlf activity can be measured from the access density, but keep in mind the target frequency range. This is a longer time range than mode return times. In order to detect autosave activity, the timing window is suitably sized to a few seconds or more. The history buffer keeps the same vlf activity as, for example, the previous 3 vlf occurrences. If a predetermined pattern is detected in the history buffer, the periodic mode is started. One simplest pattern is that the previous 3 vlf occurrence values are within a certain tolerance of one another, eg 5%. Once the periodic pattern starts, the gain factors are reduced to enter the power saving mode faster. It is assumed here that there is no performance penalty. They are restored when the periodic mode ends after the previous values have been stored. The disk drive returns to the active state when the measured vlf is within a certain tolerance of the expected vlf. The tolerance is based on statistical confidence in the vlf value and also includes a sleep mode return time that guarantees that the disk drive is ready when vlf access occurs. If the pattern is not repeated, the periodic mode ends. A simple detection mechanism for exiting the periodic mode is when access occurs within a certain tolerance of the predicted vlf, ie late or early or either.

Gain coefficients and other adjustable parameters are stored in inactive storage so that they are preserved when the disk drive is powered off. This can be accomplished by using a semiconductor storage such as FLASH RAM or by writing the values on disk. The latter case must be completed before entering the power saving mode.

Although the present invention has been shown and described above in detail with reference to the above preferred embodiments, it will be apparent to those skilled in the art that various modifications may be made to the invention without departing from the spirit and scope of the invention. And improvement can be made. It is therefore evident that the above preferred embodiments are provided as examples of the invention and are not intended to limit the invention.

Claims (34)

A recording disk with data tracks, a plurality of electrically powered components, a data controller for accessing the components for reading and writing data on the disk, and a spindle motor for rotating the disk. Disk drive components comprising: a head for writing data to or reading data from the disk; and an actuator coupled to the head for moving the head to different tracks on the disk. CLAIMS 1. A method of managing power usage in a data recording disk drive, comprising: determining a frequency of accesses to one or more of the components; Selecting an access threshold frequency from predetermined frequencies; And reducing power to one or more of the components when the determined access frequency is less than the access threshold frequency. 2. The method of claim 1, further comprising restoring power of a previously power-reduced component in response to access from the controller. 2. The disk drive of claim 1, wherein the disk drive also has servo control electronics coupled to the actuator to hold the head on a track of the disk, wherein the step of determining the frequency of accesses comprises the head on the disk. Determining the frequency of the accesses to read or write data to, wherein reducing the power comprises moving the actuator to a parking position to reduce power to the servo control electronics. How to manage power usage on a disk drive. 2. The method of claim 1, wherein determining the frequency of accesses comprises determining a frequency of accesses that cause the actuator to search across the disk, and wherein reducing the power is at a parking position for the actuator. Stopping the spindle motor by moving to. 2. The method of claim 1, wherein determining the frequency of accesses comprises estimating the access frequency from a density of accesses occurring in a time window. The method of claim 1, wherein the step of selecting an access threshold frequency from predetermined access frequencies first determines whether a standard deviation of an access frequency pattern is less than a predetermined number of minutes, and if less, the average value of the pattern. And multiplying the gain factor by. The method of claim 1, wherein the step of selecting an access threshold frequency from predetermined access frequencies first determines whether a standard deviation of the pattern of the access frequency is greater than a fraction of a predetermined mean value of the pattern, and if the Multiplying the minimum access frequency by a gain factor in the pattern. 2. The power source of claim 1, wherein the step of selecting an access threshold frequency comprises calculating an access threshold frequency from maximum and minimum access frequencies in a predetermined set of access frequencies (SET). How to manage your use. 10. The method of claim 8, wherein selecting an access threshold frequency comprises multiplying the maximum and minimum access frequencies by a predetermined gain factor. 10. The method of claim 9, further comprising adjusting the gain factor. 2. The method of claim 1, further comprising determining whether the accesses are a periodic pattern to recover power of the power-reduced component prior to a next periodic access. A recording disk with data tracks, a spindle motor for rotating the disk, a head for writing data to or reading data from the disk, and a data controller for accessing the head for reading and writing data. An actuator connected to the head to move the head to different tracks and to a dataless parking position, and a servo control connected to the actuator to hold the head on a data track during read / write of data. CLAIMS 1. A method of managing power usage in a data recording disk drive operable in at least two power saving modes with electronic circuitry, the method comprising: determining a frequency of read or write accesses; Storing a value of predetermined access frequencies; Calculating a first access threshold frequency from the stored value of predetermined access frequencies; When the read or write access frequency is less than the first threshold frequency, moving the actuator to a parking position and reducing the power of the servo control electronics to cause the disk drive to enter into a first sleep mode operation; And managing power usage in the disk drive operable in at least two power saving modes. 13. The method of claim 12, wherein a second access threshold frequency is calculated from the stored value of predetermined access frequencies, and when the read or write access frequency is less than the second threshold frequency, the power to the spindle motor is reduced so that the And causing the disk drive to enter into a second sleep mode of operation, wherein the disk drive operable in at least two power saving modes. 14. The method of claim 13, wherein the step of reducing power to the spindle motor moves the actuator to the parking position while at the same time essentially reducing power to the spindle motor without first entering the disk drive into the first mode. And entering a second power saving mode. 2. A method for managing power usage in a disk drive operable in at least two power saving modes. 13. The method of claim 12, wherein calculating the first access threshold frequency comprises calculating an access threshold frequency from the maximum and minimum access frequencies at a predetermined set of access frequencies. A method of managing power usage on a disk drive capable of operating in modes. 16. The method of claim 15, wherein calculating the first access threshold frequency comprises multiplying the maximum and minimum access frequencies by a gain factor. How to manage it. 17. The method of claim 16, further comprising adjusting the gain factor. 18. The method of claim 17, wherein adjusting the gain factor selects the gain factor within a predetermined range having a first limit value representing a disk drive hibernation effect and a second limit value representing a maximum performance of the disk drive. Changing as a coefficient of performance with possible values. A method of managing power usage in a disk drive operable in at least two power saving modes. A recording disk with data tracks, a spindle motor for rotating the disk, a head for writing data to or reading data from the disk, a data controller for accessing the head for reading and writing data; An actuator connected to the head to move the head to different tracks and to a data-free parking position, and a servo control electronics connected to the actuator to hold the head on the data track during reading / writing of data. CLAIMS What is claimed is: 1. A method for managing power usage in a data recording disk drive having an entry and exit operation in a power saving mode, the method comprising: storing previously read or write accesses; Calculating an entry time into a power saving mode that is changed according to previously stored accesses after the most recent access from the stored previous accesses; Selecting a performance factor having a predetermined value within a range of a first limit value representing a maximum power saving effect on the disk drive and a second limit value representing the maximum performance of the disk drive; Adjusting the mode entry action of the disk drive as the selected performance factor using the selected performance factor to change a mode entry time calculation. . 20. The method of claim 19, wherein storing comprises storing values of previously determined access frequencies, and wherein calculating the entry time to power save mode comprises access threshold frequencies from values of previously determined access frequencies. Calculating and entering the power saving mode when the read or write access frequency is less than the threshold frequency. 21. The method of claim 20, wherein calculating the access frequency comprises using a gain factor. 22. The data of claim 21 wherein the step of using the selected performance factor to modify the mode entry time calculation comprises adjusting the gain factor by multiplying the gain factor by the selected performance factor. How to manage power usage on a recording disk drive. 20. The method of claim 19, wherein upon entering the power saving mode the actuator is moved to the parking position and power to the servo control electronics is reduced. 20. The method of claim 19, wherein upon entering the power saving mode the actuator is moved to the parking position and the spindle motor is stopped. 20. The disk drive of claim 19, wherein the disk drive may enter and exit a second power saving mode with less power consumption than the first power saving mode, and may vary according to the previously stored accesses from the stored previous accesses. Calculating an entry time into a second power saving mode that is different from the first power saving mode entering time entering the first power saving mode; In which the second mode entry action of the disc drive is adjusted as the selected performance factor using the selected performance factor to change a second mode entry time calculation. How to manage your power usage. 26. The method of claim 25, further comprising causing the disk drive to enter the second sleep mode without first entering the first sleep mode. . 20. The method of claim 19, further comprising determining whether the stored read or write accesses exhibit a periodic pattern to terminate the power saving mode prior to the next periodic access. How to. A recording disk with data tracks, a spindle motor for rotating the disk, a head for writing data to or reading data from the disk, and a data controller for accessing the head for reading and writing data. An actuator connected to the head to move the head to different tracks and to a dataless parking position, and a servo control connected to the actuator to hold the head on a data track during reading / writing of data. CLAIMS 1. A method for managing power usage in a data recording disk drive comprising an electronic circuit portion and capable of entering and exiting a power saving mode, the method comprising: determining a frequency of read or write accesses; Storing the value of the previously determined access frequencies; Calculating an access threshold frequency from the value at which the previously determined access frequencies are stored; Entering the power saving mode when the read or write access frequency is less than the threshold frequency; Detecting the power saving mode entry and exit times; If the sleep-time duration from the detected sleep-entry and end-times is usually greater than the break-even time of the preset energy, which is the sleep-time duration required to save disk drive energy, the same as the disk drive energy required to exit sleep mode. Calculating an energy penalty if less; Accumulating the calculated energy penalties; Changing the access threshold frequency when the cumulative energy penalties exceed a preset trip level such that the entry of the disk drive to the power saving mode is dynamically adjusted from a history of energy penalties. To manage power usage on disk drives for Windows. 29. The method of claim 28, further comprising: detecting times of said read and write accesses; If the time interval between successively detected access times without entering power saving mode is greater than the energy break-even time, calculating an opportunity loss penalty; Accumulating the calculated lost opportunity penalties, and changing the access threshold frequency when the cumulative lost opportunity penalties exceed a predetermined trip level so that the entry of the disk drive into the power saving mode is a history of lost opportunity penalties. The method further comprises dynamically adjusting from the method of managing power usage in a disk drive for data recording. 29. The data record of claim 28, wherein calculating the access threshold frequency comprises calculating the access threshold frequency using gain coefficients from the maximum and minimum access frequencies in a set of stored access frequencies. To manage power usage on disk drives for Windows. 29. The method of claim 28, wherein calculating the access threshold frequency comprises adjusting a gain factor. 29. The method of claim 28, wherein upon entering the power saving mode the actuator is moved to the parking position and power to the servo control electronics is reduced. 29. The method of claim 28, wherein upon entering the power saving mode the actuator is moved to the parking position and the spindle motor is stopped. 29. The method of claim 28, further comprising determining whether the accesses are a periodic pattern and thereby exiting the power saving mode prior to the next periodic access. .