JP5345990B2 - Method and computer for processing a specific process in a short time - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To process a specific process in a short time. <P>SOLUTION: A CPU package 11 includes a plurality of logical processors 61, 63, 65, and 67 capable of performing clock boosting temporarily under predetermined conditions. A control program 101 sets one of the plurality of logical processors through an ACPI interface 170 so as to perform the clock boosting when the predetermined conditions are met. The control program selects a priority process to be performed in a short time out of a plurality of processes. The control program schedules the priority process to be executed only by the logical processor for which the clock boosting is set. <P>COPYRIGHT: (C)2012,JPO&amp;INPIT

Description

  The present invention relates to a technique for processing a specific process in a short time, and more particularly to a technique for processing a specific process in a short time in a multi-processor environment capable of clock boosting.

  A multi-task operating system (OS) gives a thread a quantum (time slice) that is a time during which a central processing unit (CPU) can be used continuously at a time. The thread dispatched to the CPU subtracts a certain value from the given quantum every clock cycle, and when the given quantum reaches 0, it is placed at the end of the runnable queue and registered in the runnable queue instead. Other threads are dispatched. At this time, the scheduler compares the priority level of each thread to determine a thread to be dispatched. Therefore, the higher the priority level, the longer the CPU allocation time within a predetermined time and the earlier the processing is completed.

  The process can be divided into a foreground process that creates the window that the user is operating on and a background process that is all other processes. Users want the system to respond more quickly to foreground processes than to background processes. The OS can operate the scheduler so that the priority level of the foreground process can be increased and processing can be performed in a shorter time than the background process. In addition, a program for generating a process can operate the scheduler to increase the priority level of a specific process.

  However, a process whose priority level has been increased only increases the CPU allocation time relative to the priority level of other processes, and the priority of background processes registered in the executable queue. If the level is equal or higher, the foreground process allocation time does not increase. In addition, when the priority level of other processes is low, the CPU is monopolized by the foreground process whose priority level has been raised, and so-called starvation that the time until the CPU is assigned to another process becomes longer. A state problem occurs. The reduction in processing time due to the operation of the process priority level is effective only in relation to other processes, so there is no way to ensure that foreground processes are processed in a short time without damaging the system. Not enough.

  There is a technique called time slice boost or quantum boost as a method of giving a foreground process more quantum than a background process and processing in a short time. In time slice boost, when a window held by a process is switched to the foreground, the priority level of the process is increased to increase the time slice. However, time-slice boost is only given more quantum than other processes on the CPU, is an improvement in response time relative to background processes, and is an absolute foreground process. It does not lead to improved response time. For example, if there is only one thread in the executable queue, the processing time is determined by the operating frequency of the CPU that executes the thread and cannot be shortened beyond that.

  Patent Document 1 discloses a technique for further improving performance and reducing power consumption in a multiprocessing environment. The document rewrites the CPU affinity mask for a given task using a system power state filter that indicates which thread unit has transitioned to a power state with a short recovery time, and the scheduler A predetermined task is assigned to a thread unit that has transitioned to a power state with a shorter return time.

  Patent Document 2 discloses a technique for improving performance by selecting operating points of processor voltage and frequency based on statistical prediction of threads collectively running in all cores in a multi-core processor. To do. Specifically, the number of cores that are simultaneously active is limited to improve the operating frequency of the core that operates while complying with the power constraint conditions as a whole.

  Non-Patent Document 1 states that when any core of a multi-core processor is in an idle state and is operating below the power, current and temperature specification limits, A technique for operating at a frequency higher than the operating frequency is disclosed. Non-Patent Document 2 discloses a hard affinity technique for optimizing performance by assigning a specific process to a specific processor.

US Patent Application Publication No. 2009/0320031 Special table 2007-535721 gazette

"Intel Turbo Boost Technology in Intel Core Microarchitecture (Nehalem) Based Processors", [online], November 2008, [searched on August 10, 2010], Internet <URL: http://www.intel.co.jp/ jp / technology / turboboost /〉 Eli Dow, “Managing Processor Affinity”, [online], September 29, 2005, [Search August 10, 2010], Internet <URL: http://www.ibm.com/developerworks/jp / linux / library / l-affinity />

  In an OS corresponding to a multiprocessor system, a plurality of processes can be executed simultaneously using a plurality of CPUs. In the target multiprocessor system, the OS treats each logical CPU equivalently for each process. The OS associates a process with each CPU so that the overall efficiency is highest during scheduling. Such an association between a specific process and the CPU is called processor affinity or processor affinity. There are two methods for processor affinity: soft affinity, which schedules threads and processes with the same condition to be assigned to the same CPU as much as possible, and hard affinity, which is actively assigned to a specific CPU.

  According to soft affinity, L2 cache data and memory data used in the NUMA (Non-Uniform Memory Access) architecture can be used effectively, so the OS generally adopts soft affinity for threads. The same CPU as that executed last is assigned. On the other hand, adopting hard affinity obviously adopted soft affinity, such as when testing system performance or executing an application that requires processing with a particularly high cache hit rate. This is limited to cases where it is known that a performance superior to that of the OS algorithm can be expected.

  By the way, in the multi-core CPU, Intel has introduced turbo boost technology that operates other core CPUs at a frequency higher than the reference operating frequency when an idle core CPU exists. In Turbo Boost Technology, the operating frequency of some CPU cores rises when conditions such as temperature or power consumption are met, but the scheduler uses a predetermined algorithm without recognizing the operating frequency of the CPU core. Schedule the process. Therefore, when the turbo boost technology is adopted, an arbitrary process is assigned to the CPU core whose operating frequency increases, and only the process that is accidentally assigned to the CPU by the scheduler is executed at a high operating frequency. Therefore, the processing time of the foreground process cannot be reliably shortened only by adopting the turbo boost technology.

  Therefore, an object of the present invention is to provide a method for reducing the processing time of a specific process in a multi-processor environment. It is a further object of the present invention to provide a method that increases the chances of executing a foreground process on a logical CPU having a higher operating frequency than a background process. A further object of the present invention is to provide a computer program and a computer that realize such a method.

  The computer according to the present invention includes a plurality of logical processors capable of clock boosting that temporarily increases the operating frequency when a predetermined condition is satisfied. The control program sets the clock boost to one of the plurality of logical processors. There may be one or more logical processors for setting the clock boost. The plurality of logical processors may be realized by a plurality of processors, or may be realized by virtualizing one processor with software. The logic processor for which the clock boost is set operates at an operating frequency higher than the basic operating frequency only while the predetermined condition is satisfied, and returns to the basic operating frequency when the predetermined condition is not satisfied. An operating frequency higher than the basic operating frequency can be set in stages.

  The predetermined condition can be set based on the presence of an idle logical processor, power consumption, current consumption, and temperature. In addition, if multiple logic processors continue to have low power consumption, there are no idle logical processors, current exceeds the reference value, or power consumption exceeds the reference value. Even if it exists, it may operate at a high operating frequency as long as the temperature does not exceed the reference value. When the processor is operated in a state where the power consumption temporarily exceeds the power consumption reference value in this way, the operating frequency can be further increased, so that the process can be processed in a shorter time.

  The control program selects, as a priority process, a process that is required to be processed in a short time from among a plurality of processes generated by the computer. The priority process may be a process for generating a window, a process for which high scheduling priority is set, or a process for which high disk I / O processing priority is set. In cooperation with the OS, the control program dispatches the priority process to only one or a plurality of logical processors for which clock boost is set and executes the priority process.

  A logical processor with a clock boost setting will increase its operating frequency when the condition is met, so any process executed there will have a shorter processing time than an unconfigured logical processor, and a priority process will be dispatched. Waiting time is shortened. Furthermore, since the priority process is scheduled only to the logical processor set with the clock boost, the probability that it is executed by the logical processor in the actual clock boosted state is increased, and the processing time is further reduced. Unlike the method of changing the priority level of the process, the method of the present invention uses the characteristic that the logical processor temporarily operates at a high operating frequency, so that the absolute processing time of the priority process is shortened. Can do. Also, other processes with lower priority levels are not starved.

  In order to schedule based on the affinity with the logical processor with the clock boost setting, the control program can create an original mask such as an affinity mask created by the OS or an affinity mask rewritten by the program that generates the process. The bit pattern of the affinity mask can be rewritten based on the setting state of the clock boost. Specifically, the affinity mask can be rewritten by performing an AND operation on the original affinity mask and a mask bit indicating the setting state of the clock boost for each of the plurality of logical processors.

  By performing an AND operation, it is possible to eliminate a situation in which the priority process is preferentially executed by a logical processor other than the logical processor designated by the program. If any logical processor is found to be equivalent to the process, the affinity mask can be rewritten to specify only the logical processor with the clock boost setting. Good. By rewriting in this way, the priority process can be executed only by the logical processor in which the clock boost is set reliably.

  The method of the present invention can be applied to a newly generated process each time. The control program obtains the usage rate of the logical processor for which clock boost is set, and the affinity mask is set so that the newly generated priority process is preferentially executed only when the usage rate is less than the predetermined value. Can be rewritten. When the usage rate is equal to or higher than the specified value, it is faster to relax the affinity so that the process scheduler dispatches not only logical processors with clock boost settings but also unconfigured logical processors. This is because the probability of being able to be processed may increase.

  Further, the control program can acquire the temperature of the logical processor, and the affinity mask can be rewritten so that the newly generated priority process is preferentially executed only when the temperature is lower than a predetermined value. When the logic processor is hot, the process scheduler runs only on the logic processor with the clock boost setting because the probability of actually operating at a high operating frequency is low even if the clock boost setting is set. This is because the probability of processing in a shorter time may be higher when a logical processor to be executed including other logical processors is selected.

  A logical processor in which clock boost is set also executes a non-priority process, and a load applied to the logical processor by each process also changes with time. When the usage rate of the logical processor for which clock boost is set is greater than or equal to a predetermined value, it can be determined whether or not there is a logical processor that can be further set for clock boost. Then, when the usage rate of the logical processor with the clock boost setting exceeds a predetermined value and there is a logical processor capable of setting the clock boost, an additional clock boost is added to the new logical processor. The load can be distributed by setting.

  At this time, rewrite the affinity mask so that the priority process that is preferentially executed and / or the priority process that is not preferentially executed are dispatched including the logical processor that is additionally set with clock boost. Can do. Some priority processes are not preferentially executed due to conditions such as the usage rate of the logical processor or the temperature of the logical processor when it is created. If there is a priority process that is not preferentially executed when the usage rate of a logical processor for which clock boost is set falls below a predetermined value, the affinity mask is rewritten so that the priority process is preferentially executed. Can do. Therefore, it is possible to preferentially execute a priority process that has not been preferentially executed once under the conditions of use rate and temperature when the process is generated, based on the subsequent operation state of the logical processor.

  When the number of logical processors for setting clock boost is increased, the affinity of the priority process with respect to the logical processor that actually performs the clock boost is reduced, and the difference from the non-priority process is reduced. Thus, there is a substantial limit to the number of logical processors that can be set for clock boost among the overall logical processors. Therefore, when the usage rate of the logical processor for which the clock boost is set increases to a predetermined value or more, there may be no logical processor that can additionally set the clock boost. At this time, if the usage rate of the logical processor for which clock boost is not set is less than the predetermined value, the control program rewrites the affinity mask so that the priority process is executed even for the logical processor for which clock boost is not set. Process affinity can be relaxed.

  The present invention has provided a method for reducing the processing time of a specific process in a multi-processor environment. Furthermore, the present invention can provide a method for increasing the chance that a foreground process is executed by a logical CPU having a higher operating frequency than a background process. Further, according to the present invention, it is possible to provide a computer program and a computer that realize such a method.

It is a functional block diagram which shows the main hardware constitutions of the computer system concerning this Embodiment. It is a functional block diagram which shows the structure of the hardware and software of the priority process processing system implement | achieved in a computer. It is a flowchart which shows the procedure which performs a priority process in a priority process processing system. It is a flowchart which shows the procedure which performs a priority process in a priority process processing system. It is a flowchart which shows the procedure which performs a priority process in a priority process processing system. It is a flowchart which shows the procedure which performs a priority process in a priority process processing system. It is a figure explaining the data structure of an affinity mask and a performance filter, and the method of an AND operation. It is a figure which shows an example of the data structure of a process list.

[Computer system configuration]
FIG. 1 is a functional block diagram showing a main hardware configuration of a computer system 10 according to the present embodiment. The computer system 10 includes a CPU package 11, a main memory 13, an input device 15, a hard disk drive (HDD) 17, a liquid crystal display (LCD) 19, a BIOS_ROM 21, and the like connected to the bus 23. Yes. The CPU package 11 is a multi-core processor in which a plurality of CPU cores are housed in one package, and supports Intel's turbo boost technology as an example.

  Turbo Boost Technology is within the range of thermal design power consumption (TDP) while improving the performance by temporarily increasing the operating frequency of any CPU core when CPU package 11 has thermal margin. It is a method of storing. In the CPU package 11, the OS requires a temporary increase in performance, and the average value of power consumption and current measured during the most recent predetermined time is below a predetermined value such as TDP, and the temperature of the CPU package is below a predetermined value. In some cases, the operating frequency of any of the CPU cores is increased by capturing some opportunity. When the condition is not satisfied, the operating frequency of the CPU core is returned to the basic operating frequency.

  Such a temporary increase in the operating frequency of the CPU core is referred to as clock boost, and power consumption, current consumption, and temperature for clock boost are referred to as boost conditions. When performing clock boost, a specific CPU core can gradually increase the operating frequency, or the number of CPU cores to be clock boosted can be increased. When returning to the basic operating frequency from the clock-boosted high operating frequency, the reverse procedure can be used.

  The trigger for clock boost includes a method of detecting an idle state of some CPU cores, a method of using a change in environmental temperature, and a method of using a change in platform temperature. In addition, Intel's new generation of turbo-boost technology will temporarily make TDP when the power consumption or temperature of the platform surrounding the CPU package or the CPU package is low, unless the power consumption or temperature reaches the upper limit. Increase the operating frequency beyond.

  With these new generations of turbo boost technology, if you run high-load applications when the platform power consumption and temperature are low, the actual power consumption and temperature will be used as a reference using the previous headroom. Since it can be operated at a higher operating frequency than the previous turbo boost technology only for a short time until it rises to the value, there is an advantage that the response time can be greatly shortened. Both old and new generation turbo boost technologies require a specific process to be scheduled on a CPU core that always clock boosts in order to use that specific process for a short time. .

  Note that the present invention is not limited to multi-processor systems that support Intel's turbo boost technology, but can be applied to other multi-processor systems that can perform equivalent clock boost. is there. Further, the present invention can be applied to a case where a plurality of single core CPU packages are provided in one computer system and only a part of the CPU packages are clock boosted. In addition, in the technology called Intel's Hyper-Threading (trademark), one processor core is caused to operate like a plurality of processor cores with respect to the OS. The present invention can also be applied to such virtualized multiprocessors.

  Furthermore, the present invention can be applied to a system that realizes a multi-processor environment by virtualizing with software such as Intel (registered trademark) VT of Intel Corporation or AMD-V of AMD Corporation. Therefore, hereinafter, the multi-processor unit that is the target of clock boosting will be referred to as a logical CPU in the sense that it includes both hardware and software. The BIOS_ROM 21 stores a BIOS that conforms to the ACPI standard.

[Priority process]
In this embodiment, a foreground process (hereinafter referred to as a priority process) in which a process or a thread that is an execution unit of the process has a high priority and a high-speed process is required is referred to as another background process (hereinafter referred to as a non-priority process). And so on. All processes have at least one thread, and one or more threads are executed as scheduling targets for the CPU package 11. Therefore, in the case of execution of a process, it means that a thread included in the process is executed, and a setting for a process also means a setting for a thread when it is inherited by a thread.

  It is common for a user using a computer system to expect a short response time for the user interface process. Thus, the first group of priority processes is defined as the user interface process that creates the window and actually displays it on the LCD 19. Of the user interface processes, the process displaying the foreground window displaying the window in the foreground can be defined as a priority process having a higher priority.

  The OS scheduler selects a thread to be dispatched to the logical CPU based on the priority level of the thread. The priority level of a thread is determined by the priority class of the process and the relative priority assigned to the threads in the process. A program can set a priority class when creating a process and set a relative priority when creating a thread. Since a thread with a high priority level is dispatched with a higher priority than a thread with a low priority level, the allocation time of a logical CPU in a predetermined time becomes longer and can be processed in a shorter time. Since the user expects a thread with a high priority class to be processed in a short time, the second group of priority processes is defined as a priority class or a process with a high priority level.

  As the operating frequency of the processor increases, the disk I / O may become a bottleneck. In order to solve this problem, for example, in Windows Vista (registered trademark), a process such as an anti-virus program, a defragmentation program, or a mail sending / receiving program running in the background is not affected. The I / O processing priority of the background process can be lowered. Therefore, the third group of priority processes is defined as a thread with a high disk I / O priority. Furthermore, a process such as a compiler or an anti-virus program that does not belong to any of the first group to the third group but needs to be a target of a priority process due to individual circumstances of the user is set as a priority process of the fourth group Define.

[Configuration of priority process processing system]
FIG. 2 is a functional block diagram showing a hardware and software configuration of the priority process processing system realized in the computer 10. The priority process processing system 100 can be mainly composed of a control program 101, an OS 150, an ACPI interface 170, and a CPU package 11. However, it is necessary to use other devices such as the main memory 13 and the bus 23 in order for the software to be executed by the CPU package 11 and exhibit functions, but is not particularly necessary for the description of the present invention. Excluded from FIG.

  The CPU package 11 includes four logical CPUs 61, 63, 65, 67 and a temperature sensor (not shown) for detecting the temperature of the die. The CPU package 11 can receive data and a P state regarding the boost condition of the CPU package 11 from the ACPI interface 170, and can autonomously control clock boost for one or more logical CPUs with a predetermined algorithm. . That is, the clock boost control by the CPU package 11 is performed independently from the scheduling by the OS 150. Further, the OS 150 does not recognize which of the CPU packages 11 is a logical CPU for which clock boost is set and whether or not the clock boost is currently actually performed.

  The control program 101 performs independent processing in cooperation with the OS 150 in order to process the priority process in the CPU package 11 in a short time. Details of the processing will be described in the priority processing procedure according to the flowcharts of FIGS. The white list 103 records an identifier of a program that generates a fourth group priority process that cannot be specified by the control program 101. Alternatively, the control program 101 does not specify the priority processes of the first group to the third group, and the identifiers of programs that generate all priority processes may be registered in the white list. The user can register such program identifiers in the white list 103 in advance through an interface provided by the control program 101.

  The performance filter 105 is a 4-bit mask bit corresponding to the number of logical CPUs 61, 63, 65, and 67, and registers the identifier of the logical CPU to be clock-boosted when the boost condition is satisfied. When starting the operation, the control program 101 calls a system call function to obtain a valid logical CPU identifier from the OS 150 or the ACPI interface 170, and selects some of the logical CPUs as targets for clock boosting. Then, a bit pattern is set in the performance filter 105. By setting the P state in the ACPI register corresponding to the logical CPU set as the clock boost target in the performance filter 105 by the control program 101, the clock boost setting for the logical CPU becomes valid.

  The logical CPU that is the target of clock boosting temporarily increases its operating frequency only when the boost condition is satisfied, but the multiple logical CPUs that are set to clock boost are actually all at the same time. It is not guaranteed to do. As the number of logical CPUs for which clock boost is set increases, the number of logical CPUs for which the time ratio for actual clock boosting decreases increases. A priority process that is assigned such a logical CPU cannot benefit from clock boost. If clock boost is set for all logical CPUs, there is no difference in processor affinity between the priority process and the non-priority process.

  Accordingly, it is necessary to appropriately select the ratio of the number of logical CPUs for setting the clock boost to the total number of logical CPUs. As an example, the control program 101 first selects 50% of the total number of logical CPUs or the minimum number of logical CPUs as the target of clock boosting as a default, and then selects the logical CPUs targeted for clock boosting. It is possible to set an optimum clock boost by adding when the CPU usage rate is equal to or higher than a predetermined value or decreasing when the CPU usage rate is lower than the predetermined value.

  The process list 107 registers the bit pattern of the affinity mask created for the priority process. The control program 101 calls a system call function to enumerate a process created at that time or hook a newly created process to determine whether each process is a priority process. Further, the bit pattern of the affinity mask is registered in the process list 107.

  The control program 101 stores, in the process list 107, the bit pattern of the affinity mask set by default by the OS 150 for the priority process and the bit pattern after the program that generated the process rewrites the default affinity mask. sign up. Hereinafter, these two types of affinity masks are referred to as original affinity masks. Furthermore, a bit pattern after the control program 101 rewrites the original affinity mask is also registered in the process list 107.

The CPU usage rate meter 151 measures the CPU usage rates of the logical CPUs 61, 63, 65, and 67. The CPU usage rate Y can be calculated by the following program using the following equation, where Ui is the user mode time of the idle process, Ki is the kernel mode time of the idle process, and Et is the elapsed time.
Y = (1− (Ui + Ki) / Et) × 100%

  The process scheduler 153 determines logical CPUs 61, 63, 65 and 67 to which threads are assigned with a predetermined algorithm based on the affinity mask 155 and dispatches them from the executable queues 161, 163, 165 and 167. When creating a process, the program can set the priority class of the process, and can further set the relative priority of the threads. The OS 150 determines the priority level of the thread based on the priority class and the relative priority. The process scheduler 153 determines the order in which threads are dispatched based on the priority level of each thread.

  The affinity mask 155 is a mask bit generated by the OS 150 for each process, and the number of bits corresponds to the number of logical CPUs 61, 63, 65, and 67. The affinity mask 155 may be an original one or may be rewritten by the control program 101. The bit pattern of the affinity mask 155 indicates the processor affinity of the process and specifies the logical CPU on which the process is executed. The bit pattern of affinity mask 155 generated for a process is inherited by threads belonging to that process. Therefore, the affinity mask 155 corresponds to a process and a subordinate thread at the same time. The affinity mask 155 and priority level become part of the data structure that is to be scheduled by the process scheduler 153 registering in the executable queue.

  The process scheduler 153 determines a logical CPU to dispatch for each thread based on the bit pattern of the affinity mask 155. When a plurality of bits are set in the affinity mask 155, the process scheduler 153 dispatches to any of the set logical CPUs based on a predetermined algorithm such as a round robin method. In the executable queue 157, a data structure of a thread that is in an executable state is registered. Each executable queue 161, 163, 165, 167 corresponds to each logical CPU 61, 63, 65, 67.

  The process scheduler 153 registers the data structure of the thread to which a predetermined quantum prepared by default is added in the executable queue 157. Further, it is registered in one of the executable queues 161, 163, 165, and 167 corresponding to the logical CPU designated by the bit pattern of the affinity mask 155. The thread registered in one of the executable queues 161, 163, 165, and 167 is eventually dispatched to the corresponding logical CPUs 61, 63, 65, and 67 according to the priority level with other threads.

  For example, when a thread dispatched to the logical CPU 61 consumes a given quantum, it is registered at the end of the corresponding executable queue 161. Further, when a thread having a high priority level is registered in the executable queue 161 in the middle of execution, it is preempted and registered at the head of the executable queue 161 even before all the quantum is consumed. A thread that has entered a standby state due to I / O waiting or the like is registered again in the executable queue 157, and then passes through the executable queues 161, 163, 165, 167, and any one of the logical CPUs 61, 63, 65, 67. To be dispatched.

  The ACPI interface 170 includes a BIOS and an ACPI register (not shown). The BIOS can control the clock boost of the CPU package 11 by setting the P state in the ACPI register. The BIOS sets the three objects of _PSS (Performance Supported States), _PSD (P-State Dependency), and _PPC (Performance Present Capabilities) in the ACPI register, thereby setting a predetermined logical CPU to a predetermined operating frequency. You can boost it.

  In one example, a logical CPU in the P0 state in which _PPC is set to 0 operates with the operating frequency set to _PSS as the upper limit operating frequency by clock boosting when the boost condition is satisfied. A logical CPU in the P1 state in which 1 is set to _PPC operates with the basic operating frequency as the upper limit operating frequency without clock boosting even if the boost condition is satisfied. The control program 101 sets the bit pattern of the performance filter 105 as the P state in the logical CPU to be clock boosted through the ACPI register of the ACPI interface 170. The ACPI interface 170 further provides the CPU package 11 with data such as power consumption, current, and temperature necessary for determining the boost condition.

  Each software shown in FIG. 2 is stored in the HDD 17 and loaded into the main memory 13 at the boot stage. Each software can be regarded as hardware that, when executed mainly by the CPU package 11 and the main memory 13, causes the computer 10 to perform a predetermined function in cooperation with the hardware. The OS 150 is not particularly limited as long as it corresponds to a multi-processor system and a multi-task. The present invention can also be realized without modifying existing OS and clock boost algorithms.

[Procedure for processing priority processes]
3, 4, 5, and 6 are flowcharts illustrating a procedure for executing a priority process in the priority process processing system 100. FIG. 3 shows a procedure for processing a process generated until the control program 101 starts an operation and a newly generated process. FIG. 4 shows a procedure for dynamically changing the clock boost setting in accordance with the CPU usage rate of the logical CPU for which the clock boost setting is set.

  FIG. 5 shows a procedure for preferentially executing a priority process that is not preferentially executed according to a subsequent decrease in the CPU usage rate under the conditions of the CPU usage rate and temperature when the process is generated. FIG. 6 shows a procedure for relaxing the processor affinity of the priority process when the CPU usage rate of the logical CPU for which the clock boost is set increases. FIG. 7 is a diagram for explaining the data structure of the affinity mask 155 and the performance filter 105 and the logical product operation method.

  In block 201, the computer 10 is activated and the program stored in the HDD 17 starts to boot. When the boot file is sequentially loaded into the main memory 13 and execution is started, each program generates a process at block 203. The OS 150 or the ACPI interface 170 checks a valid logical CPU mounted on the CPU package 11. Here, it is assumed that all four logical CPUs 61, 63, 65, and 67 are enabled. The OS 150 sets the _PSS, _PSD, and _PPC methods held by default in the logical CPUs 61, 63, 65, and 67 through the ACPI interface 170.

  In the default P state, for example, the OS 150 sets only the logical CPU 61 to the P0 state, and sets the other logical CPUs 63, 65, and 67 to the P1 state. In this case, the CPU package 11 raises only the operating frequency of the logical CPU 61 above the basic operating frequency if the boost condition is satisfied, and returns to the basic operating frequency when the boost condition is not satisfied. The OS 150 creates a default affinity mask 155 for each of the generated processes.

  Each bit of the affinity mask 155 is set to 1 so that the default affinity mask 155 can be dispatched to any valid logical CPU 61, 63, 65, 67. This state is shown as an affinity mask 155a in FIG. Further, a program for generating a process can rewrite the affinity mask 155 set by default by calling a system call function.

  As an example, a state where the program rewrites the default affinity mask 155 so that only the logical CPUs 65 and 67 are assigned is shown as an affinity mask 155c in FIG. The affinity masks 155a and 155c all correspond to the original affinity mask. The process scheduler 153 schedules not a process but a thread generated as a process execution unit by a program. The process affinity mask 155a is inherited by the thread affinity mask generated inside the process. The thread generating program and the control program 101 can also rewrite the thread affinity mask.

  The process scheduler 153 can freely assign threads to any logical CPU 61, 63, 65, 67 based on the affinity mask 155a. Process scheduler 153 can also assign threads to either logical CPU 65 or logical CPU 67 based on affinity mask 155c. However, the scheduler 153 can also schedule based on a so-called soft affinity method so that a thread having the same condition is assigned to the last executed logical CPU so that the thread does not frequently move between logical CPUs.

  The CPU package 11 controls the operating frequency of the logical CPU 61 independently of the process scheduler 153. If the logical CPU 61 is accidentally clock boosted while the thread to which the logical CPU 61 is assigned is registered in the executable queue 161, the thread waits shorter than the threads registered in the executable queues 163, 165, and 167. Dispatched on time. Further, the thread dispatched to the logical CPU 61 is processed in a shorter time than other threads registered in the executable queue 161 if the logical CPU 61 happens to be clock-boosted at that time. Any thread that has an affinity mask 155a has an opportunity to be dispatched to the clock-boosted logical CPU 61, but the probability is 1/4, and when dispatched to the logical CPU 61, the logical CPU 61 is clock-boosted. Not always. Up to this point, the process control procedure is based on the conventional method.

  The control program 101 also constitutes a part of the boot file, and the control program 101 starts operation at block 205. Since the control program 101 starts its operation at the initial stage of booting, the priority process of the boot file can be processed in a short time in the subsequent boot stage. For example, if it is desired to operate an application relating to mail as soon as possible after activation to receive mail, the process generated by the application can be selected as a priority process and executed as soon as possible during booting. Alternatively, the control program 101 may start operation after the final stage of booting or after booting is completed.

The control program 101 sets bits corresponding to the logical CPUs 61 and 63 to 1 and sets bits corresponding to the logical CPUs 65 and 67 to 0 in the performance filter 105. The control program 101 sets the P state in the ACPI interface 170 based on the performance filter 105 in block 207. The BIOS sets the logical CPUs 61 and 63 to the P0 state, and sets the logical CPUs 65 and 67 to the P1 state. The performance filter 105 at this time is shown as the performance filter 105a in FIGS. 7A and 7B.

  As a result, in the CPU package 11, the logical CPUs 65 and 67 operate with the basic operating frequency as the maximum operating frequency, and the logical CPUs 61 and 63 operate at an operating frequency higher than the basic operating frequency when the boost condition is satisfied. When not established, the basic operating frequency is set to the maximum operating frequency. When the boost condition is satisfied, the CPU package 11 may clock-boost the logical CPUs 61 and 63 at the same time, or clock-boost only the logical CPU 61 first, and if there is a thermal margin, the logical CPU 63 The operating frequency may be controlled to additionally boost the clock. Each logical CPU 61, 63, 65, 67 may operate at an operating frequency lower than the basic operating frequency when in an idle state.

  In block 209, the control program 101 calls a system call function to identify a priority process from among the processes created at that time. A priority process that initially enumerates all currently created processes to obtain process handles, and in which the process that creates the window and actually displays it is classified in the first group As specified. Next, the control program 101 calls a system call function to acquire each priority class or priority level of the remaining processes excluding the priority process of the first group, and the priority class or priority level is predetermined. Processes above the level are identified as priority processes that are classified into the second group.

  For example, in Windows (registered trademark), when a process is created, an application calls a system call function to specify a priority class of Real-time, High, Above Normal, Normal, Below Normal, or Idle. Can do. When an application creates a thread, it can call a system call function to set the relative priority of the thread. Windows (registered trademark) determines the priority level of the thread based on the specified priority class and relative priority. At this time, the control program 101 can set a process in which a priority class of Above Normal or higher is designated as a priority process classified into the second group.

  Further, the control program 101 calls the system call function to obtain the I / O processing priority of each disk of the remaining processes except the priority processes of the first group and the second group, and the I / O. A process having a processing priority of a predetermined level or higher is specified as a priority process classified into the third group. Specifically, the application can set the I / O processing priority for the disk in five stages: Critical, High, Normal, Low, and Very Low, and these data are stored in the I / O manager of the OS by the I / O manager. It is set as a flag in an IRP (I / O Request Packet) communicating with the O driver. At this time, the control program 101 can set a process specifying an I / O processing priority higher than High as a priority process classified into the third group.

  Finally, the control program 101 calls the system call function and, among the remaining processes excluding the priority processes of the third group from the first group, the priority process generated by the program registered in the white list is If so, it is identified as the fourth group priority process. Each time the control program 101 specifies a priority process, it calls a system call function to rewrite the original affinity masks 155a and 155c corresponding to the process or thread.

  For each process, the control program 101 acquires the bit pattern of the original affinity masks 155a and 155c created in the block 203, performs a logical AND operation with the bit pattern of the performance filter 105a, and FIG. , Mask bits 171a and 171b shown in FIG. 7B are calculated. In the case of FIG. 7A, the affinity mask 155b is created by rewriting the bit pattern of the original affinity mask 155a with the bit pattern of the mask bit 171a.

  As a result, since the process scheduler 153 schedules the thread generated by the process based on the affinity mask 155b, the thread is dispatched only to the logical CPU 61 or the logical CPU 63 and is not executed by the logical CPUs 65 and 67. At this time, the process scheduler 153 can schedule a thread having the affinity mask 155b between the logical CPU 61 and the logical CPU 63 by a round robin method. A priority process that is executed only by the logical CPUs 61 and 63 for which clock boost is set in this way is referred to as a priority process that is preferentially executed.

  The logical CPUs 61 and 63 are also dispatched with non-priority process threads that have the original affinity mask 155a. The thread of the priority process having the rewritten affinity mask 155b is dispatched from the executable queues 161 and 163 to the logical CPUs 61 and 63 on the basis of the priority level relationship with other priority processes and non-priority processes. The order is determined. However, since the logical CPUs 61 and 63 can process in a shorter time than the logical CPUs 65 and 67 when any thread is dispatched when the clock is boosted, the process up to the dispatch in the priority process executable queues 161 and 163 can be performed. The waiting time is shortened.

  In addition, since the priority process thread that is preferentially executed is dispatched only to the logical CPUs 61 and 63 that are clock-boosted, there is a high probability that the thread is actually clock-boosted when the thread is executed. For these two reasons, a priority process having processor affinity only for the logical CPUs 61 and 63 is processed in a shorter time than a non-priority process.

  In the case of FIG. 7B, if the bit pattern of the original affinity mask 155c and the bit pattern of the performance filter 105a are ANDed, all bits become 0 as in the mask bit 171b. In this case, the control program 101 does not rewrite the affinity mask 155c set by default. Therefore, such priority processes will only be dispatched to logical CPUs 65, 67 that do not clock boost. A priority process executed only by the logical CPUs 65 and 67 for which clock boost is not set as described above is referred to as a priority process that is not preferentially executed.

  As described above, the reason why the logical CPUs 61 and 63 forcibly set the clock boost is not executed even in the priority process is to prevent the logical CPU designated by the program from being changed. When it is clear that any logical CPU 61, 63, 65, 67 is equivalent to the program, the control program 101 can directly rewrite the affinity mask 155c with the bit pattern of the performance filter 105a. . In that case, whatever the bit pattern of the original affinity mask 155c, it can be dispatched only to the logical CPUs 61, 63 that always clock-boost that thread.

  In block 211, the control program 101 registers the priority process identifier and the corresponding bit pattern of the affinity mask 155 in the process list 107. The control program 101 registers each bit pattern of the original affinity mask and the affinity mask rewritten by the control program 101. FIG. 8 shows an example of the data structure of the process list 107. In FIG. 8, the original bit pattern of the affinity mask 155a rewritten by the control program 101 is omitted.

  The priority processes 1, 2, 3, and 4 indicate the bit pattern of the affinity mask of the priority process to be preferentially executed. The priority processes 1 and 2 indicate a bit pattern in which the original affinity mask created by default so that all bits are set to 1 by the OS 150, and the priority processes 3 and 4 are programs that generate processes. Shows a bit pattern in which the affinity mask created by default is rewritten. Threads under the priority processes 1 and 2 are dispatched only to the logical CPU 61 or the logical CPU 63. Threads under the priority process 3 are dispatched only to the logical CPU 61, and threads under the priority process 4 are dispatched only to the logical CPU 63.

  Therefore, the priority processes 1, 2, 3, and 4 are always executed only by the logical CPUs 61 and 63 that perform clock boosting. The priority processes 11, 12, and 13 indicate the affinity mask bit pattern in which the process generation program rewrites the affinity mask created by the OS 150 by default, and the priority processes 11, 12, and 13 are under the priority processes 11, 12, and 13. A certain thread is not preferentially executed because it is dispatched only to the logical CPU 65 or the logical CPU 67 that does not clock-boost.

  At block 213, a new process is created. The control program 101 hooks a newly generated process using DLL injection. The control program 101 calls a system call function to determine whether the process is a priority process in the same procedure as in block 209. If it is determined that the process is not a priority process, the process scheduler 153 dispatches the non-priority process based on the original affinity masks 155a and 155c, and returns to block 213 to process the next new process.

  If the control program 101 determines in block 215 that the new process is a priority process, the process proceeds to block 217. The control program 101 acquires the temperature of the CPU package 11 or each of the logical CPUs 61, 63, 65, 67 through the ACPI interface 170, and the temperature of the CPU package 11 is less than a predetermined value or all of the logical CPUs 61, 63, 65, 67. It is checked whether the temperature is lower than a predetermined value. If the temperature is lower than the predetermined value, it is determined that the temperature condition is satisfied.

  Further, the control program 101 acquires the CPU usage rates of the logical CPU 61 and the logical CPU 63 that are the targets of clock boost from the CPU usage rate meter 151 and checks whether or not they are less than the predetermined value. Determines that the usage rate condition is satisfied. If either the temperature condition or the usage rate condition is not satisfied, the control program 101 proceeds to block 221 to register the newly generated process as a priority process, and the original affinity masks 155a and 155c are not rewritten.

  The scheduler 153 dispatches threads belonging to the priority process based on the original affinity masks 155a and 155c. An example of the priority process in this case is shown as priority processes 21, 22, and 23 in FIG. As described above, the priority process 21 that can be executed by either the logical CPUs 61 and 63 in which the clock boost is set according to the temperature condition or the usage rate condition and the logical CPUs 65 and 67 in which the clock boost is not set. , 22 and 23 are also priority processes that are not preferentially executed.

  Even if the temperature condition or usage rate condition is not satisfied, if the affinity mask is rewritten so that the process generation program specifies only the logical CPUs 61 and 63, it is adopted as the original affinity mask and preferentially executed. Such a state is outside the scope of the process control of the present invention. The process scheduler 153 dispatches the threads under the priority processes 21, 22, and 23 to any of the logical CPUs 61, 63, 65, and 67. When the registration to the process list 107 is completed in block 221, the control program 101 returns to block 213 and waits for the generation of a new process.

  In the procedure of block 217, even if it is a priority process, if the CPU usage rate at that time of the logical CPUs 61 and 63 to be clock boosted is high, the priority process is not executed. The reason is that when the CPU usage rate of the logical CPUs 61 and 63 is high, many priority processes and non-priority processes or processes with a large load are executed there, so that the new priority process also executes the other logical CPUs 65 and 67. This is because the possibility of processing in a shorter time is higher in the case of inclusion first.

  Also, when the temperature is high, it is considered that the frequency of the logical CPUs 61 and 63 actually performing the clock boost is low, and similarly, the other logical CPUs 65 and 67 may be processed in a shorter time if they are included in the execution destination. Because it may be expensive. When the CPU usage rate of the logical CPU 61 for which the clock boost is set is high and the CPU usage rate of the logical CPU 63 is low, the affinity mask 155 is set so that the control program 101 assigns only the logical CPU 63 to the new priority process. You may rewrite it.

  When both the temperature condition and the usage rate condition are satisfied in block 217, the process proceeds to block 219, and the control program 101 rewrites the affinity mask 155 for the process by the method described in block 209. Further, the process is registered in the process list 107. In the procedure of blocks 217 to 221, whether or not the priority process is preferentially executed is controlled based on the temperature condition and usage rate condition of the CPU package 11 every time a new process is generated.

  Non-priority processes are also dispatched to the logical CPUs 61 and 63, and their loads also change over time. In this embodiment, since the priority class of the priority process and the priority level of the thread are not changed, if the number of threads registered in the executable queues 161 and 162 of the logical CPUs 61 and 63 increases, the CPU usage of the logical CPUs 61 and 63 is increased. There is a possibility that the waiting time until the rate is increased and the priority process is dispatched becomes long.

  In order to eliminate this state, in block 301 of FIG. 4 following block 219 of FIG. 3, the control program 101 periodically acquires the CPU usage rate of each logical CPU 61, 63, 65, 67 from the CPU usage rate meter 151. . In block 303, it is determined whether or not the CPU usage rates of the logical CPUs 61 and 63 for which clock boost is set are equal to or greater than a predetermined value. If the CPU usage rate of any of the logical CPUs is less than the predetermined value, the process proceeds to block 401 in FIG. 5, and if the CPU usage rate of any of the logical CPUs 61 and 63 is greater than or equal to the predetermined value, the process proceeds to block 305. Note that the CPU usage rate threshold of the block 303 can be made larger than the CPU usage rate threshold of the block 217.

  In block 305, the control program 101 refers to the performance filter 105 to determine whether there is a logical CPU that can be further set for clock boost. If the ratio of the number of logical CPUs that allow the setting of the clock boost to the total number of logical CPUs already exceeds the predetermined value, the clock boost cannot be set for other logical CPUs. The process proceeds to block 421 in FIG.

  For example, when the clock boost can be additionally set for the logical CPU 65, the process proceeds to block 307. In block 307, the control program 101 additionally rewrites the bit pattern of the performance filter 105 so that the bit corresponding to the logical CPU 65 becomes 1. Further, the logical CPU 65 is additionally changed to the P0 state through the ACPI interface 170 based on the rewritten bit pattern.

  In block 309, the control program 101 calls a system call function to rewrite the bit pattern of the affinity mask 155 of the priority process selected from the process list 107 of FIG. The priority process selected as the target of rewriting is a group of priority processes 1, 2, 3, and 4 that are preferentially executed, a group of priority processes 11, 12, 13, and 14 that are not preferentially executed under the affinity conditions specified by the program It can be any of the groups of priority processes 21, 22, and 23 that are not preferentially executed due to a rate condition or a temperature condition.

  Alternatively, the control program 101 can select all priority processes registered in the process list 107 as rewrite targets. The control program 101 rewrites the affinity mask 155 of the selected priority process by a method of performing an AND operation between the performance filter 105 and the original affinity masks 155a and 155c as performed in the block 209 of FIG. be able to.

  As a result, the affinity mask is rewritten so that the priority processes such as priority processes 1 and 2 having the original affinity mask whose bit corresponding to the logical CPU 65 is 1 are also preferentially executed by the logical CPU 65. A priority process such as the priority processes 11 and 13 in which the original affinity mask does not specify the logical CPUs 61 and 63 but specifies the logical CPU 65 is preferentially executed for the first time. Furthermore, priority processes 21, 22, and 23 that are not preferentially executed under the usage rate condition or temperature condition of block 217 are also preferentially executed. In block 311, the process list 107 is updated for the process in which the affinity mask 155 is rewritten.

  As described above, the schedule control of the process scheduler 153 and the clock boost control of the CPU package 101 are performed asynchronously. Even if the clock boost is set, the frequency of the actual clock boost depends on the state of the system at that time. Change. As the number of logical processors to be clock boosted is increased, the difference between the priority process and the non-priority process is reduced. The optimal number of logical processors for clock boosting is the number at which the average processing time of all priority processes is minimized.

  Therefore, in cases where the CPU utilization of a clock-boosted logical processor has decreased, the number of logical processors to be clock-boosted has been reduced to increase the frequency with which the logical processor to which the priority process is dispatched is clock-boosted. Is more effective. In block 313, when the CPU usage rate of the logical processor for which the clock boost setting has been set falls below a predetermined value, the control program 101 returns to block 307 to change the P state and the number of logical processors to be clock boosted. Can be reduced. In block 309, the affinity mask 155 is rewritten based on the procedure in block 209 in FIG. 3, and the process list 107 is updated in block 311. If the CPU usage rate is greater than or equal to the predetermined value in block 313, the process returns to block 213 in FIG.

  Even if the CPU usage rates of the logical CPUs 61 and 63 once exceed a predetermined value in the block 217 of FIG. 3, the state of the subsequent processes may change and the CPU usage rates may decrease. On the other hand, in block 217, a priority process that is not preferentially executed due to a temperature condition or a usage rate condition occurs. When there is a margin in the logical CPUs 61 and 63, it is desirable to shift the priority process that is not preferentially executed to a state that is preferentially executed.

  In block 401 of FIG. 5 following block 303, the control program 101 determines whether or not the CPU usage rates of the logical CPUs 61 and 63 to be clock-boosted are less than a predetermined value. The predetermined value in block 401 is smaller than the predetermined value in block 303. If the CPU usage rate of any logical CPU is equal to or greater than the predetermined value, the process returns to block 213 in FIG. In any logical CPU, if the CPU usage rate is less than the predetermined value, the process proceeds to block 403.

In block 403, the control program 101 refers to the process list 107, and the priority processes 21, 23, which are not preferentially executed because the logical CPUs 61, 63 that are the targets of the clock boost and the logical CPUs 65, 67 that are not the targets are assigned. Whether or not there is 25 is determined. Note that the priority processes 11, 12, and 13 that are not preferentially executed because only the logical CPUs 65 and 67 that are not subject to clock boost are assigned are designated to be executed by the logical CPUs 61 and 63. Since it is not, exclude it from the target of migration.

  If there is no priority process that is not subject to priority execution, the process returns to block 213. If there is a target priority process, the process proceeds to block 405, and the control program 101 uses the original affinity mask 155a adopted in block 209 in FIG. Rewrite the original affinity mask 155. In block 427, the control program 101 updates the process list 107 for the process in which the affinity mask 155 is rewritten.

  When the CPU usage rate of the logic CPUs 61 and 63 for which the clock boost is set increases, even if there is no logic CPU that can additionally set the clock boost, the logic for which the clock boost is not set If the CPU usage rates of the CPUs 65 and 67 are low, processing may be faster if the processor affinity is relaxed and the priority process is not preferentially executed. In block 421 in FIG. 6 following block 305 in FIG. 4, the control program 101 checks whether the CPU usage rates of the logical CPUs 65 and 67 that are not the target of clock boost are less than a predetermined value.

  When the CPU usage rate of either of the logical CPUs 65 and 67 is equal to or greater than a predetermined value, it is determined that the load of the logical CPUs 61 and 63 that are the targets of clock boost cannot be distributed to the other logical CPUs 65 and 67. Return to block 213 of FIG. When the CPU usage rates of both the logical CPUs 65 and 67 are less than the predetermined value, the process shifts to block 423, and the control program 101 refers to the process list 107, and a predetermined number of processes among the priority processes that are preferentially executed. To the original affinity mask bit pattern.

  Alternatively, when it is clear that all the logical CPUs 61, 63, 65, and 67 are equivalent to the process, the control program 101 dispatches such priority processes in a distributed manner to the logical CPUs 61, 63, 65, and 67. Thus, the affinity mask 155 is rewritten like the original affinity mask 155a of FIG. Alternatively, the control program 101 rewrites the affinity mask 155 so that these processes are dispatched only to the logical CPUs 65 and 67. In block 425, the control program 101 updates the process list 107 for the priority process for which the affinity mask 155 has been rewritten, and returns to block 213.

  So far, priority processes have been treated equally if they belong to one of four groups. Some priority processes want to be processed in a shorter time than other priority processes. For example, for a foreground window among processes with windows, the user expects to get a response in a shorter time. The ACPI interface can set the logical CPU to different operating frequencies even in the same P state. By using this characteristic, the affinity mask can be rewritten so that the highest priority process that needs to be processed in a shorter time in the priority process operates only with a logical CPU having a higher operating frequency.

  In order to minimize the average processing time of all priority processes, the number of logical CPUs with optimal clock boost settings always changes with the operating state of the system. In FIG. 4, the method of dynamically changing the P-state and the affinity mask based on the CPU usage rate has been described. However, in order to control the number of logical CPUs, a dummy process that is dispatched only to each logical CPU is generated. There is also a method for measuring the processing time.

  For example, each logical CPU 61, 63, 65, 67 executes a dummy process with an intermediate priority level, grasps the actual clock boost status based on the processing time, and changes the P state. You can also. Alternatively, it is also possible to change the P-state by obtaining the frequency or time during which each logical CPU is clock boosted directly from the ACPI interface 170. Further, the control program 101 can dynamically change the P state based on the temperature and average power consumption of each logical CPU measured by the ACPI interface 170.

  Although the present invention has been described with the specific embodiments shown in the drawings, the present invention is not limited to the embodiments shown in the drawings, and is known so far as long as the effects of the present invention are achieved. It goes without saying that any configuration can be adopted.

DESCRIPTION OF SYMBOLS 10 ... Computer 11 ... CPU package 100 ... Priority process processing system 101 ... Control program 150 ... Operating system

Claims (14)

In a computer comprising a plurality of logical processors capable of clock boosting that temporarily increases the operating frequency of some logical processors based on conditions related to the magnitude of the load ,
Setting the clock boost to any of the plurality of logical processors;
Determining whether the newly created process is a preferred process that displays a foreground window;
Obtaining a usage rate of a logical processor in which the clock boost is set;
Rewriting the affinity mask of the priority process when the usage rate is less than a predetermined value, and dispatching the priority process only to the logical processor for which the clock boost is set to execute the processing. Computer program to let you. The computer program according to claim 1 , wherein the step of preferentially executing includes ANDing a mask bit indicating an original affinity mask and a setting state of the clock boost for each of the plurality of logical processors. . 3. The computer program according to claim 1 , wherein the step of preferential execution includes a step of rewriting the affinity mask so as to specify only a logical processor for which the clock boost is set. Obtaining a temperature of the logical processor;
The computer program according to any one of claims 1 to claim 3 wherein the temperature is al the affinity mask rewrites when less than the predetermined value. Determining whether there is a logical processor for which the clock boost is not set;
When there is a logical processor for which the clock boost is not set and the usage rate of the logical processor for which the clock boost is set is equal to or higher than a predetermined value, the clock boost is added to the non-set logical processor. The computer program according to any one of claims 1 to 4 , further comprising a step of making an appropriate setting. 6. The computer program according to claim 5 , further comprising the step of rewriting the affinity mask of the priority process so that the priority process with the rewritten affinity mask is executed only by the logical processor set with the clock boost. . When the usage rate of the logical processor for which the clock boost is set is less than a predetermined value, it is determined whether there is a priority process that is dispatched to the logical processor for which the clock boost is not set and is not preferentially executed. Steps,
The computer program according to claim 1 , further comprising a step of rewriting the affinity mask so that the priority process that is not preferentially executed is preferentially executed. Obtaining a utilization of a logical processor for which the clock boost is not set ;
When the usage rate of the logical processor to which the clock boost is set is equal to or higher than a predetermined value and the usage rate of the logical processor to which the clock boost is not set is less than the predetermined value, the priority process is not set to the clock boost. The computer program according to claim 1 , further comprising a step of rewriting the affinity mask so as to be executed by a setting logical processor. The priority process, the computer program as claimed in any one of claims 8 to determine based on thread priority. The priority process, the computer program according to any one of claims 1 to 9, which is determined based on the priority of the disk I / O processing. A computer comprising a computer-readable recording medium storing the computer program according to any one of claims 1 to 10. A computer with a multi-processor system,
A plurality of logical processors capable of clock boosting to temporarily increase the operating frequency of some logical processors based on conditions associated with the magnitude of the load ;
Setting means for setting the clock boost to any of the plurality of logical processors;
Determining means for determining whether or not the newly generated process is a priority process for displaying a foreground window;
Usage rate acquisition means for acquiring a usage rate of a logical processor in which the clock boost is set;
Execution means for rewriting the affinity mask of the priority process when the usage rate is less than a predetermined value, and dispatching the priority process only to the logical processor for which the clock boost is set to execute the priority process. Having a computer. The computer according to claim 12 , wherein the condition related to the magnitude of the load is set in a range in which the temperature of the logical processor is less than a reference value and the power consumption exceeds the thermal design power consumption. The computer according to claim 12 or 13 , wherein the plurality of logical processors are configured by a multi-core processor in which a plurality of processor cores are mounted in one package.

Images