WO2015096031A1

WO2015096031A1 - Method and apparatus for allocating thread shared resource

Info

Publication number: WO2015096031A1
Application number: PCT/CN2013/090326
Authority: WO
Inventors: 陈云; 宋昆鹏; 崔晓松
Original assignee: 华为技术有限公司
Priority date: 2013-12-24
Filing date: 2013-12-24
Publication date: 2015-07-02
Also published as: CN105378652B; CN105378652A

Abstract

Embodiments of the present invention provide a method and an apparatus for allocating a thread shared resource. The method comprises: determining, according to a long delay operation judgment signal sent by a target thread, that the target thread performs a long delay operation; counting the number of times that the target thread performs a long delay operation in a preset time; determining a replaced weight value of the target resource according to the number of times that the target thread performs a long delay operation, a preset system weight coefficient, a preset weight adjustment policy value and frequencies of the target resource being accessed in the target thread; and reallocating the target resource according to the replaced weight value. In the embodiments of the present invention, allocation of a multi-thread shared resource is more proper, waste of a resource is avoided, the probability that important resources of some important threads are replaced, and it is avoided that performance of a target thread is decreased because after a target thread releases some important resources but needs to regain the resource when the resources are reused, thereby ensuring performance of the target thread.

Description

Thread sharing resource allocation method and device

Technical field

The present invention relates to communications technologies, and in particular, to a thread sharing resource allocation method and apparatus. Background technique

Simultaneous Multi Threading (SMT) technology is a processor technology. The SMT structure enables the processor to execute multiple instructions from different threads and mine single-thread internal instructions by copying the structure state on the processor. Level-parallelism and thread-level parallelism between threads reduce horizontal waste and vertical waste in the processor's internal pipeline, maximizing wide-emission, out-of-order superscalar processing to improve processor utilization. The above SMT structure combines the capability of superscalar multi-emission and the traditional multi-threaded architecture to hide latency. The difference is that the traditional multi-threaded architecture shares the execution resources of the processor through fast context switching, and the above SMT structure Keep multiple contexts active at the same time to compete for execution resources. This SMT structure improves resource utilization by thread sharing of resources.

In the prior art, the long delay operation of each thread is judged, and the long delay operation refers to an operation in which the processor needs dozens or even hundreds of processor clocks to return a result, if it is determined that a thread has a long delay Operation, the resources occupied by the thread in the shared resource are released.

However, with the prior art, in the case where the competition of the shared resources is not fierce, the performance of the above-mentioned threads is degraded; and the priority of the key threads in the system to the shared resources cannot be guaranteed, thereby affecting the key performance indicators of the system. Summary of the invention

The embodiment of the invention provides a thread sharing resource allocation method and device, which are used to solve the problem that the existing shared resource allocation method affects thread performance.

A first aspect of the embodiments of the present invention provides a thread sharing resource allocation method, including: determining, according to a long delay operation judgment signal sent by a target thread, that the target thread performs a long delay operation;

Counting the number of times the target thread performs a long delay operation within a preset time period; Determining the replaced weight value of the target resource according to the number of times the target thread performs the long delay operation, the preset system weight coefficient, the preset weight adjustment policy value, and the visited frequency of the target resource in the target thread, where The target resource is a multi-thread shared resource occupied by the target thread;

Reassigning the target resource according to the replaced weight value.

With reference to the first aspect, in a first possible implementation manner of the first aspect, determining, by the long delay operation determining signal sent by the target thread, that the target thread performs a long delay operation, the method includes: receiving the target thread to send Long delay operation judgment signal;

Determining, according to the long delay operation judgment signal, that the target thread performs a long delay operation. With reference to the first possible implementation manner of the first aspect, in the second possible implementation manner of the first aspect, the long delay operation determining signal includes: a long delay operation signal, and the long delay The corresponding access information of the operation signal; the long delay operation includes: an operation of accessing the main memory due to a cache miss, or an access operation of the remote shared memory.

With reference to the first aspect to any one of the second possible implementation manners of the first aspect, in a third possible implementation manner of the first aspect, the determining, by the Before the target thread performs a long delay operation, it also includes:

Obtaining the preset system weight coefficient and the preset weight adjustment policy value.

With reference to the first aspect, in a fourth possible implementation manner of the first aspect, the re-allocating the target resource according to the replaced weight includes:

According to the replaced weight, the target resource is preferentially replaced, or the probability that the target resource is replaced is reduced.

A second aspect of the embodiments of the present invention provides a thread sharing resource allocation apparatus, including: a first determining module, configured to determine, according to a long delay operation determining signal sent by a target thread, that the target thread performs a long delay operation;

a counting module, configured to count a number of times that the target thread performs a long delay operation in a preset time; a second determining module, configured to perform a long delay operation according to the target thread, a preset system weight coefficient, and a preset Determining, by the weight adjustment policy value, the accessed frequency of the target resource in the target thread, the replaced weight value of the target resource, where the target resource is a multi-thread shared resource occupied by the target thread;

And an allocating module, configured to reallocate the target resource according to the replaced weight value. With reference to the second aspect, in a first possible implementation manner of the second aspect, the first determining module is specifically configured to receive a long delay operation determining signal sent by the target thread; The determination signal determines that the target thread has performed a long delay operation.

With reference to the first possible implementation manner of the second aspect, in the second possible implementation manner of the second aspect, the long delay operation determining signal includes: a long delay operation signal, and the long delay The corresponding access information of the operation signal; the long delay operation includes: an operation of accessing the main memory due to a cache miss, or an access operation of the remote shared memory.

With reference to any one of the second aspect to the second possible implementation of the second aspect, in a third possible implementation manner of the second aspect, the device further includes:

And an obtaining module, configured to acquire the preset system weight coefficient and the preset weight adjustment policy value before determining that the target thread performs a long delay operation according to the long delay operation judgment signal sent by the target thread.

With reference to the second aspect, in a first possible implementation manner of the second aspect, the assigning module is specifically configured to: the target resource is preferentially replaced according to the replaced weight, or the target resource is reduced The probability of replacement.

A third aspect of the embodiments of the present invention provides a thread sharing resource allocation apparatus, including: a memory, configured to store an instruction;

a processor, coupled to the memory, configured to execute an instruction stored in the memory, to determine that the target thread has performed a long delay operation according to a long delay operation determination signal sent by the target thread; The number of times the target thread performs a long delay operation; the number of times of the long delay operation by the target thread, the preset system weight coefficient, the preset weight adjustment policy value, and the access of the target resource in the target thread Determining, by frequency, or the like, the replaced weight value of the target resource, where the target resource is a multi-thread shared resource occupied by the target thread; and the target resource is reallocated according to the replaced weight value.

With reference to the third aspect, in a first possible implementation manner of the third aspect, the processor is configured to receive a long delay operation determination signal sent by the target thread, and determine a signal according to the long delay operation. It is determined that the target thread has performed a long delay operation.

With reference to the first possible implementation manner of the third aspect, in the second possible implementation manner of the third aspect, the long delay operation determining signal includes: a long delay operation signal, and the long delay The corresponding access information of the operation signal; the long delay operation includes: accessing due to a cache miss The main memory operation, or the access operation of the remote shared memory.

With reference to any one of the third aspect to the second possible implementation manner of the third aspect, in a third possible implementation manner of the third aspect, the processor is further configured to send the length according to the target thread The delay operation determining signal determines that the target system weight coefficient and the preset weight adjustment policy value are obtained before the target thread performs a long delay operation.

With reference to the third aspect, in a fourth possible implementation manner of the third aspect, the processor is specifically configured to: the target resource is preferentially replaced according to the replaced weight, or the target resource is replaced The probability.

In this embodiment, determining, according to the long delay operation judgment signal sent by the target thread, that the target thread performs a long delay operation, and counting the number of long delay operations, and then performing the long delay operation according to the target thread, The system weight coefficient, the preset weight adjustment policy value, and the accessed frequency of the target resource in the target thread determine the replaced weight value of the target resource, and the target resource is reallocated when needed according to the replaced weight value. It realizes more reasonable allocation of multi-thread shared resources. According to the specific situation, some target resources are preferentially replaced, so that other threads can compete, avoiding waste of resources, and reducing the probability that important resources of some key threads are replaced, and avoiding After the target thread releases some important resources and then reacquires it again, the target thread performance is degraded, thus ensuring the performance of the target thread. DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief description of the drawings used in the embodiments or the prior art description will be briefly described below. Obviously, the drawings in the following description It is a certain embodiment of the present invention, and other drawings can be obtained from those skilled in the art without any inventive labor.

1 is a schematic flowchart of Embodiment 1 of a thread sharing resource allocation method provided by the present invention; FIG. 2 is a schematic structural diagram of Embodiment 1 of a thread sharing resource allocation apparatus provided by the present invention; FIG. 3 is a schematic diagram of a thread sharing resource allocation apparatus provided by the present invention; FIG. 4 is a schematic structural diagram of Embodiment 3 of a thread sharing resource allocation apparatus according to the present invention. detailed description The technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. It is a partial embodiment of the invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

1 is a schematic flowchart of Embodiment 1 of a thread sharing resource allocation method provided by the present invention. The method is implemented in a multi-thread system. As shown in FIG. 1, the method includes:

S101. Determine, according to the Long Latency Event (LLE) judgment signal sent by the target thread, that the target thread performs a long delay operation.

S102. Count the number of times that the target thread performs a long delay operation within a preset time. Specifically, each time a long delay operation is performed, 1 is added to the number of statistics. That is, after determining that the target thread has performed a long delay operation according to the long delay operation judgment signal, the addition operation is performed.

In the specific implementation process, statistics are periodically performed, and can be timed by a time window timer. The preset time can refer to a period of time. When the time window timer reaches the threshold, the previous count is cleared, that is, from the next cycle. Start recalculating.

5103. Determine, according to the number of times that the target thread performs the long delay operation, the preset system weight coefficient, the preset weight adjustment policy value, and the visited frequency of the target resource in the target thread, the replaced weight value of the target resource. The target resource is a multi-thread shared resource occupied by the target thread.

5104. Re-allocate the target resource according to the replaced weight value. Specifically, the target resource may be reallocated when needed. That is, resource re-allocation is not necessarily performed when the time window timer reaches the threshold, but resource re-allocation is performed when other threads or target threads need to apply for new shared resources.

In a specific implementation process, the foregoing target resource is re-allocated according to the replaced weight value, and the target resource may be preferentially replaced according to the replaced weight value, or the probability that the target resource is replaced may be reduced. The shared resource occupied by the target thread is preferentially replaced, or the probability that the shared resource occupied by the target thread is replaced is reduced. Specifically, the preset weight adjustment policy value can be adjusted. For example, the preset weight adjustment policy value can be set to 1, indicating that the longer the long delay operation, the smaller the replaced weight value, and the less the long delay operation is. The replacement weight value is larger. Set the preset weight adjustment policy value to 0 to indicate that the longer the long delay operation, the larger the replacement weight value, and the smaller the long delay operation, the smaller the replacement weight value. Then set the replacement weight coefficient within a range The shared resources occupied by the target thread are preferentially replaced. In this case, the contribution of the resources to the target thread is small or the target thread belongs to the background application (that is, its running time does not affect the work of the whole system), and these resources are Released, in order to let the thread that needs these resources to compete, then the target thread to compete for the more needed resources of the target thread to replace; in another range, reduce the probability that the shared resources occupied by the target thread are replaced, this In this case, it is indicated that these resources are more important to the target thread, and try not to let these resources be replaced.

In this embodiment, determining, according to the long delay operation judgment signal sent by the target thread, that the target thread performs a long delay operation, and counting the number of long delay operations, and then performing the long delay operation according to the target thread, The system weight coefficient, the preset weight adjustment policy value, and the accessed frequency of the target resource in the target thread determine the replaced weight value of the target resource, and the target resource is reallocated when needed according to the replaced weight value. It realizes more reasonable allocation of multi-thread shared resources. According to the specific situation, some target resources are preferentially replaced, so that other threads can compete, avoiding waste of resources, and reducing the probability that important resources of some key threads are replaced, and avoiding After the target thread releases some important resources and then reacquires it again, the target thread performance is degraded, thus ensuring the performance of the target thread.

Further, the foregoing determining, according to the LLE determination signal sent by the target thread, that the target thread performs a long delay operation, specifically: receiving an LLE determination signal sent by the target thread; determining, according to the LLE determination signal, that the target thread performs a long delay operating.

Specifically, in the embodiment of the present invention, the long delay operation specifically refers to an operation that requires dozens or even hundreds of processor clocks to return a result, and may include: accessing a main memory operation due to a cache miss, or remote sharing Memory access operations, but not limited to this. When the target thread has a long-delay operation, for example, the last-level cache is lost, the remote data is accessed, and the like, the target thread sends a long-delay operation determination signal, and the LLE determination signal can be sent to the embodiment of the present invention. The thread sharing resource allocation device is judged by the long delay judging module in the device, and the long delay judging module can judge whether the long delay operation is based on the judgment signal, and if the long delay operation is performed, the target thread is extended. The number of operations is increased by one. It should be noted that the long delay operation determination signal may include: the long delay operation signal (LLE-like), and the corresponding access information of the long delay operation signal. The corresponding access information may be: an access request, target address information of the access, read and write information, target thread information of the access, and the like. In the specific implementation process, which signals can be used to judge the LLE, can be configured by the whole system according to actual needs. Further, before determining that the target thread performs the long delay operation according to the LLE determination signal sent by the target thread, the method further includes acquiring the preset system weight coefficient and the preset weight adjustment policy value. In the specific implementation process, the preset system weight coefficient and the preset weight adjustment strategy are configured by the system administrator according to the application requirements of the whole system, for example, according to the application scenario, setting the preset weight adjustment policy value to 1, to represent The longer the long delay operation, the smaller the replacement weight value. The smaller the long delay operation, the larger the replacement weight value. Set the preset weight adjustment policy value to 0 to indicate that the long delay operation is more, and the replacement weight value is replaced. The larger, the less the long delay operation, the smaller the replacement weight value. However, it is not limited to this, and the above system administrator can also dynamically adjust the above settings according to actual conditions during system operation.

In the specific implementation process, the number of times of performing the long delay operation, the preset system weight coefficient, the preset weight adjustment policy, and the visited frequency of the target resource in the target thread are determined by using the target thread to determine the replaced weight value of the target resource. In this example, the weight adjustment policy value is 1, to indicate that the longer the long delay operation, the smaller the replacement weight value, and the smaller the long delay operation, the larger the replacement weight value. The default weight adjustment policy value is W _ Wr (0, the default system weight coefficient is W (0, the target frequency of the target resource in the target thread is Re c fy (0, the target thread performs long delay operation). The number of times is LLe _ cnt(i) , and the replaced weight value C of the target resource can be calculated according to the following formula: W(0: ( 1 ) If W _ str(i) = 1, Cost(i) = W(i I LLe _ cnt(i) * Re cenly(i); ( 2 ) If W _ str(i) = 0, Cost(i) = W(i) * Re cenly(i) * LLe _ cnt(i) Where "*,, means multiplication, and 'Τ" means division.

For example, in the application process, in the application process: (1) Under normal circumstances, a reasonable preset system weight coefficient and a preset weight adjustment policy value are set, so that the higher the long delay operation frequency, the probability that the target resource is preferentially replaced. The larger, to reduce the interference of the target thread to other threads in the multi-threaded system.

(2) Take the example of the cache resource in the SMT processor. In some applications, the preset should be adjusted if the cache of the second or third level (L2/L3) can be turned off or not. The system weight coefficient makes the cache line in the first level cache as low as possible because of the long delay operation of other threads, so that the penalty is too large. It should be noted that the competition of resources in the first level cache by each thread is based on the Cache Line. (3) At some time, the number of threads running simultaneously in the system is relatively small. At this time, the competition of shared resources among multiple threads is small. The preset system weight value can be adjusted to minimize the probability of target resources being replaced, and the target thread is reduced. Exchange and exchange of resources. (4) The length of the preset time also affects the adjustment frequency of the replacement weight value. In some application scenarios, a smaller preset time is set. The period of the meter is small, so that the replacement weight value can be adjusted in a shorter time.

A person skilled in the art can understand that all or part of the steps of implementing the above method embodiments may be completed by using hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, and the program is executed when executed. The foregoing steps include the steps of the foregoing method embodiments; and the foregoing storage medium includes: a medium that can store program codes, such as a ROM, a RAM, a magnetic disk, or an optical disk.

FIG. 2 is a schematic structural diagram of Embodiment 1 of a thread sharing resource allocation apparatus according to the present invention. As shown in FIG. 2, the apparatus includes: a first determining module 201, a counting module 202, a second determining module 203, and an allocating module 204. among them:

The first determining module 201 is configured to determine, according to the long delay operation determining signal sent by the target thread, that the target thread performs a long delay operation.

The counting module 202 is configured to count the number of times the target thread performs a long delay operation within a preset time.

The second determining module 203 is configured to determine the target according to the number of times the target thread performs a long delay operation, a preset system weight coefficient, a preset weight adjustment policy value, and an accessed frequency of a target resource in the target thread. The replaced weight value of the resource, where the target resource is a multi-thread shared resource occupied by the target thread.

The allocating module 204 is configured to re-allocate the target resource according to the replaced weight value.

Further, the first determining module 201 is specifically configured to receive a long delay operation determining signal sent by the target thread, and determine, according to the long delay operation determining signal, that the target thread performs a long delay operation.

It should be noted that the long delay operation determination signal includes: a long delay operation signal, and access information corresponding to the long delay operation signal; the long delay operation includes: accessing the main memory due to a cache miss The operation, or access operation of the remote shared memory.

FIG. 3 is a schematic structural diagram of Embodiment 2 of a thread sharing resource allocation apparatus according to the present invention. On the basis of FIG. 2, the apparatus includes:

The obtaining module 205 is configured to obtain the preset system weight coefficient and the preset weight adjustment policy value before determining that the target thread performs a long delay operation according to the long delay operation determining signal sent by the target thread. In addition, the foregoing allocation module 204 is specifically configured to: the target resource is preferentially replaced according to the replaced weight, or the probability that the target resource is replaced is reduced.

The foregoing modules are used to implement the foregoing method embodiments, and the implementation principles and technical effects are similar, and details are not described herein again.

4 is a schematic structural diagram of Embodiment 3 of a thread sharing resource allocation apparatus according to the present invention. As shown in FIG. 4, the apparatus includes: a memory 401 and a processor 402, where

The memory 401 is configured to store an instruction.

The processor 402, coupled to the memory 401, is configured to execute an instruction stored in the memory 401, specifically, configured to determine, according to the long delay operation determination signal sent by the target thread, that the target thread performs a long delay operation; The number of times the target thread performs a long delay operation within a preset time; the number of times of the long delay operation by the target thread, the preset system weight coefficient, the preset weight adjustment policy value, and the target resource in the target thread The replaced weight value of the target resource is determined by the frequency of access, etc., wherein the target resource is a multi-thread shared resource occupied by the target thread; and the target resource is reallocated according to the replaced weight value.

Further, the processor 402 is specifically configured to receive a long delay operation determination signal sent by the target thread, and determine, according to the long delay operation determination signal, that the target thread performs a long delay operation.

The processor 402 is specifically configured to: the target resource is preferentially replaced according to the replaced weight, or reduce a probability that the target resource is replaced.

In addition, the processor 402 is further configured to acquire the preset system weight coefficient and the preset weight adjustment before determining that the target thread performs a long delay operation according to the long delay operation determination signal sent by the target thread. Policy value.

Finally, it should be noted that the above embodiments are only for explaining the technical solutions of the present invention, and are not intended to be limiting thereof; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art will understand that The technical solutions described in the foregoing embodiments may be modified, or some or all of the technical features may be equivalently replaced; and the modifications or substitutions do not deviate from the technical solutions of the embodiments of the present invention. range.

Claims

claims

1. A thread shared resource allocation method, characterized by including:

Determine that the target thread has performed a long-delay operation based on the long-delay operation judgment signal sent by the target thread;

Count the number of long-delay operations performed by the target thread within the preset time;

The replaced weight value of the target resource is determined according to the number of long-delay operations performed by the target thread, the preset system weight coefficient, the preset weight adjustment strategy value, and the access frequency of the target resource in the target thread, where , the target resource is a multi-thread shared resource occupied by the target thread;

The target resource is reallocated according to the replaced weight value.

2. The method according to claim 1, characterized in that determining that the target thread has performed a long-delay operation based on the long-delay operation judgment signal sent by the target thread includes:

Receive the long-delay operation judgment signal sent by the target thread;

It is determined according to the long-delay operation judgment signal that the target thread has performed a long-delay operation.

3. The method according to claim 2, characterized in that, the long delay operation judgment signal includes: a long delay operation signal, and access information corresponding to the long delay operation signal; the long delay operation signal Operations include: accessing main memory due to cache misses, or accessing remote shared memory.

4. The method according to any one of claims 1-3, characterized in that, before determining that the target thread has performed a long-delay operation based on the long-delay operation judgment signal sent by the target thread, it further includes:

Obtain the preset system weight coefficient and the preset weight adjustment strategy value.

5. The method according to claim 1, characterized in that: redistributing the target resource according to the replaced weight includes:

According to the replacement weight, the target resource is replaced with priority, or the probability of the target resource being replaced is reduced.

6. A thread shared resource allocation device, characterized by including:

The first determination module is used to determine that the target thread has performed a long-delay operation based on the long-delay operation judgment signal sent by the target thread;

A counting module used to count the number of long-delay operations performed by the target thread within a preset time; The second determination module is used to determine the target resource based on the number of long-delay operations performed by the target thread, the preset system weight coefficient, the preset weight adjustment strategy value, and the access frequency of the target resource in the target thread. The replaced weight value of , wherein the target resource is a multi-thread shared resource occupied by the target thread;

An allocation module, configured to reallocate the target resource according to the replaced weight value.

7. The device according to claim 6, wherein the first determination module is specifically configured to receive a long-delay operation judgment signal sent by the target thread; and determine the long-delay operation judgment signal according to the long-delay operation judgment signal. The above target thread performed a long-delay operation.

8. The device according to claim 7, wherein the long delay operation judgment signal includes: a long delay operation signal, and access information corresponding to the long delay operation signal; the long delay operation signal Operations include: accessing main memory due to cache misses, or accessing remote shared memory.

9. The device according to any one of claims 6 to 8, further comprising: an acquisition module, configured to determine that the target thread has performed a long delay based on the long delay operation judgment signal sent by the target thread. Before operation, obtain the preset system weight coefficient and the preset weight adjustment strategy value.

10. The device according to claim 6, characterized in that the allocation module is specifically configured to allow the target resource to be replaced first according to the replacement weight, or to reduce the probability of the target resource being replaced.

11. A thread shared resource allocation device, characterized by including:

Memory, used to store instructions;

A processor, coupled to the memory, configured to execute instructions stored in the memory, for determining that the target thread has performed a long-delay operation based on a long-delay operation judgment signal sent by the target thread; statistical preset The number of times the target thread performs long-delay operations within the time period; According to the number of times the target thread performs long-delay operations, the preset system weight coefficient, the preset weight adjustment strategy value, and the access of the target resource in the target thread Frequency, etc. determine the replaced weight value of the target resource, wherein the target resource is a multi-thread shared resource occupied by the target thread; the target resource is reallocated according to the replaced weight value.

12. The device according to claim 11, wherein the processor is specifically configured to receive a long-delay operation judgment signal sent by the target thread; and based on the long-delay operation judgment signal No. It is determined that the target thread has performed a long-delay operation.

13. The device according to claim 12, wherein the long delay operation judgment signal includes: a long delay operation signal, and access information corresponding to the long delay operation signal; the long delay operation signal Operations include: accessing main memory due to cache misses, or accessing remote shared memory.

14. The device according to any one of claims 11 to 13, wherein the processor is further configured to determine that the target thread has performed a long delay based on the long delay operation judgment signal sent by the target thread. Before operation, obtain the preset system weight coefficient and the preset weight adjustment strategy value.

15. The apparatus according to claim 11, wherein the processor is specifically configured to allow the target resource to be replaced first according to the replacement weight, or to reduce the probability of the target resource being replaced.