WO2015196369A1

WO2015196369A1 - Distributed cache control method and apparatus

Info

Publication number: WO2015196369A1
Application number: PCT/CN2014/080630
Authority: WO
Inventors: 程德华
Original assignee: 华为技术有限公司
Priority date: 2014-06-24
Filing date: 2014-06-24
Publication date: 2015-12-30
Also published as: CN105393518A; CN105393518B

Abstract

The embodiment of the present invention provides a distributed cache control method and apparatus. The distributed cache control method provided by the present invention includes: collecting client statistic information of the cache nodes in the cluster system; computing the average expected value of users according to the client statistic information; determining the resource adjustment mode of the system according to the average expected value of the users and an expected threshold; performing node contraction operation or node expansion operation on the cache nodes in the cluster system according to the determined resource adjustment mode. The embodiment of the present invention performs corresponding adjustment strategy on the resources of the cluster system by monitoring user satisfaction to the Quality of Service (Qos) of the cluster system, thus improving the quality of service of the client.

Description

Distributed cache control method and device

Technical field

The embodiments of the present invention relate to computer technologies, and in particular, to a distributed cache control method and apparatus. Background technique

With the development and widespread use of computer technology, caches have been widely used in the cache function of cluster systems, especially the development of distributed cache clusters, which greatly improves the speed of data cache. And efficiency.

With the increase of data traffic in the cluster system and the expansion of the size of the distributed cache cluster, the resource allocation of the distributed cache nodes in the management control cluster system becomes an important factor for optimizing system resources; for example, high-performance distributed memory is currently used. Clusters of object caching systems (Memcached), such as social networking service sites (Facebook), typically perform cluster optimization management by: reducing storage data latency; adjusting cluster deployment by assessing the impact of load on cache misses Ways, specifically, analyzing the relationship between memory cache, memory replacement algorithm and cache miss, providing a more efficient Render Algorithm; or configuring the resource by changing the size of the cache by defining a predefined length optimization.

However, the prior art management method of the cluster system usually only by monitoring the resource utilization of the cluster system, and formulating a strategy for optimizing resource allocation of the cluster system. Summary of the invention

The embodiment of the present invention provides a distributed cache control method and apparatus, and performs a corresponding adjustment strategy on the resources of the cluster system by monitoring the satisfaction degree of the user to the Quality of Service (Qos) of the cluster system.

In a first aspect, an embodiment of the present invention provides a distributed cache control method, including: collecting client statistics information of a cache node in a cluster system;

Calculating an average expected value of the user according to the client statistical information; Determining a resource adjustment manner for the cluster system according to an average expected value of the user and a desired threshold;

Performing a node shrink operation or a node expansion operation on the cache node in the cluster system according to the determined resource adjustment manner.

In a first possible implementation manner of the first aspect, the determining, by the user expected value and the expected threshold, a resource adjustment manner for the cluster system, including:

Determining, when the average expected value of the user is greater than a first desired threshold, performing the node shrinking operation on the cluster system; or

When the average expected value of the user is less than a second desired threshold, it is determined that the node expansion operation is performed on the cluster system.

According to the first aspect or the first possible implementation manner of the first aspect, in a second possible implementation manner, the client statistics include the cluster data access response time, a cluster cache hit ratio, and cluster data. Throughput

And the calculating the average expected value of the user according to the client statistics, including: calculating, according to the cluster data access response time, the cluster cache hit ratio, and the cluster data throughput, the user to each cache The expected value of the node is: wherein, the data fetch response time indicating the cache node i, the _ ^ ^ represents the cache hit ratio of the cache node i, and the TH ^ represents the data throughput of the cache node i , the l 1 ₂ and ^3 represent weights;

Calculating the average expected value of the user ^: according to the expected value of the cache node by the user. According to the second possible implementation manner of the first aspect, in a third possible implementation, before the determining, by the user, the resource adjustment manner of the cluster system according to the average expected value and the expected threshold of the user, the method further includes:

Collecting resource utilization information of the cache node in the cluster system;

And determining the resource adjustment manner for the cluster system according to the average expected value and the expected threshold of the user, including:

Obtaining, according to the collected resource utilization information of the cache node and the client statistical information, a relationship between an expected value of the user for each cache node and resource utilization information of the cache node; according to the user to each cache node Expectation value and the desired threshold, and the relationship The line determines a resource adjustment manner for each cache node in the cluster system;

And performing the node shrinking operation or the node expanding operation on the cache node in the cluster system according to the determined resource adjustment manner, including:

And performing resource adjustment on each cache node in the cluster system according to the determined resource adjustment manner and the resource utilization information of the cache node.

According to a third possible implementation manner of the first aspect, in a fourth possible implementation, the resource utilization information of the cache node includes a hardware parameter of the cache node, an operating system parameter, and the cache node is a configuration parameter in the cluster system, where the operating system parameter is used to calculate a load value of the cache node;

And obtaining, according to the collected resource utilization information of the cache node and the client statistical information, a relationship curve between the expected value of each cache node and the resource utilization information of the cache node, including:

Obtaining, according to the hardware parameter of the cache node, the load value of the cache node, the configuration parameter of the cache node in the cluster system, and the client statistics, the expected value of the user to each cache node according to the The relationship between the hardware parameters of the cache node, the load value of the cache node, and the configuration parameters of the cache node in the cluster system is:

E _n = f(X _n J _n , Z _n ) ;

The X indicates the hardware parameter of the cache node, the 1 indicates the load value of the cache node, and the 2 indicates the configuration parameter of the cache node in the cluster system;

And determining a resource adjustment manner for each cache node in the cluster system according to the expected value of the cache node and the expected threshold, and the relationship curve, including:

Determining, by changing at least one of the x, the y, and the Z, to each cache in the cluster system when the „ is greater than the first desired threshold or less than the second desired threshold The node performs resource adjustment such that the adjusted value is less than the first desired threshold and greater than the second desired threshold.

According to a fourth possible implementation manner of the first aspect, in a fifth possible implementation, the method further includes: calculating, according to a load value of the cache node, a current average load value of the cluster system; When the first expected threshold is greater than the second expected threshold, a resource adjustment manner for the cluster system is determined according to a current average load value and a load threshold of the cluster system. According to the fifth possible implementation manner of the first aspect, in a sixth possible implementation, the determining, by the current average load value and the load threshold of the cluster system, a resource adjustment manner for the cluster system, including :

Determining, when the current average load value of the cluster system is greater than the first load threshold, performing the node expansion operation on the cluster system; or

When the current average load value of the cluster system is less than the second load threshold, it is determined that the node shrink operation is performed on the cluster system.

In a second aspect, the embodiment of the present invention provides a distributed cache control apparatus, including: a collection module, configured to collect client statistics information of a cache node in a cluster system; and a calculation module, configured to collect a client according to the collection module The statistical information calculates the average expected value of the user;

a decision module, configured to determine, according to an average expected value of the user obtained by the computing module, a resource adjustment manner to the cluster system according to a desired threshold;

And an execution module, configured to perform a node shrink operation or a node expansion operation on the cache node in the cluster system according to the resource adjustment manner determined by the decision module.

In a first possible implementation manner of the second aspect, the determining module is specifically configured to: when the average expected value of the user is greater than a first expected threshold, determine to perform the node shrinking operation on the cluster system; or And determining, when the average expected value of the user is less than a second expected threshold, performing the node expansion operation on the cluster system.

According to the second aspect or the first possible implementation manner of the second aspect, in a second possible implementation, the client statistics include the cluster data access response time, the cluster cache hit ratio, and the cluster data. Throughput

The calculation module is specifically configured to calculate, according to the cluster data access response time, the cluster cache hit ratio, and the cluster data throughput, an expected value of the user for each cache node: where Γ _; — represents the data fetch response time of the cache node i, the _^^ represents the cache hit ratio of the cache node i, the 7H _; _^. Representing the data throughput of the cache node i, the ^, ^, and ^ represent weights; and further calculating the average expected value of the user ^: according to the expected value of the cache node by the user. According to a second possible implementation manner of the second aspect, in a third possible implementation, the collecting module is further configured to collect, before the determining module determines a resource adjustment manner of the cluster system, Resource utilization information of a cache node in the cluster system;

The determining module, comprising: an obtaining unit, configured to acquire, according to resource utilization information of the cache node and the client statistical information collected by the collecting module, an expected value of the user for each cache node and the cache node Relationship curve of resource utilization information;

a determining unit, configured to determine, according to the expected value of each cache node and the expected threshold, and a relationship curve acquired by the acquiring unit, a resource adjustment manner for each cache node in the cluster system;

The execution module is specifically configured to perform resource adjustment on each cache node in the cluster system according to the resource adjustment manner determined by the determining unit and the resource utilization information of the cache node collected by the collection module.

According to a third possible implementation manner of the second aspect, in a fourth possible implementation, the resource utilization information of the cache node includes a hardware parameter of the cache node, an operating system parameter, and the cache node is a configuration parameter in the cluster system, where the operating system parameter is used to calculate a load value of the cache node;

The obtaining unit is specifically configured to acquire, according to hardware parameters of the cache node, a load value of the cache node, configuration parameters of the cache node in the cluster system, and statistics of the client. The relationship between the expected value of the user for each cache node according to the hardware parameters of the cache node, the load value of the cache node, and the configuration parameter of the cache node in the cluster system is:

E _n = f(X _n J _n , Z _n ) ;

The X indicates the hardware parameter of the cache node, the } represents the load value of the cache node, and the ^ represents the configuration parameter of the cache node in the cluster system;

The determining unit is specifically configured to determine, by changing at least one of the X, the y, and the Z, when the „ is greater than the first desired threshold or less than the second desired threshold The resource adjustment is performed on each cache node in the cluster system, such that the adjusted „ is less than the first desired threshold and greater than the second desired threshold.

According to a fourth possible implementation manner of the second aspect, in a fifth possible implementation, the calculating module is further configured to calculate a current status of the cluster system according to a load value of the cache node Average load value;

The decision module is further configured to determine, according to the current average load value and the load threshold of the cluster system, when determining that the ^ is less than the first expected threshold and greater than the second expected threshold The resource adjustment method of the cluster system.

According to a fifth possible implementation manner of the second aspect, in a sixth possible implementation, the determining, by the determining, determining, by using a current average load value and a load threshold of the cluster system, a resource adjustment manner for the cluster system Specifically, when the current average load value of the cluster system is greater than the first load threshold, determining to perform the node expansion operation on the cluster system; or, the current average load value of the cluster system is smaller than the second load. At the threshold, it is determined that the node shrink operation is performed on the cluster system.

The distributed cache control method and device provided by the embodiment of the present invention collects the client's statistical information of each cache node in the cluster system to obtain the average expected value of the user for the system, and according to the average expected value and the expected threshold. The resource adjustment scheme of the cluster system is compared to perform corresponding node contraction or node expansion operation on the cache node in the system. In this embodiment, the user's satisfaction with the cluster system QoS is monitored, and the resources of the cluster system are adjusted accordingly. The strategy improves the quality of the client's service. 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 flowchart of a distributed cache control method according to Embodiment 1 of the present invention; FIG. 2 is a schematic structural diagram of a cluster system provided by the embodiment shown in FIG. 1;

FIG. 3 is a flowchart of a distributed cache control method according to Embodiment 2 of the present invention; FIG. 4 is a flowchart of a distributed cache control method according to Embodiment 3 of the present invention; A schematic diagram of a structure of a distributed cache control device provided in Example 4;

FIG. 6 is a schematic structural diagram of a distributed cache control apparatus according to Embodiment 5 of the present invention; FIG. FIG. 7 is a schematic structural diagram of a cluster control server according to Embodiment 6 of the present invention. detailed description

The technical solutions in the embodiments of the present invention are clearly and completely described in the following with reference to 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 a person of ordinary skill in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.

Embodiment 1

FIG. 1 is a flowchart of a distributed cache control method according to Embodiment 1 of the present invention. The method of this embodiment is applicable to the case where the cache node is allocated resources in the cluster system. The method may be performed by a distributed cache control device or a cluster control server, which is usually implemented in hardware and/or software, and may be integrated in the memory of the server, for example integrated in the processor. In the chip, it is executed by the processor. The method of this embodiment includes the following steps:

S110. Collect client statistics of the cache node in the cluster system.

A cluster system includes a plurality of interconnected computers, which can usually be a distributed system. A cluster control server controls a plurality of clients connected thereto. The client may be referred to as a cache node in the following embodiments. The data processing in the data processing usually adopts the distributed cache clustering technology. This embodiment is described by taking the Memcached clustering technology that is currently used as an example. As shown in FIG. 2, the structure of the clustering system provided in the embodiment shown in FIG. The cluster control server is connected to n cache nodes, such as Memcached server 1 to Memcached server n, each cache node is a separate client, including a server and a memory, and the n cache nodes are An application server connection that provides an application for each buffer node. The execution entities in this embodiment and the following embodiments are all cluster control servers, and the server collects client statistics information of each buffer node connected thereto, and the client statistics information may be user behavior perception of the client, usually based on the pair of users. The Qos satisfaction survey, in particular, can collect the client statistical information in a preset time window, and provide basic information for the control method provided in this embodiment.

S120. Calculate an average expected value of the user according to the client statistical information. In this embodiment, the collected client statistics information is information of each cache node connected to the cluster control server, and the hardware resources and system configurations of each cache node may be different or the same, therefore, the user The satisfaction of each cached QoS may also be different. In a specific implementation, the user's average expected value of the user's expected value of each buffer node may be calculated to obtain a comprehensive evaluation of the overall QoS of the cluster system, for example, According to the collected client statistics, the average expected value of the cluster system is obtained by using the optimization theory algorithm of operations research.

S130. Determine a resource adjustment manner for the cluster system according to the average expected value of the user and the expected threshold.

Currently, the conventional methods for resource allocation and adjustment in a cluster system include: reducing the storage data delay, adjusting the cluster deployment mode by evaluating the impact of the load on the cache miss, or changing the size of the cache by making a predefined step size for resource adjustment. Obviously, the prior art generally adopts a single analysis and judgment rule for resource adjustment. The error rate of the data cache in the cluster system is high, and the resource control is performed according to the resource monitoring of the cache node itself, and cannot be It reflects the user's satisfaction with the QoS of each cache node. In this embodiment, the resource adjustment of the cluster system is determined according to the user's demand for Qos. For example, a desired threshold may be preset, and a resource adjustment scheme may be decided by comparing the expected threshold with the average expected value of the user.

For example, the specific manner of adjusting the resource adjustment may be: determining that the node shrinking operation is performed on the cluster system when the average expected value of the user is greater than the first expected threshold; or determining that the average expected value of the user is less than the second expected threshold. A node expansion operation is performed on the cluster system. It should be noted that the first expected threshold and the second expected threshold are the expected threshold ranges. When the calculated average expected threshold exceeds the range, the resource of the cluster system needs to be adjusted accordingly; on the one hand, the average expected value is high. In this range, the user is more satisfied with the QoS of the cache node in the cluster system, that is, the cost of the current resource configuration of the cluster system is higher. Therefore, some cache nodes in the system can be shut down to reduce system resources. Consumption, on the other hand, when the average expected value is lower than the range, the user is less satisfied with the QoS of the cache node in the cluster system, that is, the current resource configuration of the cluster system cannot meet the user's needs, and therefore, the system can be increased. The number of cache nodes to increase user satisfaction and system efficiency.

It should be noted that, in this embodiment, when the average expected value of the user is within the range of the expected threshold, the user's satisfaction with the QoS is expected to be expected, and the resources of the cluster system need not be adjusted, and the S110 may be restarted. Collect client statistics for cache nodes in the cluster system The distributed cache control method provided in this embodiment may be executed cyclically.

S140. Perform a node shrink operation or a node expansion operation on the cache node in the cluster system according to the determined resource adjustment manner.

In this embodiment, the foregoing adjustment policy is performed on the resources of the cluster system by comparing the average expected value of the user with the expected threshold. In a specific implementation, each of the cluster systems may be operated by an operation instruction according to the indication of the foregoing policy. The cache node performs a shrinking or expanding operation, and the execution of the operation corresponds to the resource adjustment policy described above, for example, changing the number of cache nodes in the cluster system.

The distributed cache control method provided in this embodiment collects client statistics of each cache node in the cluster system to obtain an average expected value of the system, and determines the cluster system according to the comparison between the average expected value and the expected threshold. The resource adjustment scheme is implemented to perform corresponding node contraction or node expansion operation on the cache node in the system. In this embodiment, the user is satisfied with the QoS of the cluster system, and the corresponding adjustment strategy is implemented on the resources of the cluster system, thereby improving the client. Service quality of the terminal; further, dynamically adjust the resource utilization of the cluster system by monitoring the average expected value of the user, and reduce the consumption of system resources as much as possible while ensuring that the user's QoS of the cluster system is expected, and correspondingly improve The efficiency of the use of system resources.

Embodiment 2

FIG. 3 is a flowchart of a distributed cache control method according to Embodiment 2 of the present invention. As shown in FIG. 3, the method provided in this embodiment includes the following steps:

S210. Collect client statistics of the cache node in the cluster system.

Specifically, the specific implementation manner of S210 refers to S110 in Embodiment 1.

S220. Collect resource utilization information of the cache node in the cluster system.

In this embodiment, not only the client statistics of each cache node in the cluster system may be collected, but also the user is satisfied with the QoS of the QoS node, and the resource utilization information of each cache node may be collected, and the cache nodes themselves from the cluster system. The resources utilization information is monitored. For example, the resource utilization information of the cache node may include a central processing unit (CPU), memory, and network input/output (Input/Output, referred to as: 10) utilization, The embodiment monitors resource monitoring and user perception of the cluster system to provide a high performance distributed cache control scheme.

It should be noted that, in this embodiment, there is no clear sequence of S220 and S210, and this embodiment is shown by S220 after S210. In a specific implementation, the collection of the client statistics and the resource utilization information of the cache node may be implemented by using a code instruction. For example, the code for collecting resource information provided by the embodiment is specifically:

Struct memcached— stats {〃Define statistics data structure

Int requests;

// memcached user command mode of operation

Int gets;

Int multigets;

Int sets;

Int adds;

Int replaces;

Int deletes;

Int hits;

Int misses;

Struct stat response—time;

Struct stat get- size;

Struct timeval last— time;

};

Void printGlobalStats(struct config* config) {

Pthread—mutex—lock(&stats—lock);

Struct timeval currentTime;

Gettimeofday(&currentTime, NULL);

/ / Define the metric indicator calculation formula

Double timeDiff = currentTime.tv— sec - global— stats.last_time.tv — sec + 1 e-6 * (currentTime .tv_sec - global_stats.last_time.tv_sec);

Double rps = global— stats.requests/timeDiff;

Double std = getStdDev(&global_stats.response_time);

}

S230. Acquire, according to the collected resource utilization information of the cache node and the client statistical information, a relationship between a user expectation value of each cache node and resource utilization information of the cache node. In this embodiment, the resource utilization information of the cache node may include, for example, a hardware parameter (Hardware Parameter) of the cache node, an operating system parameter (Operating System Parameter), and configuration parameters of each cache node in the cluster system (Memcached Configuration). Parameter) , specifically, the operating system parameter represents the working state of the cache node, that is, the utilization of the node, and is used to calculate the load value of the cache node, which can be represented by the CPU, the memory, and the network IO.

For example, the definition of x is the eigenvalue of the hardware parameter of the cache node, which represents the total number of distributed cache nodes in the cluster system where the cache node is currently located. Therefore, the value of X „ represents the number of cache nodes currently composed by the cluster system. The division of resources is achieved through configuration files in the system.

Definition 1 is the eigenvalue of the operating system parameter of the cache node, and represents the working state of each cache node. This embodiment is described by taking the CPU, memory, and network IO utilization as an example. Therefore, the value of y„ corresponds to different cache nodes. The working state, the value of which is the resource utilization of the cache node. Through the dynamic adjustment of load balancing, the load value of the cache node can be expressed as:

γ _ * CPU _i + β * MEM _i + χ * NET _i (工) " _ 3 ,

In the above (1), CPUi, MEM, , Γ _; respectively represent the CPU, memory, and network IO utilization of the i-th cache node, where ", and ^ are weights; specifically, the i-th cache node Load value.

The definition Z is the characteristic value of the configuration parameter of the cache node in the cluster system, and represents the parameter setting of different Memcached in the cache node. Therefore, the value of z„ corresponds to the running parameter of Memcached of different cache nodes, and the parameter is configured by the Memcached command. Line or profile settings, 2 „ can be expressed as:

Z _n e {Conf _{ , Conf ₂ , ... , Conf _{ , ... , Conf _n }, (2) In the above (2), is the Memcached parameter configuration file ID of the i-th cache node,

(2) represents the set of configuration files for each cache node in the cluster system.

It should be noted that when the Memcached cluster application is first executed, it can be run using the default Memcached configuration file.

In a specific implementation, the expected value of the cache node obtained by the user in the embodiment may be expressed as a function according to a hardware parameter of the cache node, a load value of the cache node, and a configuration parameter of the cache node in the cluster system. :

E _n = f(X _n , Y _n , Z _n ) , (3 ) The functional relationship represented by (3) above can be obtained by long-term monitoring of the collection of historical empirical data. For example, a function approximation algorithm can be used to obtain a variation curve of „„, ynz„.

S240. Determine a resource adjustment manner of each cache node in the cluster system according to a user expectation value and a desired threshold for each cache node, and the relationship curve.

In this embodiment, for each cache node, the expected value of the user for each cache node has been obtained along with the corresponding X„, ! and the change curve, that is, X„, ! The change of ^ or ^ has an effect on the „, so, according to the comparison with the expected threshold, the resource adjustment strategy of each cache node in the cluster system can be determined, so that the adjusted „ is within the range of the desired threshold.

Specifically, when „ is greater than the first desired threshold or less than the second desired threshold, determining to adjust resources of each cache node in the cluster system by changing at least one of x _n , 1 and z „, so that the adjusted It is smaller than the first desired threshold and greater than the second desired threshold. It should be noted that the distributed cache control method provided in this embodiment separately performs resource adjustment policies on each cache node in the cluster system, so that the user The satisfaction level of the cache node Qos can reach the expected value.

S250. Perform resource adjustment on each cache node in the cluster system according to the determined resource adjustment manner and the resource utilization information of the cache node.

Different from the above-mentioned first embodiment, the embodiment specifically adjusts the resource utilization of each buffer node in the cluster system, which is different from the overall resource of the cluster system in the above embodiment; specifically, for the user-to-partial part The node whose expected value of the buffer node is within the desired threshold range does not need to perform resource adjustment. For the node whose expected value of the partial buffer node exceeds the expected threshold range, the buffer node may be respectively corresponding to the corresponding relationship of (3) above. The resources are redistributed, for example, the CPU usage of the partial cache node is turned off or turned on, the proportion of the buffer function in the memory is adjusted, the bandwidth of the network is used, and the like. Therefore, the method provided in this embodiment only needs to be satisfactory to the user. The cache node that achieves the expected effect performs resource adjustment, which saves the cost of resource adjustment by the cluster system, reduces the bit error rate of the cluster system, and increases the flexibility of the resource adjustment scheme.

The distributed cache control method provided in this embodiment collects client statistics information and resource utilization information of each cache node in the cluster system, and acquires a relationship between the expected value of each cache node and the resource utilization information of the corresponding cache node. Determining, according to the comparison between the expected value of each cache node and the expected threshold, the resource adjustment scheme of the cache node in the cluster system, so as to implement resource adjustment for the cache node that does not achieve the desired effect in the cluster system. The embodiment combines the user satisfaction with the cache node Qos and the resource utilization information of the corresponding node, and adjusts the resource of the cache node in the cluster system, which is beneficial to reducing the error rate of the cluster system. Further, this embodiment provides The distributed cache control method only needs to perform resource adjustment on the cache node whose user satisfaction degree does not achieve the expected effect, which saves the cost of resource adjustment of the cluster system, and increases the flexibility of the resource adjustment scheme.

Embodiment 3

FIG. 4 is a flowchart of a distributed cache control method according to Embodiment 3 of the present invention. As shown in FIG. 4, the method provided in this embodiment includes the following steps:

S310. Collect client statistics of the cache node in the cluster system.

S320. Calculate an average expected value of the user according to the client statistical information.

Specifically, the specific implementation manners of S310~S320 refer to S110~S120 in the first embodiment. The client statistics collected in this embodiment may include a cluster data access response time (Cluster Date Access Response Time), and a cluster cache. Cache Hit Rate and Cluster Date Throughput Rate; S320 specifically includes: calculating a user pair according to the cluster data access response time, the cluster cache hit ratio, and the cluster data throughput The expected value of the cache node is: (4) where Γ _ϊρ represents the data access response time of the cache node i, Ri—one represents the cache hit ratio of the cache node i, and THL^ represents the data throughput of the cache node i, Λ 】, ^ and represent the weight; and then calculate the average expected value of the user ^: according to the user's expected value for each cache node.

S330, collecting resource utilization information of the cache node in the cluster system, where the resource utilization information of the cache node may also include hardware parameters of the cache node, operating system parameters, and configuration parameters of each cache node in the cluster system, the operation System parameters are used to calculate the load value of the cache node.

For details, refer to S220 in Embodiment 1 for the specific implementation manner of S330.

It should be noted that, in this embodiment, there is no clear existing sequence of S330 and S310~S320. This embodiment is shown by S330 after S310~S320.

S340. Calculate a load value of each cache node, and calculate a current average load value of the cluster system according to a load value of each cache node.

S350, determining whether: is smaller than the first desired threshold and greater than the second desired threshold; if not, executing S360; if yes, executing S370 S360, determining a resource adjustment manner for the cluster system according to the sum threshold and the expected threshold.

Specifically, the specific implementation manner of S360 refers to S130 in Embodiment 1.

S370. Determine a resource adjustment manner for the cluster system according to a current average load value and a load threshold of the cluster system.

In this embodiment, the satisfaction degree of the user to the QoS can reach the expected requirement, and the method for adjusting the resource can be further determined according to the comparison between the current average load value of the cluster system and the load threshold; for example, the specific manner of adjusting the resource decision The method may be: when the current average load value of the cluster system is greater than the first load threshold, determine that the node expansion operation is performed on the cluster system; or, when the current average load value of the cluster system is less than the second load threshold, determine The cluster system performs the node shrink operation. It should be noted that, the first load threshold and the second load threshold are load threshold ranges. When the calculated current load threshold of the cluster system exceeds the range, the resource of the cluster system needs to be adjusted accordingly; When the current average load value is higher than the range, the current load of each cache node in the cluster system is higher. Therefore, the system load can be reduced by increasing the number of cache nodes in the system. On the other hand, the current average load value is lower than the current load value. In the scope, it indicates that the load of each cache node in the cluster system is not saturated. Therefore, the system resource consumption can be reduced by turning off some cache nodes in the system.

It should be noted that, in the embodiment, the current average load value of the cluster system is within the range of the load threshold, indicating that the resource configuration of the cluster system conforms to the load balancing principle on the premise that the user's satisfaction with the QoS is expected. The configuration of the distributed cache control method provided in this embodiment may be performed cyclically. The configuration of the distributed cache control method provided in this embodiment may be performed cyclically.

S380. Perform a node shrink operation or a node expansion operation on the cache node in the cluster system according to the determined resource adjustment manner.

In another implementation manner of the embodiment, according to the current average load value and the load threshold of the cluster system, the decision of the resource adjustment mode of the cluster system may also be: Cache nodes satisfying > ε in the cluster system form an migrating node The set, the cache nodes satisfying ≥ ε constitute a set of move-in nodes, where ^: represents the current average load value of the cluster system, and f is a preset ratio obtained by long-term historical monitoring data. According to the currently collected resource utilization information, if it is calculated that the set of the moved node is much smaller than the number of the set of the migrated node, it is determined that the node expansion operation is performed on the cluster system, and the new cache node is added to the set of the moved node, that is, The number of X„ is increased, and under the condition of guaranteeing user service, data migration is gradually performed by the application background; if it is calculated that the set of the moved node is much larger than the number of the set of the moved node, it is determined that the node is executed for the cluster system. In the specific implementation, in the specific implementation, the cache node with smaller load value in the cluster system can be selected to perform the merge operation, that is, the number of X„ is reduced, and the data migration by the application background is also performed under the condition of ensuring user service. .

The distributed cache control method provided in this embodiment determines the client's statistical information and resource utilization information of each cache node in the cluster system according to the average expected value of the system and the current load average value of the system. The resource adjustment scheme of the cluster system ensures the quality of service of the client, and further improves the efficiency of using system resources.

It should be noted that if the current resource of the cluster system is limited by the cost and the user's average expected value cannot be adjusted within the expected threshold, the range of the expected threshold may be changed to adapt to the hardware and software configuration of the current cluster system. Generally, the requirements of the cluster data access response time, the cluster cache hit ratio, or the cluster data throughput can be changed, especially the cluster cache hit ratio, which directly limits the concurrent access volume of the Memcached client, that is, directly limits the number of access request connections. The method provided by the embodiment is advantageous for maintaining the normal working state of the cluster system and finding a solution for solving the resource depletion.

Embodiment 4

FIG. 5 is a schematic structural diagram of a distributed cache control apparatus according to Embodiment 4 of the present invention. As shown in FIG. 5, the distributed cache control apparatus provided in this embodiment specifically includes: a collection module 11, a calculation module 12, a decision module 13 and an execution module 14.

The collecting module 11 is configured to collect client statistics of the cache node in the cluster system. The calculating module 12 is configured to calculate an average expected value of the user according to the client statistics collected by the collecting module 11.

The decision module 13 is configured to determine a resource adjustment manner for the cluster system according to the average expected value of the user obtained by the calculation module 12 and the expected threshold.

The executing module 14 is configured to perform a node contraction operation or a node expansion operation on the cache node in the cluster system according to the resource adjustment manner determined by the decision module 13.

In this embodiment, the decision module 13 is specifically configured to: the average expected value of the user is greater than the first When the threshold is expected, it is determined that the node shrink operation is performed on the cluster system; or, when the average expected value of the user is less than the second desired threshold, it is determined that the node expansion operation is performed on the cluster system. It should be noted that the first expected threshold and the second expected threshold are the expected threshold ranges. When the calculated average expected threshold exceeds the range, the resource of the cluster system needs to be adjusted accordingly; on the one hand, the average expected value is high. In this range, the user is more satisfied with the QoS of the cache node in the cluster system, that is, the cost of the current resource configuration of the cluster system is higher. Therefore, some cache nodes in the system can be shut down to reduce system resources. Consumption, on the other hand, when the average expected value is lower than the range, the user is less satisfied with the QoS of the cache node in the cluster system, that is, the current resource configuration of the cluster system cannot meet the user's needs, and therefore, the system can be increased. The number of cache nodes to increase user satisfaction and system efficiency.

It should be noted that, in this embodiment, when the average expected value of the user is within the range of the expected threshold, the user's satisfaction with the QoS is expected to be expected, and the resources of the cluster system need not be adjusted, and the collection module 11 can continue to execute. The operation of collecting the client statistics information of the cache node in the cluster system may be performed cyclically by the distributed cache control device provided in this embodiment.

The distributed cache control device provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the first embodiment of the present invention, and has a corresponding function module, and the implementation principle and the technical effect thereof are similar, and details are not described herein again.

Embodiment 5

FIG. 6 is a schematic structural diagram of a distributed cache control apparatus according to Embodiment 5 of the present invention. As shown in FIG. 6, the apparatus provided in this embodiment is based on the apparatus structure shown in FIG. 5, and the collection module 11 is further configured to collect the cluster system before the decision module 13 determines the resource adjustment mode of the cluster system. In the present embodiment, the collection module 11 does not have a clear sequence of collection of client statistical information and resource utilization information.

Correspondingly, the decision module 13 includes: an obtaining unit 15 configured to acquire, according to the resource utilization information of the cache node and the client statistical information collected by the collection module 11, the expected value of the cache node and the resource utilization information of the cache node. The determining unit 16 is configured to determine a resource adjustment manner of each cache node in the cluster system according to a desired value and a desired threshold of each cache node and a relationship curve acquired by the obtaining unit 15 .

In this embodiment, the resource utilization information of the cache node collected by the collection module 11 may include the The hardware parameter of the cache node, the operating system parameter, and the configuration parameter of the cache node in the cluster system, and the operating system parameter is used to calculate the load value of the cache node. In a specific implementation, the obtaining unit 15 is configured to obtain, according to the hardware parameter of the cache node, the load value of the cache node, the configuration parameter of the cache node in the cluster system, and the statistics information of the client, The expected value of the cache node may be represented by the function form (3) according to the hardware parameter of the cache node, the load value of the cache node, and the configuration parameter of the cache node in the cluster system, and the formula may be Through the long-term monitoring of the collection of historical experience data, for example, a function approximation algorithm can be used to obtain a curve of „随, „ and ∑„; accordingly, the determining unit 16 is specifically used to be greater than the first desired threshold or less than the second when the desired threshold, it is determined ", Y _n ^ and adjusting at least one of the resources of each cache node in the cluster system by changing the X, less than the first desired value such that the adjusted threshold and greater than a second desired threshold.

It should be noted that the distributed cache control apparatus provided in this embodiment separately formulates resource adjustment policies for each cache node in the cluster system, so that the user's satisfaction level with each cache node Qos can reach the expected value requirement.

In this embodiment, the execution module 14 is further configured to perform resource adjustment on each cache node in the cluster system according to the resource adjustment manner determined by the determining unit 16 and the resource utilization information of the cache node collected by the collection module 11.

The distributed cache control device provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the second embodiment of the present invention, and has a corresponding function module, and the implementation principle and the technical effect thereof are similar, and details are not described herein again.

In another implementation manner of the embodiment, the client statistics collected by the collection module 11 include a cluster data access response time, a cluster cache hit ratio, and a cluster data throughput. Accordingly, the calculation module 12 is specifically configured to The cluster data fetch response time, the cluster cache hit ratio, and the cluster data throughput, and the expected value of the user for each cache node is calculated as the above formula (4), and the average expected value of the user is calculated according to the expected value of the cache node by the user^: .

Further, the collection module 11 collects the resource utilization information of the cache node, and the operating system parameter therein can be used to calculate the load value of the cache node, that is, the calculation module 12, and is further configured to calculate the load value according to the cache node. The current average load value of the cluster system; correspondingly, the decision module 13 is further configured to determine, according to the current average load value and the load threshold of the cluster system, when determining that: less than the first expected threshold and greater than the second expected threshold Resource adjustment of the cluster system The whole mode; on the other hand, when it is determined that ^: is greater than the first desired threshold, or is less than the second desired threshold, the resource adjustment mode for the cluster system is determined according to the ^: and the expected threshold.

In a specific implementation, the determining module 13 determines the resource adjustment mode of the cluster system according to the current average load value and the load threshold of the cluster system, and specifically includes: when the current average load value of the cluster system is greater than the first load threshold. And determining to perform a node expansion operation on the cluster system; or, when the current average load value of the cluster system is less than a second load threshold, determining to perform the node shrink operation on the cluster system. The first load threshold and the second load threshold are load threshold ranges. When the calculated current load threshold of the cluster system exceeds the range, the resource of the cluster system needs to be adjusted accordingly. On the one hand, the current average load value is high. In this range, the current load of each cache node in the cluster system is high. Therefore, the system load can be reduced by increasing the number of cache nodes in the system. On the other hand, when the current average load value is lower than the range, The load of each cache node in the cluster system is not saturated. Therefore, the system resource consumption can be reduced by turning off some cache nodes in the system.

It should be noted that, in the embodiment, the current average load value of the cluster system is within the range of the load threshold, indicating that the resource configuration of the cluster system conforms to the load balancing principle on the premise that the user's satisfaction with the QoS is expected. The operation of the distributed cache control device provided in this embodiment may be performed by the collection module 11 to continue the operation of collecting the client statistics and the resource utilization information of the cache node in the cluster system. It is done in a loop.

The distributed cache control device provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the third embodiment of the present invention, and has a corresponding function module, and the implementation principle and the technical effect thereof are similar, and details are not described herein again.

Embodiment 6

FIG. 7 is a schematic structural diagram of a cluster control server according to Embodiment 6 of the present invention. As shown in FIG. 7, the cluster control server provided in this embodiment specifically includes: a collector 21, a calculator 22, and a processor 23.

The collector 21 is configured to collect client statistics of the cache node in the cluster system. The calculator 22 is configured to calculate the average expected value of the user based on the client statistics collected by the collector 21.

The processor 23 is configured to obtain an average expected value of the user according to the calculator 22, and a desired threshold The value determines how the resource is adjusted for the cluster system.

The processor 23 is further configured to perform a node shrink operation or a node expansion operation on the cache node in the cluster system according to the determined resource adjustment manner.

In this embodiment, the processor 23 determines that the resource adjustment manner of the cluster system is specifically: when the average expected value of the user is greater than the first expected threshold, determining that the node shrink operation is performed on the cluster system; or, the average expected value of the user When less than the second desired threshold, it is determined that the node expansion operation is performed on the cluster system. It should be noted that the first expected threshold and the second expected threshold are the expected threshold ranges. When the calculated average expected threshold exceeds the range, the resource of the cluster system needs to be adjusted accordingly; on the one hand, the average expected value is high. In this range, the user is more satisfied with the QoS of the cache node in the cluster system, that is, the cost of the current resource configuration of the cluster system is higher. Therefore, some cache nodes in the system can be shut down to reduce system resources. Consumption, on the other hand, when the average expected value is lower than the range, the user is less satisfied with the QoS of the cache node in the cluster system, that is, the current resource configuration of the cluster system cannot meet the user's needs, and therefore, the system can be increased. The number of cache nodes to increase user satisfaction and system efficiency.

It should be noted that, in this embodiment, when the average expected value of the user is within the range of the expected threshold, the user's satisfaction with the QoS is expected to meet the expected requirements, and the resources of the cluster system need not be adjusted, and the collector 21 can continue to execute. The operation of collecting the client statistics information of the cache node in the cluster system may be performed cyclically by the cluster control server provided in this embodiment.

The cluster control server provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the first embodiment of the present invention, and has a corresponding physical device. The implementation principle and technical effects are similar, and details are not described herein again.

Example 7

FIG. 7 is a schematic structural diagram of a cluster control server according to Embodiment 7 of the present invention. In this embodiment, the collector 21 is further configured to collect the resource utilization information of the cache node in the cluster system before the processor 23 determines the resource adjustment mode of the cluster system. 21 There is no clear sequence of collection of client-side statistics and resource utilization information.

Correspondingly, the determining, by the processor 23, the resource adjustment manner of the cluster system may be: obtaining the expected value of the user for each cache node and the cache node according to the resource utilization information and the client statistics information of the cache node collected by the collector 21 Resource utilization information relationship curve; The expected value of each cache node, the expected threshold, and the acquired relationship curve determine the resource adjustment mode of each cache node in the cluster system.

In this embodiment, the resource utilization information of the cache node collected by the collector 21 may include hardware parameters of the cache node, operating system parameters, and configuration parameters of the cache node in the cluster system, where the operating system parameters are used for calculation. The load value of this cache node. In a specific implementation, the processor 23 obtains a relationship between the expected value of the cache node and the resource utilization information of the cache node, and may be: according to the hardware parameter of the cache node, the load value of the cache node, and the a configuration parameter of the cache node in the cluster system, and the client statistics, obtaining a user expected value for each cache node according to a hardware parameter of the cache node, a load value of the cache node, and the cache node in the cluster system The relationship between the configuration parameters can be expressed in the form of a function (3) above, which can be obtained by long-term monitoring of the collection of historical empirical data. For example, a function approximation algorithm can be used to obtain a curve of „, 、, and ∑„. Correspondingly, the processor 23 determines a resource adjustment manner of each cache node in the cluster system, which may be: when the value is greater than the first desired threshold or less than the second desired threshold, the determination is determined by changing at least X, ynz A resource adjustment for each cache node in the cluster system So that the adjusted is less than the first desired threshold and greater than the second desired threshold.

It should be noted that the cluster control server provided in this embodiment separately formulates resource adjustment policies for each cache node in the cluster system, so that the user can satisfy the expected value of each cache node Qos.

In this embodiment, the processor 23 is further configured to adjust the manner and the collector according to the determined resource.

The resource utilization information of the collected cache nodes is separately adjusted for resources in each cache node in the cluster system.

The cluster control server provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the second embodiment of the present invention, and has a corresponding physical device, and the implementation principle and technical effects thereof are similar, and details are not described herein again.

In another implementation manner of the embodiment, the client statistics collected by the collector 21 include a cluster data access response time, a cluster cache hit ratio, and a cluster data throughput. Accordingly, the calculator 22 is specifically configured to The cluster data fetch response time, the cluster cache hit ratio, and the cluster data throughput, and the expected value of the user for each cache node is calculated as the above formula (4), and the average expected value of the user is calculated according to the expected value of the cache node by the user ^: . Further, the collector 21 collects the resource utilization information of the cache node, and the operating system parameter therein can be used to calculate the load value of the cache node, that is, the calculator 22, and is further configured to calculate the load value according to the cache node. The current average load value of the cluster system; correspondingly, the processor 23 is further configured to determine, according to the current average load value and the load threshold of the cluster system, when determining that: less than the first expected threshold and greater than the second expected threshold The resource adjustment mode of the cluster system; on the other hand, when it is determined that ^ is greater than the first desired threshold or less than the second desired threshold, the resource adjustment mode for the cluster system is determined according to the ^: and the expected threshold.

In a specific implementation, the processor 23 determines the resource adjustment mode of the cluster system according to the current average load value and the load threshold of the cluster system, and specifically includes: when the current average load value of the cluster system is greater than the first load threshold. And determining to perform a node expansion operation on the cluster system; or, when the current average load value of the cluster system is less than a second load threshold, determining to perform the node shrink operation on the cluster system. The first load threshold and the second load threshold are load threshold ranges. When the calculated current load threshold of the cluster system exceeds the range, the resource of the cluster system needs to be adjusted accordingly. On the one hand, the current average load value is high. In this range, the current load of each cache node in the cluster system is high. Therefore, the system load can be reduced by increasing the number of cache nodes in the system. On the other hand, when the current average load value is lower than the range, The load of each cache node in the cluster system is not saturated. Therefore, the system resource consumption can be reduced by turning off some cache nodes in the system.

It should be noted that, in the embodiment, the current average load value of the cluster system is within the range of the load threshold, indicating that the resource configuration of the cluster system conforms to the load balancing principle on the premise that the user's satisfaction with the QoS is expected. The operation of collecting the cluster statistics of the cache node in the cluster system may be performed by the collector 21, and the execution process of the cluster control server provided in this embodiment may be a loop. ongoing.

The cluster control server provided by the embodiment of the present invention is configured to perform the distributed cache control method provided by the third embodiment of the present invention, and has a corresponding physical device. The implementation principle and technical effects are similar, and details are not described herein again.

It will be understood by those skilled in the art that all or part of the steps of implementing the above method embodiments may be performed by hardware related to the program instructions. The aforementioned program can be stored in a computer readable storage medium. The program, when executed, performs the steps including the various method embodiments described above; The foregoing storage medium includes various media that can store program codes, such as a ROM, a RAM, a magnetic disk, or an optical disk.

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

Claim

A distributed cache control method, comprising:

Collect client statistics of cache nodes in the cluster system;

Calculating an average expected value of the user according to the client statistical information;

Determining a resource adjustment manner for the cluster system according to an average expected value of the user and a desired threshold;

The method according to claim 1, wherein the determining a resource adjustment manner for the cluster system according to the user expected value and a desired threshold includes:

The method according to claim 1 or 2, wherein the client statistical information includes the cluster data access response response time, a cluster cache hit ratio, and a cluster data throughput;

And the calculating the average expected value of the user according to the client statistics, including: calculating, according to the cluster data access response time, the cluster cache hit ratio, and the cluster data throughput, the user to each cache The expected value of the node is: wherein, the data fetch response time indicating the cache node i, the _^ represents the cache hit ratio of the cache node i, and the 7/, ._^ represents the cache node i Data throughput, the l ^ and Λ represent weights;

Calculating the average expected value of the user ^: according to the expected value of the cache node by the user.

The method according to claim 3, wherein, before determining the resource adjustment manner for the cluster system according to the average expected value and the expected threshold of the user, the method further includes:

Obtaining, according to the collected resource utilization information of the cache node and the client statistical information Determining, by the user, a relationship between the expected value of each cache node and the resource utilization information of the cache node; determining, according to the expected value of each cache node and the expected threshold, and the relationship curve, for each of the cluster systems The resource adjustment mode of the cache node;

The method according to claim 4, wherein the resource utilization information of the cache node includes hardware parameters of the cache node, operating system parameters, and configuration parameters of the cache node in the cluster system. The operating system parameter is used to calculate a load value of the cache node; and the obtaining, according to the collected resource utilization information of the cache node and the client statistical information, the expected value and location of each cache node by the user. The relationship curve of the resource utilization information of the cache node includes:

E _n ^ f(X _n , Y _n , Z ;

The X indicates the hardware parameter of the cache node, the load value of the cache node, and the parameter indicates the configuration parameter of the cache node in the cluster system;

Determining, by changing at least one of the X, the 1 and the Z, to each cache node in the cluster system when the „ is greater than the first desired threshold or less than the second desired threshold Resource adjustment is performed such that the adjusted „ is less than the first desired threshold and greater than the second desired threshold.

The method according to claim 5, further comprising: calculating a current average load value of the cluster system according to a load value of the cache node;

Determining that the ^: is smaller than the first expected threshold and greater than the second expected threshold, determining resource adjustment to the cluster system according to a current average load value and a load threshold of the cluster system the way.

The method according to claim 6, wherein the determining a resource adjustment manner for the cluster system according to a current average load value and a load threshold of the cluster system, including: When the average load value is greater than the first load threshold, determining to perform the node expansion operation on the cluster system; or

8. A distributed cache control device, comprising:

a collection module, configured to collect client statistics of a cache node in the cluster system; and a calculation module, configured to calculate an average expected value of the user according to the client statistics collected by the collection module;

The device according to claim 8, wherein the determining module is configured to: when the average expected value of the user is greater than a first expected threshold, determine to perform the node shrinking operation on the cluster system; Alternatively, when the average expected value of the user is less than a second expected threshold, determining to perform the node expansion operation on the cluster system.

The device according to claim 8 or 9, wherein the client statistical information includes the cluster data memory response response time, a cluster cache hit ratio, and a cluster data throughput; And calculating, according to the cluster data memory response response time, the cluster cache hit ratio, and the cluster data throughput, an expected value of the user for each cache node:

E _n

_ cachehit i _data;

The data storage response time indicating the cache node i, the _^^ indicates the cache hit ratio of the cache node i, and the TH, . _ indicates the data throughput of the cache node i, And ^ represent weights; and then calculate an average expected value of the user ^: according to the expected value of the user for each cache node.

The device according to claim 10, wherein the collecting module is further used for Collecting resource utilization information of the cache node in the cluster system before the determining module determines a resource adjustment manner of the cluster system;

The device according to claim 11, wherein the resource utilization information of the cache node includes hardware parameters of the cache node, operating system parameters, and configuration parameters of the cache node in the cluster system. The operating system parameter is used to calculate a load value of the cache node;

E _n = f(X _nn , Z _n ) ;

The determining unit is specifically configured to determine, by changing at least one of the X, the 1 and the Z, when the „ is greater than the first desired threshold or less than the second desired threshold Performing resource adjustment on each cache node in the cluster system such that the adjusted value is less than the first desired threshold and greater than the second desired threshold.

The device according to claim 12, wherein the calculating module is further configured to calculate a current average load value of the cluster system according to a load value of the cache node;

The decision module is further configured to determine that the ^ is less than the first desired threshold and When the second expected threshold is greater than the current average load value and the load threshold of the cluster system, a resource adjustment manner for the cluster system is determined.

The apparatus according to claim 13, wherein the determining module determines a resource adjustment manner for the cluster system according to a current average load value and a load threshold of the cluster system, specifically: When the current average load value of the system is greater than the first load threshold, determining to perform the node expansion operation on the cluster system; or determining, when the current average load value of the cluster system is less than the second load threshold, determining the cluster The system performs the node contraction operation.