TW201327205A - Managing method for hardware performance and cloud computing system - Google Patents

Abstract

A managing method for hardware performance and a cloud computing system are provided. The cloud computing system includes a plurality of node devices deployed with a plurality of node resource pools and a managing node for performing the managing method. The managing method for hardware performance includes the following steps. A load for the node resource pools are detected to judge a bottleneck and a corresponding bottleneck resource pool. At least one of convert nodes is evaluated and selected among the node devices of the node resource pools other than the bottleneck resource pool. The at least one of convert nodes is changed and deployed from its original node resource pool to the bottleneck resource pool.

Description

Hardware performance management method and cloud computing system

The present invention relates to a performance management technology for cloud computing, and in particular to a hardware performance management method and a cloud computing system.

Cloud Computing technology is a combination of a large number of servers (or nodes) through the Internet to form high-speed computing and integrated storage with a large amount of storage capacity, which emphasizes limited resources at the local end. Use the network to obtain remote computing resources, storage resources or services. Cloud computing technology can share or divide these nodes through virtualization and automation technologies, and operate the web pages of these services through terminal interfaces such as the Internet and browsers to perform various operations and work.

These numerous nodes are aggregated into a server group. Due to the large number of these nodes, how can the server cluster automatically cancel the bottleneck and provide higher performance when a certain bottleneck of the server cluster causes bottlenecks and affects the overall performance of the system. An important topic in many cloud computing systems.

The invention provides a hardware performance management method and a cloud computing system, which detects whether a bottleneck phenomenon occurs in various node resource pools, and automatically adjusts and reallocates the role division of the servers, thereby effectively solving the bottleneck phenomenon automatically. Cloud computing systems provide higher performance.

The invention provides a hardware performance management method suitable for a cloud computing system. The cloud computing system includes a plurality of node devices, and the node devices are configured in a plurality of node resource pools. The management method includes the following steps. The load of the resource pools of these nodes is detected to determine a bottleneck phenomenon and a bottleneck resource pool corresponding to the bottleneck. Wherein the node resource pools comprise the node resource pools. At least one transition node is evaluated and selected from these node devices of other node resource pools other than the bottleneck resource pool. The conversion node is changed to redistribute the conversion node from the original node resource pool to the bottleneck resource pool.

In an embodiment of the present invention, the foregoing management method further includes the following steps. A normal threshold and a bottleneck threshold are respectively set for each node resource pool. When the load of one of the resource pools of the nodes is lower than the corresponding normal threshold, it indicates that one of the resource pools of the nodes is in a normal state. When the load of one of the node resource pools is higher than the corresponding bottleneck threshold, it indicates that the bottleneck phenomenon occurs in one of the node resource pools and becomes the bottleneck resource pool.

In an embodiment of the invention, evaluating the conversion node comprises the following steps. After the bottleneck threshold is estimated to estimate that the conversion node is allocated from the original node resource pool to the bottleneck resource pool, the load of the original node resource pool and the load of the bottleneck resource pool should each be smaller than the corresponding bottleneck threshold.

In an embodiment of the invention, changing the conversion node comprises the following steps. Retrieving a node-related database and obtaining node-related data of the conversion node. Adjusting node related data of the conversion node, so that the conversion node is modified from the original node resource pool to the bottleneck resource pool. The conversion node is isolated from the cloud computing system. Adjusting the conversion node according to the bottleneck resource pool. And, the conversion node is rejoined to the cloud computing system.

In an embodiment of the invention, isolating the conversion node comprises the following steps. Migrating a plurality of virtual machines in the conversion node from the conversion node to other node devices of the original node resource pool. And, closing a plurality of service programs executed by the conversion node.

In an embodiment of the invention, isolating the conversion node further comprises the following steps. A node association database is set to isolate node related data of the conversion node.

In an embodiment of the invention, the load of the node resource pool includes a respective operational load, a spatial load, and/or a combination thereof of the node resource pools.

In an embodiment of the present invention, the node resource pool includes a service resource pool, a computing resource pool, a storage resource pool, and/or a combination thereof.

From another point of view, the present invention provides a cloud computing system that includes a plurality of node devices and a management node. The node devices are coupled to each other through a network and configured in a plurality of node resource pools. The management node is coupled to the node device through the network to detect the load of the resource pools of the nodes, and determine the bottleneck phenomenon and the bottleneck resource pool corresponding to the bottleneck, wherein the node resource pool includes the node resource pool. The management node evaluates and selects at least one transition node from node devices of other node resource pools other than the bottleneck resource pool, and alters the transition node to redistribute the transition node from the original node resource pool to the bottleneck resource pool.

Please refer to the above description for the remaining implementation details of the cloud computing system, and will not be described here.

Based on the above, the cloud computing system of the embodiment of the present invention sets different load upper limits for each node resource pool, and detects the operation status of each node resource pool. When a bottleneck occurs in a specific node resource pool and no other redundant nodes are available for support, the cloud computing system can select some nodes from the normal operation node and no bottlenecks, and put them into the specific node resource pool. (In other words, it is to redistribute the role division of some nodes) to reduce the bottleneck. Therefore, by automatically adjusting and redistributing the role division of these servers, the cloud computing system can effectively solve bottlenecks automatically and improve its hardware operation performance, providing higher performance.

The above described features and advantages of the present invention will be more apparent from the following description.

DETAILED DESCRIPTION OF THE INVENTION Reference will now be made in detail to the exemplary embodiments embodiments In addition, wherever possible, the elements and/

FIG. 1 is a schematic diagram of a cloud computing system 100 according to an embodiment of the present invention. For example, an embodiment of the present invention provides a Infrastructure as a Service (IaaS)-based Container Data Center. As the cloud computing system 100. As shown in FIG. 1, the cloud computing system 100 of this embodiment may include at least one container. In the cabinet described in this embodiment, each cabinet includes multiple racks (RACKs), each rack also has multiple slots, and each slot may also include one or more servers ( Or called a node device). Each cabinet has a similar composition. For convenience of description, a cabinet is taken as an example in this embodiment.

Referring to FIG. 1 , the cloud computing system 100 includes a plurality of node devices. When the cloud computing system of the cloud computing system 100 is deployed, the cloud computing system 100 is configured in a plurality of node resource pools. In other words, the node devices can be classified into three types of nodes, that is, a service resource pool 110, a computing resource pool 120, and a storage resource pool 130, and the service resources. Pool 110 can also be more detailed according to its service functions. Thus, the node resource pool includes a service resource pool 110, a computing resource pool 120, a storage resource pool 130, and/or a combination thereof.

In this embodiment, the service resource pool 110 includes i node devices 112_1~112_i, the computing resource pool 120 includes j node devices 122_1~122_j, and the storage resource pool 130 includes k node devices 132_1~132_k, i, j, k. All are non-negative integers. The node devices 112_1~112_i, the node devices 122_1~122_j, and the node devices 132_1~132_k may also be referred to as service nodes 112_1~112_i, compute nodes 122_1~122_j, and storage nodes 132_1~132_k, respectively, in this embodiment. The node devices are all coupled to the second layer switch 140, so as to be coupled to each other, communicate, and transmit information through the area network. The embodiments of the present invention may also be coupled to the node devices through other types of network methods, such as the Internet, the wireless network, etc., and are not described herein.

The service resource pool 110 and the service nodes located therein can be subdivided according to the service function, such as a physical installer service, a physical manager service, a log processing service, and a virtual manager service. , Application Programming Interface (API) services, virtual resource provisioning services, database services, storage manager services, load balance services, and security services. ..Wait. The computing resource pool 120 and the computing nodes located therein are used to provide computing services. The storage resource pool 130 and the storage nodes located therein are used to provide storage services.

In other words, the service nodes 112_1~112_i mainly provide services of a plurality of virtual machines (VMs), which are executed by the computing resources pool 120 composed of the computing nodes 122_1~122_j. The required storage space is provided by the storage resource pool 130 composed of the storage nodes 132_1~132_k. Each service node 112_1~112_i provides different services for the user according to the different software they execute. In contrast, the computing nodes 122_1~122_j disposed in the computing resource pool 120 or the storage nodes 132_1~132_k in the storage resource pool 130 respectively perform similar software programs, so that they are easy to integrate with each other for huge computation or storage. data.

The cloud computing system also includes a management node for monitoring and adjusting the load of each node device. The management node may be one of the foregoing node devices or another monitoring device independent of the node device. In this embodiment, the node device 112_2 located in the service resource pool 110 serves as a management node. The management node 112_2 includes a bottleneck monitoring module 150, a node selection module 160, a data access module 170, a node isolation module 180, a node deployment module 190, and a node addition module 195. These functional modules will be described in detail below. Description. In addition, when the cloud operating system deploys each node device, node related data corresponding to each node device is obtained, and the cloud operating system integrates the node related data into the node related database DB. For reference by the management node 112_2. In this embodiment, the node association database DB is disposed in the service node 112_1, but the present invention is not limited thereto. Other embodiments may also place the node association database DB in any node device.

Although some of the node devices are specifically applied to the specific roles of the cloud computing system 100 (for example, dedicated to a specific node resource pool), some types of node devices can support multiple cloud resources, and are not limited to playing the cloud computing system. 100 specific roles. For example, some node devices have much better computing performance than other node devices, but their ability to provide services or store data is far less than other nodes. In this case, these node devices can be specifically classified into the computing resource pool 120. As a compute node. However, many node devices have good computing performance, can provide better services and data storage, and thus can be used as service nodes, storage nodes or computing nodes. That is to say, such a node device is not limited to being a specific role of the cloud computing system 100 only because of its hardware design. Although the more compatible node devices have been determined to belong to a specific node resource pool when the cloud operating system is configured, in a specific case, the management module 112_2 also changes the roles of the node devices.

The so-called "Bottleneck" is the performance load or space load in the resource pool of each node of the cloud computing system 100, and there is no other spare node available for support. The cloud computing system 100 is already in a bottleneck. For example, when the average usage rate of the central processing unit (CPU) of each of the computing nodes 122_1 122 122_j in the entire computing resource pool 120 is too high, it is referred to as a performance bottleneck. For another example, when the storage space remaining in the storage nodes 122_1~122_k in the storage resource pool 130 is insufficient, it is referred to as a space bottleneck.

In this case, when the cloud computing system 100 has detected a bottleneck and the entire cabinet has no redundant standby nodes, the cloud computing system 100 of the embodiment of the present invention can select certain nodes from the node resource pool where no bottleneck occurs. The device performs role change, so that the selected node devices act as one of the node resource pools that are bottlenecks, thereby eliminating the bottleneck phenomenon of the entire cloud computing system 100. Of course, the embodiment of the present invention must consider whether the changed node devices can bear the performance of the switched node resource pool.

The following is a description of the management method of hardware performance by the cloud computing system 100 to which it is applied. 2 is a flow chart showing a method for managing hardware performance according to an embodiment of the invention. Referring to FIG. 1 and FIG. 2, in step S210, the bottleneck monitoring module 150 in the management node 112_2 detects the load of the node resource pools 110-130, and in step S220, the bottleneck monitoring module 150 determines whether it occurs. The bottleneck phenomenon and the node resource pool corresponding to the bottleneck. Here, the node resource pool corresponding to the bottleneck phenomenon is called a bottleneck resource pool.

In this embodiment, the management node 112_2 sets a normal threshold and a bottleneck threshold for each of the node resource pools 110-130 to determine the current operation of each node resource pool. In detail, the bottleneck monitoring module 150 of the embodiment detects the load of each node device, and the load of each node device includes the computing load of each node device and the space load (that is, the used storage space), and The average load of each node resource pool 110~130 is calculated through the node related database DB. Thereby, the load of the node resource pool includes the average computational load, spatial load, and/or a combination thereof of the respective node devices in the resource pools of the nodes.

In this embodiment, the normal threshold of the performance bottleneck is set to 70%, and the bottleneck threshold of the performance bottleneck is set to 80%. That is to say, when the average performance rate of the CPU in the node resource pool is less than 70%, and the average performance of the CPU in the original point resource pool after the change is changed, the average performance rate of the CPU in the original point resource pool needs to be lower than 70%. %. On the other hand, when the average CPU usage of the CPU in the node resource pool is greater than 80%, and the role of the installed node device changes, the bottleneck resource pool after the change should be less than 80% to meet the assessment.

In this embodiment, the normal threshold of the space bottleneck is set to 80%, and the bottleneck threshold of the space bottleneck is set to 90%. That is to say, when the value of the used resource pool (used space/all storage space) is less than 80%, and after a node device changes the role, the original point resource pool is changed (used space/all) The value of the storage space still needs to be less than 80%. The value of the used resource pool (used space/all storage space) in the node resource pool is greater than 80%, and the value of the changed bottleneck resource pool (used space/all storage space) after a node device is changed. It should be less than 80% to meet its assessment.

After the load of each node resource pool is calculated, when the load of the node resource pool is lower than the corresponding normal threshold, the bottleneck monitoring module 150 can determine that the resource pools of the nodes are in a normal state, and no bottleneck occurs. When the load of the node resource pool is higher than the corresponding normal threshold but lower than the corresponding bottleneck threshold, the bottleneck monitoring module 150 can determine that the resource pool of the node is in a high load condition, but has not yet reached the bottleneck phenomenon described above. 』. However, when the load of the node resource pool has reached or is higher than its corresponding bottleneck threshold, it indicates that the corresponding load of the resource pool of the node is near full load, and the bottleneck monitoring module 150 can determine that the resource pool of the node has occurred. "Bottling phenomenon". The "bottleneck phenomenon" is, for example, that the performance load in the computing resource pool 120 is insufficient to load the current computing amount, or the storable space in the storage resource pool 130 is already lower than the preset spare space.

For convenience of explanation, the present embodiment assumes that the storage resource pool 130 has a problem of a space bottleneck. Therefore, when the bottleneck monitoring module 150 determines that the bottleneck phenomenon has occurred, and has determined that the node resource pool corresponding to the bottleneck phenomenon (that is, the storage resource pool 130) has occurred, the process proceeds from step S220 to step S230, and the node selection module 160 At least one conversion node is evaluated and selected from a node device of a node resource pool other than the bottleneck resource pool. In other words, the node selection module 160 selects node devices that can serve as storage nodes from other node resource pools that have no bottlenecks (for example, the service resource pool 110 or the computing resource pool 120), and evaluates that the node devices are After the role change is made, whether or not the cloud computing system 100 can be surely prevented from bottlenecking.

In order to prevent the node devices of the node data pools that are already in a high load state from continuously performing role transformation, the node selection module 160 of this embodiment will be from a node resource pool located in a normal phenomenon (that is, its load is low). Selecting the node device whose role is to be changed in the node data pool of the normal threshold value is not selected in the node data pool located in the high load state. In addition, the node selection module 160 also needs to estimate whether the above-mentioned conversion node after the role change (that is, after the allocation from the original node resource pool to the bottleneck resource pool) causes the original node data pool and the bottleneck according to the bottleneck threshold value. The load of the resource pool is less than the corresponding bottleneck threshold, so that it is expected that the cloud computing system 100 has no bottleneck after the role transition. The node selection module 160 of this embodiment can intuitively calculate the average performance load of the central processing unit of each node device in the same node resource pool, or intuitively calculate whether the used storage space is exceeded, to determine whether it is The bottleneck threshold has been reached or exceeded.

For example, assume that the node selection module 160 selects the computing node 122_2 as a conversion node in the computing resource pool 120 located in the normal phenomenon, the computing node 122_2 can serve as a storage node, and the computing node 122_2 can achieve the above effects after the node selection module 160 evaluates. The node selection module 160 treats these node devices as conversion nodes, thereby continuing the steps described below.

In step S240, the management node 112_2 changes the node-related data of the conversion node 122_2 in the node-associated database DB, thereby reallocating the conversion node 122_2 from the original node resource pool (that is, the computing resource pool 120) to the bottleneck resource pool. (that is, storage resource pool 130). The operation flow of step S240 can be subdivided into steps S250 to S290, and the steps are detailed one by one through the node related information table in the node association database DB.

When the cloud operating system of the cloud computing system 100 configures the node device, the node related information records of the respective node devices are organized into a node related information table. The above node related information can be obtained in the following ways. For example, when the basic input/output system (BIOS) of each node device performs a power-on self-test (POST) program, it dynamically obtains its node-related data (for example, central processing unit, memory, hard disk, network). Cards and other related information), and obtain other node-related data through, for example, SMBIOS data structures (types 0, 1, 2, and OEM types) and MAC addresses in the network card EEPROM (for example, products of node devices) Data, BIOS information, node type), and finally transfer these data to the Baseboard Management Controller (BMC) of each node device via IPMI OEM instructions. In addition, the BMC can dynamically obtain information such as the BMC network card, such as the BMC network card MAC address, IP address and its bandwidth, to enrich its node related information. The following uses Table (1) as an example of the node-related database DB and the information about the nodes therein. The five pieces of node related information recorded in the table (1) are sequentially taken from the service nodes 112_1, 112_i, the computing nodes 122_1, 122_2, and the storage node 132_1 of FIG. 1, respectively.

Table (1) includes 10 fields, respectively recording the MAC address of the Baseboard Management Controller (BMC) network card in each node device, the IP address and bandwidth of the BMC network card, the MAC address of the system network card, and the IP of the system network card. Address and bandwidth, processor information (model/operation rate), memory information, hard drive information, node location, node type, and server type. The IP address of the system NIC is obtained through Network Boot.

The node address information of the node device 112_1 is that the MAC address of the BMC network card is "00:A0:D1:EC:F8:B1", and the IP address of the allocated BMC network card is "10.1.0.1". The BMC NIC has a bandwidth of 100 Mbps (bps=bits per second). The MAC address of the system network card of the node device 112_1 is "00: A0: D1: EA: 34: E1", the IP address is "10.1.0.11", and the bandwidth is 1000 Mbps. Further, the model of the central processing unit of the node device 112_1 is "Intel(R) Xeon(R) CPU E5540", and its operation frequency is "2530 MHz". The node device 112_1 includes four memory modules, DIMM1~DIMM4, and each memory module has a capacity of 8G. Further, the carrier number of the hard disk of the node device 112_1 is 1, the hard disk type is SAS (Serial Attached SCSI, SCSI = Small Computer System Interface), the hard disk capacity is 1 TB, and the hard disk rotation speed is 7200 RPM (Revolution). Per Minute) and the hard drive cache capacity is 16MB.

Referring back to FIG. 2 and referring to FIG. 1 simultaneously, in step S250, the material access module 170 retrieves the node-related database DB and obtains node-related data of the conversion node 122_2. That is, the material access module 170 can obtain node related data such as the following table (2) from the node-related database DB.

In step S260, the data access module 170 adjusts the node-related data of the conversion node 122_2 in the table (2), and restores the node-related data to the node-related database DB in the service node 112_1. This embodiment is Modify the field "node type" and "server type" in table (2) to be changed from the original "computing node" to "storage node" (as shown in the following table (3)), so that the conversion node 122_2 is from the original node. The compute node 122_2 of the resource pool (computation resource pool 120) modifies to the storage node of the bottleneck resource pool (storage resource pool 130). In other embodiments, the data access module 170 also sets the node-related database DB at this time and isolates the node-related data of the conversion node 112_2 so that other node devices cannot access the conversion node 112_2.

Then, in step S270, the node isolation module 180 isolates the conversion node 122_2 from the cloud computing system 100. In detail, the node isolation module 180 performs a number of processes to isolate the conversion node 122_2 from the cloud computing system 100. For example, the node isolation module 180 migrates a plurality of virtual machines (VMs) running in the conversion node 122_2 from the conversion node to the other node devices 122_1 122 122_j of the computing resource pool 120. And, after the transfer of the virtual machine described above, the node isolation module 180 closes all service programs executed on the conversion node 122_2.

In step S280, the node deployment module 190 adjusts the conversion node 122_2 according to the bottleneck resource pool (the storage resource pool 130). The node deployment module 190 redeploys the conversion node 122_2 according to the adjusted node type/server type, that is, the operation system required to install the bottleneck resource pool (storage resource pool 130) for the conversion node 122_2, and After the installation of the above operating system is completed, the service packages of all the storage nodes must be installed to make the conversion node 122_2 meet the requirements of the storage resource pool 130.

Finally, as shown in FIG. 3, FIG. 3 is another schematic diagram illustrating the cloud computing system 100 according to an embodiment of the present invention. Referring to FIG. 2, in step S290, the node adding module 195 will re-switch the node 122_2. The cloud computing system 100 is added and converted from the computing node 122_2 of the original node resource pool (the computing resource pool 120) to the storage node 132_x in the bottleneck resource pool (the storage resource pool 130) (as indicated by the dashed arrow 300). The data access module 170 also sets the node association database DB at this time and reopens the node related data of the original conversion node 112_2 (that is, the storage node 132_x of FIG. 3), so that other node devices can access the storage node 132_x. .

In summary, the cloud computing system in the embodiment of the present invention sets different load upper limits for each node resource pool, and detects the operation status of each node resource pool. When a bottleneck occurs in a specific node resource pool and no other redundant nodes are available for support, the cloud computing system can select some nodes from the normal operation node and no bottlenecks, and put them into the specific node resource pool. (In other words, it is to redistribute the role division of some nodes) to reduce the bottleneck. Therefore, by automatically adjusting and redistributing the role division of these servers, the cloud computing system can effectively solve bottlenecks automatically and improve its hardware operation performance, providing higher performance.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the invention, and any one of ordinary skill in the art can make some modifications and refinements without departing from the spirit and scope of the invention. The scope of the invention is defined by the scope of the appended claims.

100. . . Cloud computing system

110. . . Service resource pool

112_1~112_i. . . Service node

120. . . Computing resource pool

122_1~122_j. . . calculate node

130. . . Storage resource pool

132_1~132_k. . . Storage node

140. . . switch

150. . . Bottleneck monitoring module

160. . . Node selection module

170. . . Data access module

180. . . Node isolation module

190. . . Node deployment module

195. . . Node add module

300. . . Dotted arrow

DB. . . Node related database

S210~S290. . . step

1 is a schematic diagram illustrating a cloud computing system in accordance with an embodiment of the present invention.

2 is a flow chart showing a method for managing hardware performance according to an embodiment of the invention.

FIG. 3 is another schematic diagram illustrating a cloud computing system in accordance with an embodiment of the invention.

100. . . Cloud computing system

110. . . Service resource pool

112_1~112_i. . . Service node

120. . . Computing resource pool

122_1~122_j. . . calculate node

130. . . Storage resource pool

132_1~132_k. . . Storage node

140. . . switch

150. . . Bottleneck monitoring module

160. . . Node selection module

170. . . Data access module

180. . . Node isolation module

190. . . Node deployment module

195. . . Node add module

DB. . . Node related database

Claims (10)

A hardware performance management method is applicable to a cloud computing system, the cloud computing system includes a plurality of node devices, and the node devices are configured in a plurality of node resource pools, and the management method includes: detecting the node resource pools Load to determine a bottleneck phenomenon and a bottleneck resource pool corresponding to the bottleneck phenomenon, wherein the node resource pool includes the node resource pool; from the node devices of the other node resource pools other than the bottleneck resource pool Evaluating and selecting at least one transition node; and changing the at least one transition node to reallocate the at least one transition node from the original node resource pool to the bottleneck resource pool. For example, the management method described in claim 1 further includes: setting a normal threshold and a bottleneck threshold for each of the node resource pools; when the load of one of the node resource pools is lower than the corresponding one When the normal threshold is used, one of the node resource pools is in a normal phenomenon; and when the load of one of the node resource pools is higher than the corresponding threshold value of the bottleneck, the bottleneck occurs in one of the node resource pools And become the bottleneck resource pool. For the management method described in claim 2, the evaluating the at least one conversion node includes the following steps: after estimating the at least one conversion node from the original node resource pool to the bottleneck resource pool according to the bottleneck threshold value, the original The load of the node resource pool and the load of the bottleneck resource pool are each smaller than the corresponding bottleneck threshold. The management method according to claim 1, wherein the changing the at least one conversion node comprises the steps of: retrieving a node-related database and obtaining node-related data of the at least one conversion node; and adjusting node correlation of the at least one conversion node. Data, such that the at least one transition node is modified from the original node resource pool to the bottleneck resource pool; the at least one transition node is isolated from the cloud computing system; the at least one transition node is adjusted according to the bottleneck resource pool; At least one conversion node rejoins the cloud computing system. The management method of claim 4, the isolating the at least one conversion node comprises the steps of: migrating the plurality of virtual machines of the at least one conversion node from the at least one conversion node to the other of the original node resource pool a node device; and shutting down a plurality of service programs executed by the at least one conversion node. For the management method described in claim 5, isolating the at least one conversion node further includes the step of setting the node association database to isolate node related data of the at least one conversion node. The management method of claim 1, wherein the load of the node resource pools includes an operational load, a spatial load, and/or a combination thereof of the node resource pools. The management method of claim 1, wherein the node resource pool comprises a service resource pool, a computing resource pool, a storage resource pool, and/or a combination thereof. A cloud computing system includes: a plurality of node devices coupled to each other through a network and configured in a plurality of node resource pools; and a management node coupled to the node devices to detect the The load of the node resource pool is used to determine a bottleneck phenomenon and a bottleneck resource pool corresponding to the bottleneck phenomenon, wherein the node resource pool includes the node resource pool, and the management node is from other node resources other than the bottleneck resource pool. The at least one transition node is evaluated and selected in the node devices of the pool, and the at least one transition node is changed to reallocate the at least one transition node from the original node resource pool to the bottleneck resource pool. The cloud computing system of claim 9, wherein the management node comprises: a bottleneck monitoring module, and respectively setting a normal threshold and a bottleneck threshold for each of the node resource pools, when the nodes When the load of one of the resource pools is lower than the corresponding normal threshold, it is determined that one of the node resource pools is in a normal state, and when the load of one of the node resource pools is higher than the corresponding threshold value of the bottleneck, Determining that the bottleneck phenomenon occurs in one of the node resource pools and becomes the bottleneck resource pool; a node selection module estimates the at least one conversion node according to the bottleneck threshold value, wherein the at least one conversion node is from the original node resource pool After being allocated to the bottleneck resource pool, the load of the original node resource pool and the load of the bottleneck resource pool are each smaller than the corresponding bottleneck threshold; and a data access module retrieves a node-related database to obtain the at least one Converting node-related data of the node, and adjusting node-related data of the at least one conversion node, so that the at least one conversion node The node resource pool is modified to the bottleneck resource pool; a node isolation module isolates the at least one conversion node from the cloud computing system; and a node deployment module adjusts the at least one conversion node according to the bottleneck resource pool; A node adds a module, and the at least one conversion node is rejoined to the cloud computing system.