JP2018148382A - Network resource management device, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a network resource management device capable of leveling network resources including switches with different performance.SOLUTION: A network resource management device 100 that manages a resource of each switch arranged in a network includes: a performance estimating unit 10 that estimates the performance of each switch from the usage rate of the switch and a flow flowing through the switch; a first load calculating unit 20 that calculates a first load of the switch; a configuration change candidate determination unit 30 that determines a candidate for changing a configuration of a network to which a target switch, which is a switch whose calculated first load is higher than a predetermined threshold value; and a second load calculating unit 40 that calculates the second load of the switch with the estimated performance, based on the configuration of the network determined to be a candidate after the change.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a network resource management apparatus, a network resource management method, and a network resource management program that manage network resources including a plurality of switches.

  In some systems, it is difficult to avoid hardware resource depletion and bias. Therefore, the system administrator usually grasps the resource status, sets a maintenance period, changes the configuration, and increases hardware resources.

  On the other hand, in the case of a virtual environment, there is a technique for moving a VM (Virtual Machine) according to CPU (Central Processing Unit) and memory resources. Therefore, even if a specific VM has a high load, the configuration can be changed flexibly to some extent to solve the high load state.

  Patent Document 1 describes a virtual machine placement device that places each virtual machine on a physical server so as to guarantee the communication quality between the virtual machines. The virtual machine placement apparatus described in Patent Document 1 collects the total amount of communication handled by the switch through which communication between each virtual machine passes for each switch, so that the total does not exceed the communication capacity of the switch. Determine the physical server where the machine will be placed.

Japanese Patent Laying-Open No. 2015-225560

  On the other hand, with respect to network resources, unlike CPU and memory resources, there are many cases where resource depletion cannot be solved by monitoring only the resources of a VMM (Virtual Machine Monitor) that operates a VM. The CPU and memory completely depend on the VMM, but the network resource is like a passage through various devices. For this reason, even when a local determination is limited to a specific device, the point causing the true cause is not known, and an erroneous determination may be made.

  Therefore, it is necessary to determine the resource after understanding the entire network and dynamically change the settings. However, the performance of switches connecting networks is not constant. Therefore, even if the usage rate (that is, the load factor) of the hardware resource of the switch can be grasped, it is difficult to grasp how the resources are allocated when the network configuration is changed.

  For example, in a certain system, there are a core switch for managing a network and an edge switch connected to the core switch, and a server for providing a service and a business computer (Personal computer) are connected to the edge switch to operate. Suppose you are in a situation. Here, it is assumed that the server and the business PC exist as a FAT-PC or a VM.

  In general, the performance of the switches is not the same, and there is a performance difference between the switches. The number of servers and business PCs connected to the switch is not constant. Furthermore, the amount of traffic flowing through the switch is not the same, and there is a bias.

  For this reason, although the traffic amount of the entire network does not exceed the capacity of all the switches, from a local viewpoint, the load may be biased to a specific switch, which may affect services and operations.

  When such a situation occurs, it is recognized that an imbalance in network load has occurred due to a notification from a service user or a user in the network or a notification from network monitoring software or the like. In order to solve this situation, the network administrator needs to take measures such as manually changing the network configuration or connecting a terminal that uses a large amount of network resources to another switch.

  There are two main problems in this case. First, since it is necessary to design and evaluate the review of the network configuration, it is difficult to deal with an emergency situation that requires immediateness. Second, the downtime required for changing the configuration of the switch occurs, which affects the user. The first problem is particularly difficult to deal with. For example, in the case of a complicated network configuration, there is a high possibility that the problem cannot be solved by changing the setting from a local viewpoint.

  Specifically, the larger the network, the greater the range to be confirmed, necessitating a complicated design. For example, in order to grasp the influence of a certain switch, a phenomenon may occur in which an upper switch is traced from an edge switch and a lower switch connected to the switch and a lower switch must be traced.

  If all switches are of the same model, simply check the usage rate (load factor) of the hardware resource, change the network configuration, that is, add or subtract the load factor, and the resource problem Can be solved. However, in an actual network, the performance is often different for each switch. In that case, since the resource allocation needs to be calculated in consideration of the performance difference between the switches, enormous calculation is required to solve the problem.

  In reality, it can be said that it is full to confirm a specific server and an edge switch to which the server is connected or one switch higher than the edge switch. For this reason, for example, there is no actual operation of changing the network configuration in a spot manner by calculating the load factor of the network while expanding the range from the edge switch to the higher level switch and further higher level switch. .

  Further, the virtual machine placement device described in Patent Document 1 must grasp the communication capacity of each switch. Further, as described above, it is general that switches having different performances are abolished on the network. For this reason, as described in Patent Document 1, it is difficult to appropriately allocate resources of switches having different performances simply by allocating resources so as not to exceed the communication capacity. Therefore, in order to reduce workload and appropriately allocate resources, it is preferable that network resources can be leveled autonomously.

  Accordingly, an object of the present invention is to provide a network resource management device, a network resource management method, and a network resource management program that can level network resources including switches having different performances.

  The network resource management apparatus according to the present invention is a network resource management apparatus that manages the resources of each switch arranged in the network, and estimates the performance of each switch from the usage rate of each switch and the flow that flows through each switch. A network configuration to which a performance estimation unit, a first load calculation unit that calculates a first load of each switch, and a target switch that is a switch in which the calculated first load is higher than a predetermined threshold is connected. A configuration change candidate determination unit that determines candidates for change, and a second load calculation that calculates the second load of each switch with the estimated performance based on the configuration of the changed network determined as a candidate And a section.

  The network resource management method according to the present invention is a network resource management method for managing the resources of each switch arranged in the network, and estimates the performance of each switch from the usage rate of each switch and the flow flowing through each switch. The first load of each switch is calculated, and a candidate for changing the configuration of the network to which the target switch that is a switch whose calculated load is higher than a predetermined threshold is connected is determined as a candidate. The second load of each switch at the estimated performance is calculated based on the changed network configuration.

  A network resource management program according to the present invention is a network resource management program applied to a computer that manages the resources of each switch arranged in a network, and is based on the usage rate of each switch and the flow that flows through each switch. A performance estimation process for estimating the performance of each switch, a first load calculation process for calculating the first load of each switch, and a target switch that is a switch in which the calculated first load is higher than a predetermined threshold is connected Configuration change candidate determination process for determining candidates for changing the network configuration to be changed, and the second load of each switch at the estimated performance based on the changed network configuration determined as a candidate The second load calculation process is performed to calculate.

  According to the present invention, network resources including switches having different performances can be leveled.

It is a block diagram which shows one Embodiment of the network resource management apparatus by this invention. It is explanatory drawing which shows the example of the assumed network configuration. It is a flowchart which shows the operation example of a resource management apparatus. It is explanatory drawing which shows the specific example of the management method using a network resource management apparatus. It is explanatory drawing which shows the specific example of the management method using a network resource management apparatus.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. The network resource management apparatus of this embodiment manages the resources of each switch arranged in the network. It is assumed that the switch managed in the present embodiment implements SNMP (Simple Network Management Protocol) and implements a protocol capable of transmitting statistical information as a flow, such as sFlow or NetFlow.

  FIG. 1 is a block diagram showing an embodiment of a network resource management apparatus according to the present invention. The network resource management apparatus 100 according to the present embodiment includes a performance estimation unit 10, a first load calculation unit 20, a configuration change candidate determination unit 30, and a second load calculation unit 40.

  FIG. 2 is an explanatory diagram showing an example of a network configuration assumed in this embodiment. In the network illustrated in FIG. 2, there are several core switches that are backbones, edge switches connected to the core switches, and FAT-PCs or VMs connected to the edge switches. FAT-PC or VM has various uses, and includes those that communicate only within the network, those that are connected to an external server, and those that are open to the public as a server. Hereinafter, these are collectively referred to as a node.

  There is an implicit priority between the nodes. For example, a server or the like that provides a correction patch for its product to the outside wants to allocate resources more preferentially than other nodes, so a high priority is set. Within the network, there is a manager (MGR) that monitors the entire network. This MGR corresponds to the network resource management apparatus 100 of this embodiment.

  The MGR monitors the load of the entire network, and when there is an area where the network load occurs, the MGR scores each switch based on the processing shown in the present embodiment, and performs evaluation and configuration change. A firewall (FW) exists between the outside and the switch closest to the outside.

  A line and a broken line between the switches illustrated in FIG. 2 indicate routes on the network. Physical wiring exists between the upper and lower switches, and the network path can be flexibly changed according to the network load. The MGR periodically acquires information on these physical wirings and manages the information as internal information. This information is used as information for determining candidates for changing the network configuration.

  In this way, in order to equalize network resources including switches having different performances, it is first necessary to consider the performance difference between the switches. That is, if all the switches are of the same model, it is possible to level the resources by simply acquiring the CPU usage rate and the backplane usage rate.

  However, as described above, in most cases, switches are selected in accordance with system requirements. In addition, since switches may be added due to configuration changes or additions, it is practically impossible that the performance of all switches is the same. Therefore, it is necessary to grasp the performance difference of the switch by some method. However, for example, it is difficult to check the catalog specifications for each switch model and calculate the performance difference from the actual usage rate.

  Therefore, the performance estimation unit 10 of the present embodiment estimates the performance of each switch from the usage rate of each switch and the flow that flows through each switch. Thereby, the absolute performance of each switch is dynamically calculated. The switch performance includes switching capacity (data amount that can be processed per second) and switching capacity (number of frames that can be processed per second).

  The CPU usage rate of each switch and the traffic usage rate of the entire switch can be acquired from a management information base (MIB). This is because a switch in which MIB is mounted can acquire information from the outside, and therefore an arrangement of an agent or the like for acquiring information in each switch is unnecessary. In addition, the flow (traffic and the number of frames) flowing through each switch can be acquired by using a technology for monitoring traffic such as sFlow or netflow.

  Therefore, the performance estimation unit 10 calculates an approximate limit value of the performance of each switch by dividing the amount of flow flowing through the switch by the usage rate of the switch. For example, when the traffic flowing per unit time is 10 Gbps and the traffic usage rate of the switch is 20%, the performance estimation unit 10 can estimate the limit of the switching capacity as the performance of the switch is approximately 50 Gbps. Similarly, the performance estimation unit 10 can also estimate the switching capability of the switch from the sum of the CPU usage rate per unit time and the number of frames that have passed through the switch.

For example, when the total number of frames F ₁ processed by the switch at a certain point in time and the total number of frames F ₂ processed by the switch after a lapse of t from a certain point in time, the average value of the CPU usage rate during the period t is It can be calculated by In Equation 1, T is the number of times the CPU usage rate is measured in the period t, and _Cp is the CPU usage rate at the time of measurement.

  In this case, the switching capacity of the switch can be calculated by the following equation 2.

  The switching ability can also be calculated in the same manner as the above formulas 1 and 2.

  The first load calculation unit 20 calculates the load (first load) of each switch. The first load calculation unit 20 may use the acquired CPU usage rate as it is as the load rate of each switch. However, in consideration of actual operation, there are cases where resource allocation cannot be determined simply by the load of the switch. This is the case, for example, when network resources are to be allocated preferentially to a patch distribution server that is open to the public rather than a business PC that uses an in-house network.

  Therefore, a parameter indicating the degree (priority) to which resources should be preferentially allocated may be set for each switch and terminal. For example, α is a parameter set for a switch as a priority, β is a parameter set for a terminal, and γ is a load factor acquired from each switch. At this time, the first load calculation unit 20 may calculate the load factor with priority of each switch by the following Expression 3. Note that n is the number of terminals. As shown in Equation 3, the first load calculation unit 20 may calculate the load higher as the priority is higher. In this way, the resource status analysis considering the entire network is performed.

  The configuration change candidate determination unit 30 determines a candidate for changing the configuration of a network to which a switch having a calculated load higher than a predetermined threshold is connected. The configuration change candidate determination unit 30 may automatically determine a configuration candidate list that can secure a route to the currently connected node. In addition, the configuration change candidate determination unit 30 may determine network configuration candidates connected to the target switch in accordance with a user instruction.

  The second load calculation unit 40 calculates the load (second load) of each switch with the estimated performance based on the changed network configuration determined as a candidate. That is, the second load calculation unit 40 evaluates the resource status when the configuration is changed. Based on this evaluation, configuration changes are implemented.

Specifically, the second load calculation unit 40 calculates the load factor of each switch from the flow that flows when the configuration is changed based on the performance of each switch estimated by the performance estimation unit 10. For example, it is assumed that traffic (the number of frames) F ₃ is generated in the target switch due to a change in configuration. In this case, the second load calculation unit 40 calculates the load factor γ after resource distribution using, for example, Expression 4 shown below.

  The second load calculation unit 40 may determine whether or not the calculated load factor γ is lower than a predetermined load factor. If the calculated load factor γ is greater than or equal to a predetermined load factor, another configuration change can be considered.

  The performance estimation unit 10, the first load calculation unit 20, the configuration change candidate determination unit 30, and the second load calculation unit 40 are realized by a CPU of a computer that operates according to a program (network resource management program). For example, the program is stored in a storage unit (not shown) of the network resource management device, and the CPU reads the program, and the performance estimation unit 10, the first load calculation unit 20, and the configuration change candidate determination unit 30 according to the program. Alternatively, the second load calculation unit 40 may operate. Moreover, each of the performance estimation unit 10, the first load calculation unit 20, the configuration change candidate determination unit 30, and the second load calculation unit 40 may be realized by dedicated hardware.

  Next, the operation of the resource management apparatus of this embodiment will be described. FIG. 3 is a flowchart illustrating an operation example of the resource management apparatus according to the present embodiment. First, MGR (that is, the network resource management apparatus 100 of the present embodiment) is arranged on the network. Furthermore, priorities are set in advance for nodes (specifically, terminals and switches) to which resources are to be preferentially allocated (step S11). The priority is managed by an MGR arranged in the network. Note that in the case of a network in which all priorities are flat (that is, there is no bias in priorities), the step of setting priorities may be omitted.

  The MGR (for example, the performance estimation unit 10) regularly collects node information (step S12). Information collected by the MGR is mainly CPU usage rate, traffic information, and the like. Note that MGR may further collect values that serve as other indices. Each switch periodically transmits statistical information such as the number of packets that have passed through the switch to the MGR as a flow.

  The MGR (for example, the performance estimation unit 10) analyzes the acquired information and determines whether there is a high-load node (step S13). When a high-load node exists (Yes in step S13), the MGR (for example, the first load calculation unit 20) calculates a load factor considering the performance difference for each model (step S14). On the other hand, when there is no highly loaded node (No in step S13), the MGR repeats the process of collecting node information.

  The MGR (for example, the configuration change candidate determination unit 30) determines candidates for eliminating resource depletion within a range in which the configuration can be changed. And MGR (for example, the 2nd load calculation part 40) calculates the load factor at the time of changing the structure of a network according to the determined candidate (step S15). Specifically, the MGR calculates a load factor by selecting all lower switches and upper switches around a node where a load is generated.

  The MGR (for example, the second load calculation unit 40) determines whether or not the resource shortage can be resolved by switching the paths in the candidates (step S16). If the high load state cannot be resolved even if all candidates are applied, that is, if the resource shortage cannot be resolved by switching the paths in the candidates (No in step S16), MGR determines whether there is a higher-level switch. (Step S17).

  If a higher switch exists (Yes in step S17), the MGR (for example, the second load calculation unit 40) selects the upper switch and includes it in the candidate area (step S18). And the process after step S15 which determines the candidate for eliminating resource exhaustion is repeated. In other words, the MGR calculates a load factor by selecting a higher-order switch and a switch existing under the switch. That is, this process is repeated by expanding the area until the high load state is resolved.

  On the other hand, if the resource shortage can be resolved by switching the path in the candidate in Step S16 (Yes in Step S16), the MGR determines that the optimum candidate has been found, and changes the network path (Step S16). S19) The network high load state is eliminated.

  Next, a specific example of this embodiment will be described. 4 and 5 are explanatory diagrams showing a specific example of a management method using the network resource management apparatus of this embodiment.

  As illustrated in FIG. 4, there is a switch 301 (hereinafter referred to as SW1) closest to the outside, and a switch 310 (hereinafter referred to as switch SW10) and a switch 311 (hereinafter referred to as switch SW11) connected to SW1. The MGR 300 is connected to SW1. Further, a switch 320 (hereinafter referred to as SW100) and a switch 321 (hereinafter referred to as SW101) are connected to SW10. On the other hand, a switch 322 (hereinafter referred to as switch SW102) is physically connected to SW10, but the path is blocked by the setting of SW10.

  SW102 is connected to SW11. On the other hand, SW100 and SW101 are physically connected to SW11, but the path is blocked by the setting of SW11.

  A switch 330 (hereinafter, SW1001), a switch 331 (hereinafter, SW1002), and a switch 332 (hereinafter, SW1003) are connected to the SW101. Further, SW1001, SW1002, and SW1003 are physically connected to SW100 and SW102, respectively, but the path is blocked by the setting of each switch.

  Note that lower SWs connected to SW100 and SW102 are omitted. SW1001, SW1002, and SW1003 are edge switches, and a plurality of FAT-PCs 340 or VM341 are connected under the switch.

  In this specific example, it is assumed that the network load of the SW 101 is 93% and is in a high load state. The network load of the upper SW 10 of the SW 101 and the lower SW 1001, SW 1002, and 1003 are all 60% or less. Therefore, in this specific example, it is considered that the load state can be eliminated by the configuration change under SW101.

  Next, the MGR 300 confirms the upper switch physically connected to the lower switch of the SW 101. In addition to SW101, two switches, SW100 and SW102, are physically connected to SW1001, SW1002, and SW1003, and each load factor is 60% or less.

  The load factor of SW100 is 60%, and the load factor of SW102 is 50%. Therefore, at first glance, it may be effective to increase the number of routes to SW102. However, in practice, there is a performance difference between SW100 and SW102.

  In this specific example, it is assumed that each switch has “estimated switching capacity” described in each switch in FIG. 4 as performance. Considering the performance difference between the switches, it can be said that the resource of the SW 100 is larger than that of the SW 102. Therefore, as illustrated in FIG. 5, the MGR 300 changes the route of the SW 1003 that is a switch under the SW 101 to the SW 100. As a result, the load on SW101 is reduced, and the loads on SW100, SW101, and SW102 are leveled.

  If it is determined only by the network load (usage rate) and the route of SW1003 is switched to SW102, the usage rate of SW102 will be 100%, and resource allocation will deteriorate due to the route change. . However, in the present embodiment, the performance estimation unit 10 estimates the performance of each switch, and the second load calculation unit 40 appropriately calculates the load factor based on the estimated performance. Therefore, network resources including switches with different performance can be leveled.

  As described above, in this embodiment, the performance estimation unit 10 estimates the performance of each switch from the usage rate of each switch and the flow that flows through each switch, and the first load calculation unit 20 uses the first of each switch. Calculate the load. The configuration change candidate determination unit 30 determines a candidate for changing the configuration of a network to which a switch having a calculated load higher than a predetermined threshold is connected, and the second load calculation unit 40 is determined as a candidate. Based on the network configuration after the change, the second load of each switch at the estimated performance is calculated. Therefore, as described above, network resources including switches with different performances can be leveled.

  Further, by using the network resource management apparatus 100 of the present embodiment, even in a network in which various switches having different performance exist, the network path is dynamically changed without being aware of the performance of each switch, Network load can be leveled. In addition, the network resource management apparatus 100 according to the present embodiment can use a technology that is normally installed in a switch, such as SNMP, sFlow, and NetFlow. Therefore, it is possible to operate in a wide range of environments without limiting the types of switches.

  Next, the outline of the present invention will be described with reference to FIG. A network resource management apparatus according to the present invention is a network resource management apparatus 100 that manages resources of each switch arranged in a network, and estimates the performance of each switch from the usage rate of each switch and the flow that flows through each switch. A network to which a performance estimation unit 10 that performs calculation, a first load calculation unit 20 that calculates a first load of each switch, and a target switch that is a switch in which the calculated first load is higher than a predetermined threshold is connected Based on the configuration change candidate determination unit 30 that determines candidates for changing the configuration and the network configuration after the change determined as a candidate, the second load of each switch with the estimated performance is calculated. And a second load calculation unit 40.

  With such a configuration, network resources including switches having different performances can be leveled.

  In addition, priority is determined for each switch and a node (for example, FAT-PC, VM) connected to the switch, and the first load calculation unit 20 calculates the first load higher as the priority is higher. May be.

  Further, the configuration change candidate determination unit 30 selects all lower switches of the target switch and upper switches of the target switch as candidates for changing the network configuration, and the second load calculation unit 40 selects the selected network. It may be determined whether the second load calculated based on the configuration is less than a predetermined load.

  Further, the configuration change candidate determination unit 30 further determines whether the configuration of the network is further changed as a candidate for changing the network configuration when there is no configuration in which the second load of each switch calculated by the second load calculation unit 40 is lower than a predetermined load. The switch and the switches under the switch may be selected.

  The first load calculation unit 20 may use the load acquired from the management information base (MIB) as the first load of each switch.

DESCRIPTION OF SYMBOLS 10 Performance estimation part 20 1st load calculation part 30 Configuration change candidate determination part 40 2nd load calculation part 100 Network resource management apparatus

Claims (9)

A network resource management device that manages resources of each switch arranged in a network,
A performance estimation unit that estimates the performance of each switch from the usage rate of each switch and the flow flowing through each switch;
A first load calculation unit for calculating a first load of each switch;
A configuration change candidate determination unit that determines a candidate for changing the configuration of a network to which a target switch that is a switch in which the calculated first load is higher than a predetermined threshold;
A network resource management comprising: a second load calculating unit that calculates a second load of each switch with the estimated performance based on the network configuration after the change determined as the candidate apparatus. Priorities are established for each switch and the node connected to that switch,
The network resource management device according to claim 1, wherein the first load calculation unit calculates the first load higher as the priority is higher. The configuration change candidate determination unit selects all lower switches of the target switch and higher switches of the target switch as candidates for changing the network configuration,
The network resource management device according to claim 1, wherein the second load calculation unit determines whether or not the second load calculated based on the configuration of the selected network is lower than a predetermined load. When there is no configuration in which the second load of each switch calculated by the second load calculation unit is lower than a predetermined load, the configuration change candidate determination unit further selects a higher-level switch and the relevant switch as candidates for changing the network configuration. The network resource management device according to claim 3, wherein a switch under the switch is selected. The network resource management device according to any one of claims 1 to 4, wherein the first load calculation unit uses a load acquired from the management information base as a first load of each switch. A network resource management method for managing resources of each switch arranged in a network,
Estimating the performance of each switch from the usage rate of each switch and the flow flowing through each switch,
Calculating the first load of each switch;
Determining a candidate for changing the configuration of the network to which the target switch, which is a switch whose calculated load is higher than a predetermined threshold, is connected;
A network resource management method, comprising: calculating a second load of each switch with the estimated performance based on a network configuration after change determined as the candidate. Priorities are established for each switch and the node connected to that switch,
The network resource management method according to claim 6, wherein the higher the priority is, the higher the first load is calculated. A network resource management program applied to a computer for managing resources of each switch arranged in a network,
In the computer,
A performance estimation process for estimating the performance of each switch from the usage rate of each switch and the flow flowing through each switch;
A first load calculation process for calculating a first load of each switch;
A configuration change candidate determination process for determining a candidate for changing the configuration of a network to which a target switch that is a switch in which the calculated first load is higher than a predetermined threshold; and
A network resource management program for executing a second load calculation process for calculating a second load of each switch with the estimated performance based on the changed network configuration determined as the candidate. Priorities are established for each switch and the node connected to that switch,
On the computer,
The network resource management program according to claim 8, wherein in the first load calculation process, the higher the priority, the higher the first load is calculated.

Images