JP2009075718A - Method of managing virtual i/o path, information processing system, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To configure an I/O path virtualized between servers, and to guarantee end-to-end QoS. <P>SOLUTION: An information processing system has a global QoS management section, a local QoS management section corresponding to a processing node, and a logic I/O management section. The global QoS management section has: a QoS assignment management section that configures a virtual end-to-end I/O path in the information processing system and assigns, changes, and deletes a band to the virtual I/O path; a global QoS management table for holding state information on the entire switch fabric and all virtual I/O paths; and a band monitoring section for monitoring the state information of the entire switch fabric. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention provides a server group connected via an I / O switch such as a switch fabric, and an end-to-end data transfer amount between I / O adapters for each virtualized I / O transfer path and The present invention relates to a method for managing a communication band and an information processing system using the method.

  The trend of multi-core processors (CPUs) and low power consumption has become serious, and full-scale introduction of CPU virtualization in the server field is progressing. CPU virtualization means that a single system hardware resource (for example, CPU and memory) that could only be used by a single OS is logically divided into a plurality of hardware resources so that a plurality of OSs can be used. Technology. By virtualization of the CPU, the user can efficiently consolidate a plurality of servers into one or a small number of servers at low cost and with reduced power consumption.

  Since CPU virtualization in a conventional server system is realized on a software basis, there is a problem that overhead is large. However, in recent years, CPU vendors have been able to install a hardware mechanism for supporting the virtualization of the CPU in the CPU to reduce the overhead associated with the virtualization. From such a situation, it is considered that commoditization of the CPU virtualization function will further advance in the future.

  In conventional CPU virtualization, hardware resources such as CPU and memory were divided and used. However, due to problems such as performance and security, the I / O is not provided as a virtualization target but provided independently. It was common. However, with the growing need for low power consumption and continuous network speedup, the flow of I / O virtualization is expected to accelerate in the future.

  Regarding I / O virtualization, CPU vendors have recently introduced a chip set equipped with a hardware mechanism that supports I / O virtualization. In addition, standardization for virtualizing I / O in an I / O interface card and an I / O switch is also being promoted in PCI-SIG that develops PCI-related standards.

  As a specific example of I / O virtualization in the I / O interface card unit, there is a method in which one physical I / O interface card is shared by a plurality of OSs. For example, in Patent Document 1, a virtual host logically divided by CPU virtualization and a virtual I / O interface card having different QoS (Quality of Service) on a physical I / O interface card are supported. Technology is disclosed. In Patent Document 1, the hypervisor sets QoS weighting for each set of processing queues assigned to a system image, so that QoS can be set corresponding to a virtual I / O interface card.

 On the other hand, there are PCI-Express switches and InfiniBand switches as I / O virtualization technologies using I / O switches. In these I / O switches, a group of servers connected to the switch can transfer a plurality of I / O protocols such as IP and Fiber Channel using the same route, so that I / O can be integrated at low cost. However, when such an I / O switch is introduced, a management function for recording the connection relation of each port of the I / O switch or collecting port status information is required.

  As a method for managing the I / O switch as described above, there are a method for enhancing the functionality of the I / O switch itself, and a method for managing the I / O switch by a dedicated server without enhancing the functionality.

  As a method for enhancing the functionality of the I / O switch itself, for example, Patent Document 2 discloses a method for collecting fabric state information in a fiber channel switch system. Since the fiber channel switch system is composed of a plurality of switch fabrics, it is necessary to provide a fabric information collection method capable of monitoring the load status of each fabric within the fiber channel switch system. In order to solve this problem, in Patent Document 2, when the switch fabric is equipped with a CPU and the CPU receives a request for collecting state information, the CPU of the switch fabric collects load information.

  On the other hand, as a method for managing the I / O switch state outside the I / O switch, for example, Non-Patent Document 1 discloses the specification of the fabric management manager in ASI-SIG. The fabric management manager performs configuration management of an ASI (Advanced Switch Interconnect) switch fabric. Since the ASI switch fabric is not equipped with a processor, the fabric information is collected by accessing a register that is memory-mapped in the configuration space of the management dedicated server.

The fabric information that can be collected here includes active information corresponding to the physical port and the number of transmitted / received packets. In ASI, these pieces of information are acquired inbound using a register access packet called PI (Protocol Interface) -4 supported by ASI.
US Patent Application Publication No. 2006/0193327 A1 JP 2001-274794 A ASI Core Specification Revision 1.1 Final, ASI-SIG, 2004

  By the method disclosed in Patent Document 1, the QoS of I / O between a virtual I / O interface card sharing a single physical I / O interface card and a virtual host corresponding to the virtual I / O interface card is determined. Can be set individually.

  However, when the method disclosed in Patent Document 1 is used, only QoS between an I / O interface card and a server in an information processing system including a group of servers connected to an I / O switch can be guaranteed. In other words, in the conventional method, an I / O path is virtually configured between servers connected by an I / O interface card via an I / O switch or between a server and an I / O device. There is a problem that the QoS cannot be guaranteed (first problem).

  In other words, in an I / O interface card that supports multi-protocols, virtual I / O ports of different protocols can be configured on the physical port of the I / O switch, and the virtual port of TCP / IP can be configured on one physical port. It is possible to configure a virtualized I / O port and an iSCSI virtualized I / O port. Then, a TCP / IP virtual I / O path and a SCSI virtual I / O path are respectively configured between the I / O interface card, the I / O switch, and other computers or I / O devices. be able to. However, in Patent Document 1, a virtual I / O interface card is assigned to each physical I / O interface card for each of a plurality of virtual hosts (guest OSs), and QoS is assigned to each virtual I / O interface card. Although it can be controlled, there is a problem that the QoS cannot be secured for the virtual I / O path of TCP / IP and the virtual I / O path of iSCSI as described above. .

  In addition, in order to guarantee the QoS of a plurality of virtually configured I / O paths in the entire information processing system, port usage information of the virtual I / O interface card is provided for each configured I / O path. Need to manage. When a plurality of I / O paths are configured, it is necessary to manage port status information for each of the plurality of I / O paths in a single physical port. The technology disclosed in Patent Document 2 is based on the premise that the switch fabric is equipped with a CPU, and the CPU on the I / O switch (switch fabric) collects load information for each port for management. Therefore, there is a problem that load information for each port cannot be collected in an I / O switch without a CPU.

  On the other hand, the technique disclosed in Non-Patent Document 1 discloses an interface for a management-dedicated server to acquire state information of a port of an I / O switch. By using this interface, even when the I / O switch does not have a CPU, it is possible to collect the state information of the ports of the I / O switch constituting the server. However, according to the method of Non-Patent Document 1, as CPU virtualization and I / O virtualization become full-scale in the future, between servers or servers and I / O devices in a server group connected to an I / O switch. When virtualized I / O paths corresponding to multiple protocol types are configured between them, there is a problem that it is not possible to collect port status information corresponding to each virtualized I / O path. (Second problem).

  Further, as the scale of the I / O switch connecting the server group and the number of virtualized I / Os per physical I / O port increase, the status information of the I / O ports to be managed increases. As a result, the number of times the port state information is read increases. If the number of times I / O port status information is read increases, it affects other processes, resulting in an increase in processing time for obtaining status information and variations in processing time. (Third problem)

  As described above, a plurality of virtual I / O devices configured between servers or between a server and an I / O device in a server group connected to an I / O switch for full-scale introduction of CPU virtualization. A method for managing the data amount and transfer rate of the / O path and a system using the same are required.

  Even if a plurality of virtualized I / O paths are configured in an apparatus that performs CPU virtualization on a server connected to an I / O switch, in order to solve the first problem, a virtual Of virtualized I / O path data amount and transfer rate (bandwidth) that can guarantee QoS end-to-end for each data transfer path (path) for each I / O is required.

  In order to solve the second problem, a virtualized I / O capable of collecting status information for each I / O port virtually configured as a physical port even in an I / O switch having no CPU. A method for managing the data amount and transfer rate (bandwidth) of the O path is required.

  In order to solve the third problem, a virtual I / O port to be managed even when there are a large number of virtual I / O ports per physical port or a large-scale I / O switch There is a need for a method for managing the data amount and transfer rate (bandwidth) of the virtualized I / O path that can execute the state information read processing for each time in a certain time.

  A first problem of the present invention is that an information processing system including a group of servers connected to an I / O switch has an end of each I / O path when a plurality of virtualized I / O paths are configured. It is to provide a method for managing the data amount and transfer rate (bandwidth) of a virtualized I / O path that can guarantee QoS in a two-end manner.

  A second problem of the present invention is to collect information for each I / O port virtually configured as a physical port even in an I / O switch having no CPU in addition to the first problem. It is an object of the present invention to provide a method for managing virtualized I / O data volume and transfer rate (bandwidth).

  The third problem of the present invention should be managed even in the case where there are a large number of virtual I / O ports per physical port or a large-scale I / O switch in addition to the second problem. To provide a method for managing a data amount and a transfer rate (bandwidth) of a virtualized I / O path that can execute a state information read process for each virtualized I / O path in a predetermined time. .

  The present invention has a plurality of nodes each having an I / O adapter that can set a virtual I / O port as a physical I / O port, and connects the I / O adapter of the node to a port of an I / O switch. A method of managing a virtual I / O path in an information processing system that transfers data between the plurality of nodes via the virtual I / O port, wherein any one of the plurality of nodes is the I / O switch port information is collected, each node acquires a bandwidth for each virtual I / O port of an I / O adapter provided in each node, and any one of the plurality of nodes is The port information of the I / O switch, the band is acquired, the information of the port of the I / O switch, the node that acquired the band is the information of the source node and the information of the destination node And band included A node that receives a request to generate a virtual I / O path and receives a request to generate a virtual I / O path is a virtual node from the source node to the destination node based on the port information of the I / O switch. The node that selects the path of the I / O path, receives the virtual I / O path generation request, extracts the virtual I / O port on the path of the selected virtual I / O path, and extracts the extracted virtual I It is determined whether or not the virtual I / O path can be set from the acquired bandwidth and the bandwidth included in the generation request in the / O port, and the node that has received the virtual I / O path generation request When it is determined that an I / O path can be set, the virtual I / O path is set to a port on the virtual I / O path.

  In addition, the step in which any one of the plurality of nodes collects information on the port of the I / O switch includes a step of selecting one node based on an operating state of the plurality of nodes.

  According to the present invention, a plurality of virtualized I / O paths are configured for an information processing system that connects a plurality of processing nodes via an I / O switch, and a data transfer rate (bandwidth) is set for each I / O path. Width) can be controlled, so that end-to-end QoS can be guaranteed for each virtualized I / O path. Even an I / O switch that does not have a CPU can manage status information for each I / O port virtually configured on a physical port of an I / O adapter and collect the information. In addition, even when there are a large number of I / O switches or a large number of virtual I / O ports per physical port, I / O switch status information read processing is connected to the I / O switch. By sharing according to the operating states of a plurality of processing nodes, it is possible to minimize the influence on other communications and execute I / O switch information collection in a certain time.

  Hereinafter, an embodiment of the present invention will be described with reference to the accompanying drawings.

<First Embodiment>
FIG. 1 shows a first embodiment and is a block diagram of an information processing system to which the present invention is applied.

  The information processing system includes a plurality of processing nodes (server computers) 101 and 102 that provide services, a management node (management computer) 103 that manages these processing nodes 101 and 102, and a virtual I In order to provide an / O path (logical I / O path), the switch fabric group 105 includes a plurality of switch fabrics 170 and 171. The management node 103 is connected to a console 104 that can input and output to the administrator. Each node is connected via a backplane (not shown) and stored in a housing (not shown).

  Each of the processing nodes 101 and 102 includes a plurality of CPUs (so-called multi-core processors) 10 including a plurality of CPU (processor) cores # 0 and # 1. Since the processing nodes 101 and 102 have the same configuration, only the configuration of the processing node 101 will be described below.

  The CPU # 0 has a plurality of CPU cores # 0-0 and # 0-1, and the CPU # 1 has a plurality of CPU cores # 1-0 and # 1-1. The CPUs # 0 and # 1 are connected to the chipset 12 via the front side bus 11.

  The chip set 12 is connected to the main memory 13 via the memory bus, and accesses the main memory 13 in response to memory access requests from the CPUs # 0 and # 1. Further, the chip set 12 is connected to an I / O adapter 113 as an I / O interface card via an I / O bus 14 and is connected to the switch fabric group 105 in response to an I / O access request from the CPUs # 0 and # 1. to access. The I / O bus 14 can be configured by, for example, PCI-EXPRESS.

  The I / O adapter 113 provides a plurality of protocols, and a plurality of virtual I / O ports (hereinafter referred to as virtual I / O ports or logical I / O ports) as physical I / O ports (hereinafter referred to as “physical I / O ports”). And communicate with the switch fabric group 105 for each virtual I / O port. The I / O adapter 113 includes a controller including a register 180 described later.

  The hardware configuration of the management node 103 is the same as that of the processing node 101, and includes a CPU 20, a front side bus 21 chip set 22, a main memory 23, an I / O bus 24, and an I / O adapter 120, and an I / O adapter. 120 is connected to the switch fabric group 105.

  As will be described later, the CPU 10 loads the OS, applications, and management units to the main memory 13 and executes them.

  FIG. 2 is a block diagram of the information processing system 100 according to the first embodiment of this invention.

  The information processing system 100 includes processing nodes 101 and 102, a management node 103, a management console 104, and a switch fabric group 105. The processing nodes 101 and 102 and the management node 103 are connected to the switch fabric group 105. A common physical interface is used for connection between the switch fabric group 105 and each node, but any protocol may be used as an upper protocol. For example, it may be a unique protocol having a unique physical interface, or a unique protocol using a standard physical interface such as PCI. Alternatively, it may be a standard protocol having a standard physical interface such as PCI Express.

  For example, the switch fabric group 105 and the I / O adapters 113, 118, and 120 adopt PCI Express as a physical layer and can communicate with a plurality of protocols such as TCP / IP, FC (Fibre Channel), or SCSI as an upper layer. Composed.

  In addition, the switch fabrics 170 and 171 include registers 180 and 181 that store setting states of a plurality of physical I / O ports and virtual I / O ports that constitute each switch fabric.

  The processing nodes 101 and 102 are either a server processing node on which an application is mounted or an I / O processing node on which an I / O device (for example, a disk drive) is mounted and performs I / O processing. In the first embodiment of the present invention, the processing nodes 101 and 102 function as server processing nodes that execute applications, and only the processing nodes 101 and 102 and the management node 103 are connected to the switch fabric group 105. However, an I / O processing node on which an I / O device or the like is mounted may be connected to the switch fabric group 105. In this case, the I / O node is also connected to other nodes through a common physical interface, and the I / O node includes an I / O controller that controls the I / O device and an I / O device.

  The processing nodes 101 and 102 exchange data via the switch fabric group 105. In the first embodiment of the present invention, two processing nodes are illustrated, but any number of processing nodes may be provided.

  The management node 103 manages the overall configuration of the information processing system 100 and exchanges status information with the processing nodes 101 and 102 via the switch fabric group 105. In the first embodiment of the present invention, only one management node is illustrated, but a plurality of management nodes may be provided in order to improve reliability. In the case where a plurality of management nodes are provided, control is performed so as to ensure data consistency between the management nodes.

  The management console 104 is a computer having a CPU and a memory. A management request from an administrator, for example, a request to assign QoS (Quality of Service) to a virtual I / O path (logical I / O path) configured in the switch fabric group 105 is transmitted to the management node 103 To do. The virtual I / O path is notified to each I / O driver by the QoS allocation management units 131 and 134 of the processing nodes 101 and 102 based on a command from the management console 104 to the management node 103 as will be described later. Set by each I / O driver. Alternatively, the applications 110 and 115 of the processing nodes 101 and 102 may notify the I / O driver and set a virtual I / O port.

  Here, QoS in the present embodiment indicates that a bandwidth for each of a plurality of virtual I / O ports (virtual I / O ports) set on a physical I / O port is guaranteed. For example, as shown in FIG. 5, when the virtual I / O ports “1” and “4” are set in the physical I / O port “0” of the I / O adapter 113, the virtual I / O port “1” is set. Communicates with other processing nodes using TCP / IP, and the virtual I / O port “4” communicates with other processing nodes (for example, storage devices) using the FC protocol. At this time, guaranteeing the TCP / IP bandwidth and the FC bandwidth for each virtual I / O port (virtual I / O port) is the QoS in this embodiment.

  The switch fabric group 105 connects the processing nodes 101 and 102 and the management node 103. The internal configuration of the switch fabric group 105 may be a bus, a crossbar network, or a multistage network. Any configuration is possible as long as information can be exchanged between processing nodes and between a processing node and a management node. In the first embodiment of the present invention, an example in which the switch fabric group 105 is configured by a plurality of switch fabrics 170 and 171 configured by a non-blocking crossbar network is shown.

  The processing node 101 includes an application 110, a local QoS management unit 111, a logical I / O management unit 112, and an I / O adapter 113. The processing node 102 has the same configuration as that of the processing node 101, and only the reference numerals are different in FIG.

  The application 110 is a user program executed by the CPU 10 of the processing node 101. The local QoS management unit 111 manages the setting information of the virtual I / O ports of the processing node 101 on which the local QoS management unit 111 is mounted and the I / O adapter 113 on the processing node 101, and periodically stores these information. To the global QoS management unit 122 (described later) on the management node 103. The logical I / O management unit 112 manages the correspondence between the status information of the physical I / O adapter 113 and the virtualized I / O adapter 113 and the data transfer amount (or data transfer speed or bandwidth). The I / O adapter 113 is an interface with the switch fabric group 105.

  Here, the local QoS management unit 111 may be implemented as an independent application, or may be incorporated as a part of an application for system management. Further, the logical I / O management unit 112 may be incorporated in an I / O driver under the management of the OS 350. Alternatively, it may be incorporated as part of virtualization software such as a hypervisor instead of the I / O driver. In the first embodiment of the present invention, the local QoS management unit 111 is incorporated as a system management application, and the logical I / O management unit 112 is incorporated in an I / O driver in the OS 350. In the first embodiment of the present invention, the processing nodes 101 and 102 include the local QoS management units 111 and 116. However, when it is desired to manage the bandwidth of the I / O in the management node 103, the management node 103 May be provided with a local QoS management unit.

  The management node 103 includes an I / O adapter 120, a logical I / O management unit 121, and a global QoS management unit 122. The configuration of the I / O adapter 120 is the same as the configuration of the I / O adapter 113 of the processing node 101 described above.

  The logical I / O management unit 121 correlates the physical I / O adapter 120 on the management node 103, the port status information of the virtualized I / O adapter 120, and the data transfer amount (bandwidth). to manage.

  The global QoS management unit 122 includes state information of the entire information processing system 100, state information of processing nodes including the management node 103 (information of the local QoS management table 130), state information of the switch fabric group 105 (switch information 303), It manages state information (path information) of virtual I / O paths configured in the processing nodes 101 and 102 and the switch fabric group 105. In addition, the global QoS management unit 122 monitors the states of the switch fabric group 105 and the processing nodes 101 and 102, and transmits the virtual I / O path to each component (virtual) according to the configuration information set via the management console 104. I / O port etc.) is allocated, changed or deleted. For example, the status information such as the used bandwidth for each virtual I / O path is compared with the information at the time of setting the virtual I / O path, and if the used bandwidth does not satisfy the set bandwidth or exceeds the set bandwidth, The set bandwidth is changed in each component constituting the I / O path. Here, in the first embodiment of the present invention, the management node 103 includes the global QoS management unit 122. However, instead of the management node 103, the global QoS management unit 122 may be provided in a part of the processing node.

  The local QoS management unit 111 of the processing node 101 includes a local QoS management table 130, a QoS allocation management unit 131, and a bandwidth monitoring unit 132. The local QoS management table 130 manages port information (port information 204, logical I / O information 205) of the I / O adapter 113 mounted on the processing node 101. Details of the local QoS management table 130 will be described later (see FIG. 3). The QoS allocation management unit 131 operates in cooperation with the bandwidth monitoring unit 132, and in accordance with the contents of the local QoS management table 130, the I / O adapter 113 corresponding to a virtualized I / O path (virtual I / O path). Virtual I / O ports (virtual I / O ports) are generated, deleted, changed, and displayed. The bandwidth monitoring unit 132 periodically (predetermined predetermined status information (port information 204, logical I / O information 205) of virtual I / O ports configured in the I / O adapter 113 managed by the local QoS management unit 111). (Period)) to the global QoS management unit 122. Details of the bandwidth monitoring unit 132 will be described later (see FIG. 8).

  The global QoS management unit 122 of the management node 103 includes a global QoS management table 140, a QoS allocation management unit 141, and a bandwidth monitoring unit 142. The global QoS management table 140 manages state information (switch information 303) of the switch fabric group 105. Details of the global QoS management table 140 will be described later (see FIG. 4). The QoS management allocating unit 141 determines each related port (the I / O adapters 113 and 118 on each processing node and the switch fabric group 105 port) constituting the virtualized I / O path according to the contents of the global QoS management table 140. Allocate bandwidth to and change the associated table. Details of the QoS allocation management unit 141 will be described later (see FIG. 5). The bandwidth monitoring unit 142 monitors state information (switch information 303) of the switch fabric group 105. Details of the bandwidth monitoring unit 142 will be described later (see FIG. 6). Here, the related port is configured by information on a device from the I / O adapter 113 as a starting point to an I / O adapter as a communication destination (end point) and information on a port used by the virtual I / O path. For example, in the case of a virtual I / O path that communicates from the I / O adapter 113 of the processing node 101 to the I / O adapter 118 of the processing node 102 via the switch fabrics 170 and 171, the I / O adapter of the processing node 101 113 logical I / O information 205 includes identifiers of physical I / O ports and virtual I / O ports of the I / O adapter 113, identifiers of the switch fabrics 170 and 171 and identifiers of the ports used in each switch fabric. The identifiers of the physical I / O port and the virtual I / O port of the I / O adapter 118 serving as the end point are stored in the related port information of the logical I / O information 205. The related port information of the switch fabric group 105 can be expressed by an identifier of the switch fabric at the entrance, an identifier of the port, an identifier of the switch fabric as the exit, and an identifier of the port.

  The logical I / O management table 150 of the processing node 101 includes the state information (port information) of the physical I / O port of the I / O adapter 113 and the virtual I / O port virtually configured on the physical I / O port. 204, the correspondence between the logical I / O information 205) and the data transfer amount (bandwidth) in the virtual I / O port is held. The logical I / O management tables 151 and 152 of the processing node 102 and the management node 103 are also equivalent to the configuration of the logical I / O management table 150. Details of the logical I / O management table 150 will be described later (see FIG. 5).

  The controller 16 of the I / O adapter 113 includes a register 160. The register 160 is a memory mapped I / O mapped to the configuration space of the I / O adapter 113 of the processing node 101. Various settings of the I / O adapter 113 are recorded in the register 160. Here, in addition to various settings, the active state of the physical I / O port provided in the I / O adapter 113, the number of transmission / reception data, and the like are recorded. The register of the I / O adapter is the same for other nodes. The controller 16 of the I / O adapter 118 of the processing node 102 includes the register 161, and the controller 26 of the I / O adapter 120 of the management node 103 includes the register 162. Thus, the above data is stored. The register 161 is mapped to the configuration space of the processing node 102, and the register 162 is mapped to the configuration space managed by the management node 103. Each register is normally accessible from the mapped node, but only the management node 103 can access the registers 160 and 161 mapped to the address spaces of the processing nodes 101 and 102. At this time, the management node 103 has an address conversion unit (not shown) for accessing the registers 160 and 161 mapped to different address spaces.

  The switch fabric 170 is an I / O switch that constitutes the switch fabric group 105. The switch fabric 171 has the same configuration as the switch fabric 170. Setting of the switch fabric 170 is performed via the register 180. The register 180 is mapped to an address space (configuration space) accessible by the management node 103. The register 180 records, for example, the active state of the physical I / O port of each switch fabric, the number of transmission / reception data per physical I / O port, and the like.

  FIG. 3 is a configuration diagram of the local QoS management table 130 (133) according to the first embodiment of this invention. The local QoS management table 130 manages local state information of the processing node 101 on which the local QoS management unit 111 is mounted and the I / O adapter 113 on the processing node 101. Here, an example in which only the I / O adapter 113 is mounted on the processing node 101 is shown, but a plurality of I / O adapters may be mounted. The local QoS management table 130 holds fabric information 201, QoS service status 202, node information 203, port information 204, and logical I / O information 205. The fabric information 201 is an identifier of the switch fabric that becomes the owner of the processing node 101 in the switch fabric group 105. Here, the owner indicates a switch fabric that is the starting point of the processing node 101 in the switch fabric group 105.

  The QoS service state 202 indicates the activation state of the global QoS management unit 122. That is, a value indicating whether or not the global QoS management unit 122 of the management node 103 that manages the processing node 101 is activated is stored in the QoS service state 202.

  The node information 203 stores the identifier of the processing node 101 on which the local QoS management unit 111 is mounted and the number of physical I / O ports included in the I / O adapter 113 mounted on the processing node. If there are a plurality of I / O adapters, the number of physical I / O ports may be stored for each identifier of the I / O adapter.

  The port information 204 is generated logically (virtually) in the physical I / O port state (in use) of the I / O adapter 113, its maximum bandwidth (maximum transfer rate), and the physical I / O port. The number of virtual I / O ports stored is stored. When the I / O adapter 113 of the processing node 101 has a plurality of physical I / O ports, the port information 204 corresponding to the number of physical I / O ports is managed.

  The logical I / O information 205 is generated for each virtual I / O port virtually generated in the physical I / O port of the I / O adapter 113 on the processing node 101 on which the local QoS management unit 111 is mounted. It manages an O port identifier, a protocol identifier, a transmission source node identifier, a transfer destination node identifier, routing information, priority control information of data to be transferred, the number of related ports, related port information, and the like. Here, the number of related ports is the number of physical I / O ports that pass through the end-to-end data transfer path constituting the virtual I / O path, and the related port information includes the position information of the related ports and bandwidth control. Information.

  FIG. 4 is a configuration diagram of the global QoS management table 140 according to the first embodiment of this invention. The global QoS management table 140 holds state information of the entire switch fabric group 105 and state information of virtualized I / O (virtual I / O path) configured end-to-end in the information processing system 100. To do. Note that end-to-end refers to a path from the virtual I / O port of the I / O adapter 113 of the processing node 101 to the virtual I / O port of the I / O adapter 118 of the processing node 102, for example. The global QoS management table 140 includes fabric information 301, node information 302, switch information 303, and logical I / O information 304. The pointers stored in the information 301 to 304 indicate the storage location of the data table on the main memory 23. Point to.

  The fabric information 301 (310) is the entire information of the switch fabric. In the first embodiment of the present invention, the number of processing nodes, the number of switch fabrics, the identifier of the fabric owner, the information processing system existing in the switch fabric group 105 100 includes the number of virtual I / O paths generated. In the first embodiment of this invention, the fabric information 301 is managed by the fabric information table 310, and the global QoS management table 140 manages a pointer to the fabric information table 310.

  The node information 302 manages information on all processing nodes existing in the information processing system 100. In the first embodiment of the present invention, the node information 302 is managed by the node information table 320, and the global QoS management table 140 manages the node identifier and the corresponding pointer of the node information table 320. The node information table 320 includes, for each node identifier, the operating status of the local QoS management unit, the number of physical I / O ports of the I / O adapter installed in the node, status information for each physical I / O port, and physical I / O. The status information of the virtual I / O port generated in the port is held. Here, the status information of the physical I / O port and the status information of the virtual I / O port hold an identifier of each port and a value indicating the usage status.

  The switch information 303 is held in the switch information table 330 and manages state information of all switch fabrics existing in the switch fabric group 105. In the first embodiment of this invention, the switch information 303 is managed by the switch information table 330, and the global QoS management table 140 is the switch identifier of the switch fabrics 170 and 171 and the switch information corresponding to this identifier. A pointer to the table 330 is managed. The switch information table 330 is managed for each switch fabric identifier, and individual tables are generated for the number of physical I / O ports of the switch fabric, physical I / O port status information (usage status), and physical I / O ports. The status information (usage status) of the virtual I / O port that has been set is retained.

  The logical I / O information 304 manages configuration information and status information of a virtual I / O path (virtual I / O path) generated in the information processing system 100. These pieces of information are managed in the logical I / O information table 340 for each virtual I / O path. The logical I / O information table 340 includes a protocol identifier, a transmission source node identifier, a transmission destination node identifier, routing information, priority control information, bandwidth information, and the number of related ports constituting the virtual I / O path for each virtual I / O path. , Hold related port information.

  Since the virtual I / O path is an end-to-end virtual I / O path configured between the transmission source node and the transmission destination node, the physical I / O ports of a plurality of nodes and a plurality of switch fabrics are used. . The physical I / O ports of a plurality of nodes and a plurality of switch fabrics used to configure this virtual I / O path are referred to as related ports.

  FIG. 5 is a configuration diagram of the logical I / O management table 150 of the processing node 101 according to the first embodiment of this invention. The logical I / O management table 150 indicates the identifier of the virtual I / O port configured in the I / O adapter 113 of the processing node 101, the identifier of the physical I / O port, and the data transfer amount in the virtual I / O port. Hold. The logical I / O management table 150 includes a virtual I / O path identifier 401, a physical I / O port identifier 402, and a data transfer amount 403. The virtual I / O path identifier 401 is an identifier of an I / O path virtually configured in the information processing system 100 and is unique within the information processing system 100.

  In FIG. 5, there are two virtualized I / O ports configured on the physical I / O port 0 of the I / O adapter 113, and the virtual I / O path identifiers are 1 and 4, respectively. Show. Other virtual I / O path identifiers are configured other than the I / O adapter 113 and are not managed by the logical I / O management unit 112 on the processing node 101.

  The physical I / O port identifier 402 is an identifier of a physical I / O port included in the I / O adapter 113. When the I / O adapter 113 has two physical ports, an identifier is assigned to each physical port. Is granted. FIG. 5 shows that only the physical I / O port having the identifier “0” is used. The data transfer amount 403 holds the data transfer amount transferred on the virtual I / O port configured as the physical I / O port managed by the physical I / O port identifier as the number of packets. In FIG. 5, the data transfer amount in the virtual I / O port having the virtual I / O path identifier 1 is X0 packet, and the data transfer amount in the virtual I / O port having the virtual I / O path identifier 4 is X2 packet. is there. As the data transfer amount 403, the number of bytes and bits of data and commands transmitted and received within a predetermined time can be used.

  When attention is paid to a virtual I / O port whose virtual I / O path identifier 401 is “1”, another virtual I / O on the physical I / O port “0” in which this virtual I / O port is set. Port “4” is the associated port.

  FIG. 6 is a block diagram of the QoS assignment management unit 141 of the management node 103 according to the first embodiment of this invention. The QoS allocation management unit 141 is one of the processing functions of the global QoS management unit 122, and generates, deletes, allocates, and displays a virtual I / O path based on a command from the management console 104. The QoS allocation management unit 141 includes a virtual I / O path generation unit 501, a virtual I / O path deletion unit 502, a virtual I / O path change unit 503, and a virtual I / O path display unit 504.

  The virtual I / O path generation unit 501 includes a virtual I / O path selection unit 510 and a virtual I / O path registration unit 511. When the virtual I / O path generation unit 501 receives a virtual I / O path generation request via the management console 104, the virtual I / O path selection unit 510 selects the shortest path constituting the virtual I / O path. To do.

  Further, the virtual I / O path registration unit 511 refers to the information in the global QoS management table 140, and creates a new virtual I / O path for each related port of the shortest path selected by the virtual I / O path selection unit 510. Is set, and if it can be set, a virtual I / O path is set for the related port constituting the selected shortest route, and the virtualized I / O in the information processing system 100 is set. An O path is generated. The shortest route is, for example, a route in which the number of switch fabric stages constituting the virtual I / O path is the smallest. The virtual I / O path generation request includes the identifier of the transmission source node, the identifier of the transmission destination node, the set bandwidth of the virtual I / O path, and the priority. The flow of the virtual I / O path generation process will be described later (FIG. 9).

  The virtual I / O path deletion unit 502 receives a virtual I / O path deletion request from the outside such as the management console 104 and deletes a virtual I / O path already configured in the information processing system 100.

  The virtual I / O path change unit 503 receives a virtual I / O path change request from the outside such as the management console 104, and sets the set bandwidth and priority of the virtual I / O path already configured in the information processing system 100. change.

  The virtual I / O path display unit 504 displays all the virtual I / O paths configured in the information processing system 100 or the status information of the virtual I / O path to be requested on the management console 104 or the like.

  FIG. 7 is a configuration diagram of the bandwidth monitoring unit 142 of the management node 103 according to the first embodiment of this invention. The bandwidth monitoring unit 142 is one of the processing functions of the global QoS management unit 122. The bandwidth monitoring unit 142 includes a local QoS management information reception unit (local information acquisition unit) 601, a local QoS operation confirmation unit 602, a switch information acquisition unit 603, a global QoS information holding unit 604, and a global QoS information transmission unit 605.

  The local QoS management information receiving unit 601 receives local QoS management information from a processing node in which the local QoS management processing is operating, among the processing nodes 101 and 102 constituting the information processing system 100. The local QoS management information to be received includes update information that has been changed in the local QoS management tables 130 and 133 managed by each processing node, and the operating rate of the processing node. Note that the utilization rate (load factor) of the CPU 10 acquired by the OSs 350 and 351 can be used as the operation rate of the processing node. Further, as the operation rate, in addition to the usage rate of the CPU 10, the usage rate of the virtual I / O port (the bandwidth actually used / the set bandwidth) can be used, and the usage of a plurality of virtual I / O ports can be used. The average value of the rate may be used as the virtual I / O port usage rate for each processing node.

  The local QoS operation confirmation unit 602 refers to the global QoS management table 140 and confirms the operation status of each local QoS management unit 111, 116. Specifically, the operating status of the local QoS management unit 111 and the local QoS management unit 116 recorded in the node information 302 of the global QoS management table 140 shown in FIG. 4 is confirmed. As a result, the operating state of the processing node managed by the management node 103 can be grasped as a heartbeat indicating that each processing node is operating normally.

  The switch information acquisition unit 603 acquires the above-described switch information of the switch fabrics 170 and 171 constituting the information processing system 100. The switch information acquisition unit 603 acquires switch information of the entire switch fabric group 105 when a command is issued from the management node 103.

  The global QoS information holding unit 604 holds the contents of the global QoS management table 140 managed by the global QoS management unit 122 if the information acquired by the local QoS operation confirmation unit 602 and the switch information acquisition unit 603 is changed.

  When there is a change in the configuration information of the switch fabric group 105 or processing node managed by the global QoS management table 140, the global QoS information transmission unit 605 notifies the change contents to all the local QoS management units 111 and 116. .

  The switch information acquisition unit 603 further includes a node selection unit 610, an acquisition request transmission unit 611, and an acquisition response reception unit 612. The node selection unit 610 checks the status information of the processing node held in the global QoS management table 140 by the global QoS management unit 122 and selects a processing node with a low operation rate. The node selection unit 610 selects a processing node with a low operation rate, thereby minimizing the influence of the switch information acquisition processing on other processing such as application processing and local QoS management processing. For example, the processing node may be selected by reading the usage rate of the CPU 10 acquired by the OSs 350 and 351 and selecting the one with the lowest usage rate or the usage rate less than a predetermined threshold. As the operation rate, an average value of the ratio between the set value of the virtual I / O port bandwidth of each processing node and the actually used bandwidth (maximum bandwidth) may be used as the operation rate.

  The acquisition request sending unit 612 sends a switch information acquisition request to the processing node with the low operation rate selected by the node selection unit 610. The acquisition response reception unit 613 receives a response (switch information) to the acquisition request from the processing node that has transmitted the acquisition request.

  As described above, the bandwidth monitoring unit 142 of the management node 103 receives the local QoS management tables 130 and 133 of the processing nodes managed by the local QoS management units 111 and 116 from the respective processing nodes 101 and 102, and each processing node 101 , 102, the information such as the bandwidth of the virtual I / O port set in the I / O adapter is acquired, and the global QoS management table 140 is updated. Further, the bandwidth monitoring unit 142 notifies the processing nodes 101 and 102 of the updated information when the global QoS management table 140 is updated, and transmits a change to the virtual I / O path set by the management node 103. . Then, the bandwidth monitoring unit 142 can collect a switch information of the switch fabric group 105 by selecting a processing node having a low operation rate from a plurality of processing nodes and issuing a switch information acquisition request.

  FIG. 8 is a configuration diagram of the bandwidth monitoring unit 132 of the processing node 101 according to the first embodiment of this invention. The bandwidth monitoring unit 132 is a part of the function of the local QoS management unit 111. The bandwidth monitoring unit 135 of the local QoS management unit 116 of the processing node 102 has the same configuration as the bandwidth monitoring unit 132 of the processing node 101.

  The bandwidth monitoring unit 132 includes a logical I / O bandwidth calculation unit 701, a register information acquisition unit 702, a local QoS information holding unit 703, a local QoS information transmission unit 704, a global QoS operation confirmation unit 705, a global QoS information reception unit 706, a switch An information acquisition unit 707 is provided.

  The logical I / O bandwidth calculation unit 701 calculates a bandwidth (transfer rate) for each virtual I / O port set as a physical I / O port of the I / O adapter 113. The register information acquisition unit 702 accesses the I / O adapter 113 and acquires state information (adapter information) of the register 160 of the I / O adapter 113. The adapter information includes a physical I / O port identifier of the I / O adapter 113, a virtual I / O port identifier set for the physical I / O port, and a value indicating the usage status of the virtual I / O port. And stored in the port information 204 and the logical I / O information 205 of the local QoS management tables 130 and 133.

  The local QoS information holding unit 703 holds the information (bandwidth and adapter information) acquired by the logical I / O band calculation unit 701 and the register information acquisition unit 702 in the local QoS management table 130.

  The local QoS information transmission unit 704 sends the information of the local QoS management table 130 updated by the local QoS information holding unit 703 to the global QoS management unit 122 of the management node 103. The global QoS service operation confirmation unit 705 confirms the operation status of the global QoS management unit 122 of the management node 103.

  The global QoS information acquisition unit 706 acquires updated information among the information held in the global QoS management table 140 of the management node 103. That is, if the information (setting value) regarding the processing node 101 among the information of the global QoS management table 140 of the management node 103 has been updated, the global QoS information acquisition unit 706 updates the local QoS management table 130. As a result, the virtual I / O port settings set by the management node 103 are transmitted to the processing nodes 101 and 102.

  The switch information acquisition unit 707 includes an acquisition determination unit 720 and a switch information acquisition processing unit 721. The acquisition determination unit 720 checks whether or not there is a switch state information acquisition request sent from the bandwidth monitoring unit 142 of the global QoS management unit 122 of the management node 103. If there is a switch state acquisition request, the acquisition determination unit 720 721 is executed, all switch information of the switch fabric group 105 is acquired, and the switch information acquisition result is returned to the bandwidth monitoring unit 142 of the management node 103.

  The switch information acquisition processing unit 721 accesses the registers 160 and 161 of the switch fabrics 170 and 171 configuring the switch fabric group 105 and acquires the switch information of all ports of all the switch fabric groups 105 configuring the switch fabric group 105. To do. As described above, the switch information includes identifiers of physical I / O ports and virtual I / O ports, and status information (usage status, etc.).

  The logical I / O band calculation unit 701 further includes a difference time calculation unit 710, a logical I / O information acquisition unit 711, a logical I / O band calculation unit 712, a logical I / O information reset unit 713, and a previous monitoring time 714. Serve.

  The difference time calculation unit 710 obtains the current time and calculates the difference from the previous monitoring time recorded in the previous monitoring time 714. The logical I / O information acquisition unit 711 accesses the logical I / O management unit 112 on the processing node 101 and transfers data for each virtual I / O path held in the logical I / O management table 150 of the processing node 101. The quantity 403 is acquired.

  The logical I / O bandwidth calculation unit 712 uses the difference time calculated by the difference time calculation unit 710 and the data transfer amount for each virtual I / O path acquired by the logical I / O information acquisition unit 711. The bandwidth (transfer rate) for each port of the port and switch fabric group 105 is calculated. The logical I / O information reset unit 713 resets the data transfer amount for each virtual I / O path recorded in the logical I / O management table 150. By this processing, the bandwidth (transfer rate) per unit time for each virtual I / O port is calculated. When the data transfer amount for each virtual I / O path is reset, the previous monitoring time 714 is updated to the current time.

  FIG. 9 is a diagram illustrating a virtual I / O path generation processing flow executed by the virtual I / O path generation unit 501 of the global QoS management unit 122 of the management node 103 according to the first embodiment of this invention. . When the virtual I / O path generation unit 501 receives a virtual I / O path generation request via the switch fabric group 105 from a management program on the management console 104 or a management application on the processing node, the virtual I / O path generation processing is performed. Start (step 801). Here, the virtual I / O path generation request includes a new virtual I / O path transmission source identifier, a new virtual I / O path transmission destination identifier, a new virtual I / O path setting bandwidth (transfer rate). ), The priority of the new virtual I / O path is included as an argument.

  In step 802, the argument of the virtual I / O path generation request is decoded, and the processing node used in the new virtual I / O path and the related port of the switch fabric are determined from the transmission source identifier and transmission destination identifier included in the request. (Step 802). Here, the method for determining the related port is selected so as to form the shortest path. Alternatively, it may be selected from a plurality of route candidates. That is, the virtual I / O path generation unit 501 selects the shortest route from the start and end points of the virtual I / O path included in the received virtual I / O path generation request, and relates the ports on the selected route. Extract as a port.

  Next, after determining the related port on the virtual I / O path, the global QoS management table 140 is searched to determine whether or not the bandwidth of the requested virtual I / O path can be newly allocated for each related port. Is checked (step 803). This processing is performed, for example, when the bandwidth set for each related port and the newly added virtual I / O path bandwidth are within the physical bandwidth of each port on the virtual I / O path. Thus, it can be determined that a new virtual I / O path can be set. Alternatively, a new virtual I / O path can be set if the value obtained by adding the bandwidth used by each related port and the bandwidth of the new virtual I / O path is within the physical bandwidth of each port. You may judge. In this case, it is desirable to compare the value obtained by adding the maximum bandwidth used by the related port and the bandwidth of the new virtual I / O path with the physical bandwidth of each port on the virtual I / O path. If all the related ports can be allocated, the process of step 805 is performed, and if the allocation is impossible, the process of step 810 is performed.

  In addition, when checking whether or not a new virtual I / O path bandwidth can be allocated based on the bandwidth and priority included in the argument of the virtual I / O path generation request, the following processing is performed. Is desirable. For example, if the physical bandwidth to be newly set as a high priority virtual I / O path is insufficient and the bandwidth of a virtual I / O path with a low priority has already been allocated, a bandwidth with a low priority is assigned. It is notified to the administrator via a management node whether or not new allocation can be performed using a message or the like, and only when there is an approval from the administrator, the existing low priority virtual I / O The bandwidth of the virtual I / O path with high priority is secured by reducing the physical bandwidth of the O path. Alternatively, if the physical bandwidth of the new virtual I / O path is insufficient and a bandwidth with the same priority as the virtual I / O path to be newly assigned is already set, a message is sent to the administrator Then, the options such as whether to interrupt the new generation process, reduce the allocated bandwidth, or secure the newly allocated bandwidth with an allowable physical bandwidth are shown. Perform allocation control. Here, an example has been shown in which the administrator selects whether or not to make a new allocation. However, as another method, the bandwidth when the physical bandwidth of the newly configured virtual I / O path is insufficient in the management node in advance. A method of automating the virtual I / O path generation process without making an inquiry to the administrator by setting an allocation policy is also conceivable. For example, if a policy is set in advance to reduce the lower priority bandwidth and secure the physical bandwidth of the new virtual I / O path, the virtual I / O path generation process uses a high priority virtual If the physical bandwidth is insufficient when generating the I / OO path, the physical bandwidth of the existing virtual I / O path with low priority is reduced to secure the bandwidth of the new virtual I / O path.

  In step 805, since the bandwidth can be newly allocated for all the related ports, the newly generated virtual I / O path identifier is registered in the global QoS management table 140. Next, a new virtual I / O path identifier is registered in the logical I / O management table on the processing node having the related port (step 807). If there are a plurality of related ports, the processing in step 807 is repeated according to the number of related ports.

  In step 808, the virtual I / O path identifier is notified to the management program of the management console 104 as a response to the virtual I / O path generation request, and the generation process is terminated. The virtual I / O path identifier is a value that is uniquely assigned by the virtual I / O path generation unit 501 within the information processing system 100.

  If it is determined in step 804 that new allocation is not possible, in step 810, the console 104 is notified that a new virtual I / O path cannot be generated, and the process ends. Alternatively, instead of performing step 801, the process may return to step 802 and automatically or manually select a virtual I / O path so that the virtual I / O path passes through another route. In this case, step 810 is performed only when a virtual I / O path cannot be generated no matter what route is selected, and notification that it cannot be generated. The new virtual I / O path generation process is completed by the above processing flow. As a result, information on the generated new virtual I / O path is set in the global QoS management table 140. Information on the new virtual I / O path set in the global QoS management table 140 is acquired by the global QoS information acquisition unit 706 of the processing node 101, and the QoS allocation management unit 131 of the processing node 101 notifies the OS 350, The I / O driver in the OS 350 sets a new virtual I / O path, that is, a virtual I / O port, in the I / O adapter 113.

  FIG. 10 is a flow diagram of bandwidth monitoring processing of a virtual I / O path (virtual I / O path) in the global QoS management unit 122 of the management node 103 according to the first embodiment of this invention. When the global QoS management unit 122 starts the bandwidth monitoring process, the bandwidth monitoring process is started (step 901). The global QoS management unit 122 first sleeps for a certain period of time in order to execute the bandwidth monitoring process at predetermined time intervals (step 902). The sleep time is set short when it is desired to improve the accuracy of bandwidth monitoring, and is set long when accuracy is not required. For example, when performing bandwidth monitoring at the same level as general heartbeat processing, the sleep time is set to about 1 second.

  Next, local QoS information reception processing (the above processing of the local QoS management information receiving unit 601) is performed (step 903). In this step, local QoS management information is acquired from the processing node in which the local QoS management processing is operating among the processing nodes 101 and 102 constituting the information processing system 100. The global QoS management unit 122 can determine whether each processing node 101, 102 is operating normally from the presence or absence of a response. Then, the global QoS management unit 122 receives local QoS management information (information of the local QoS management tables 130 and 133) from all the local QoS management units 111 and 116 that are in operation.

  Subsequently, local QoS operation confirmation processing (the above processing of the local QoS operation confirmation unit 602) is performed (step 904). In the local QoS operation confirmation process, the local QoS management information sent from the processing node is checked in step 904 to confirm whether all the local QoS management units 111 and 116 are in an operating state. Alternatively, the operating status of each processing node 101 or 102 may be confirmed using information recorded in the node information table 320 of the global QoS management table 140.

  Next, the global QoS management unit 122 performs node selection processing (processing by the node selection unit 610) (step 905). In the node selection process, the operating status of each processing node included in the local QoS information is confirmed, and a processing node with a low operating rate is selected. At this time, when there are a plurality of processing nodes with a low operation rate, they may be selected at random, or after selecting a processing node by round robin, it is determined whether the processing node has a low operation rate, If the rate is high, the next processing node may be selected. Note that, as described above, for example, the usage rate of the CPU 10 of each processing node may be used as the operating rate of the processing node, and the processing node with the lowest usage rate of the CPU 10 may be selected as a processing node with a low operating rate.

  In step 906, it is determined whether or not the selected processing node is its own node (management node 103). If the selected processing node is its own node, switch information acquisition processing is performed (step 907). In the switch information acquisition processing in step 907, the registers 180 and 181 of the switch fabrics 170 and 171 constituting the switch fabric group 105 are accessed, and the status information (switch information) of all the ports of the switch fabric group 105 from the registers 180 and 181. Is read. The acquired switch information is reflected in the switch information 303 of the global QoS management table 140.

  On the other hand, if the selected processing node is not its own node in step 906, switch information acquisition request sending processing (processing of the acquisition request sending unit 611) is performed on other processing nodes (step 908). In the acquisition request sending process, an acquisition request for acquiring the status information of the switch fabric group 105 is transmitted to the selected processing node. Next, in the acquisition response reception processing (processing of the acquisition response reception unit 612), a response from the processing node to which the acquisition request is sent is received (step 909).

  Subsequently, the global QoS information update process is performed, and the local QoS management tables 130 and 133 for each processing node received in the local QoS information reception process in step 903 and the update information of the switch information acquired in the acquisition response reception process in step 909 Is reflected in the global QoS management table 140 (step 910). Finally, global QoS information transmission processing is performed (step 911). In the global QoS information transmission process, the operating status of the global QoS management unit 122 (information of the global QoS management table 140) is transmitted to all the local QoS management units 111 and 116, and the switch fabric group 105 managed by the global QoS management unit 122 is used. And the processing node configuration are changed, the information is transmitted to all local QoS management units 111 and 116. These processes continue until the bandwidth monitoring process is interrupted (step 911).

  In the first embodiment of the present invention, the local QoS information reception process (step 902) and the global QoS information transmission process (step 911) are shown separately. However, the local QoS information reception process is a local QoS management of the transmission source. Information sent in the global QoS information transmission process may be added to the response sent back to the part. In that case, the process of step 911 is not necessary.

  Through the above processing, the global QoS management unit 122 receives the local QoS management tables 130 and 133 from the processing nodes 101 and 102 and updates the global QoS management table 140. Then, the global QoS management unit 122 instructs the processing node determined to have a low operation rate to collect switch information of the switch fabric group 105. Therefore, even when the switch fabric group 105 is large or when the number of virtual I / O ports per physical I / O port is large, the state information read processing of the switch fabrics 170 and 171 that become a heavy load is performed. The processing node having a low operation rate among the plurality of processing nodes connected to the switch fabric group 105 prevents the load from being concentrated on a specific processing node, thereby collecting information on the switch fabric group 105 at a certain level. It can be executed in time.

  FIG. 11 is a diagram illustrating a flow of local bandwidth monitoring processing executed by the local QoS management units 111 and 116 of the processing nodes 101 and 102 according to the first embodiment of this invention. Since the local QoS management unit 116 of the processing node 102 also performs the same processing, only the processing node 101 will be described below.

  When the local QoS management unit 111 starts the local bandwidth monitoring process, the local QoS monitoring unit 111 starts the local bandwidth monitoring process (step 1001). Since the local bandwidth monitoring process is executed at regular time intervals, the process first sleeps for a predetermined time period (step 1002). The sleep time is set short when the accuracy of bandwidth monitoring of the I / O port of the I / O adapter 113 is improved, and is set long when accuracy is not required. For example, when performing bandwidth monitoring at the same level as general heartbeat processing, the sleep time is set to about 1 second.

  Next, global QoS information reception processing (processing of the global QoS information acquisition unit 706) is performed (step 1003). In step 1003, information on the global QoS management table 140 is received from the global QoS management unit 122 of the management node 103. This information is received from the information held in the global QoS management table 140 when the configuration of the switch fabric group 105 or the number of processing nodes is changed.

  Subsequently, a global QoS service operation confirmation process (a process of the global QoS service operation confirmation unit 705) is performed (step 1004). In the global QoS service operation confirmation process, the information in the global QoS management table 140 sent from the management node 103 is checked in step 1003 to confirm whether the global QoS management unit 122 of the management node 103 is in an operating state. Thereby, it can be confirmed that the management node 103 is operating normally.

  Next, switch information acquisition processing (processing of the switch information acquisition unit 707) is performed (step 1005). The switch information acquisition process includes an acquisition determination process (step 1006) and a switch information acquisition process (step 1007). In step 1006, the process of the acquisition determination unit 720 checks whether there is a switch information acquisition request from the management node 103. If there is no switch information acquisition request, step 1007 is not performed and the next step is performed.

  On the other hand, if there is a switch information acquisition request in step 1006, switch information acquisition processing is performed in step 1007. In the switch information acquisition processing (processing of the switch information acquisition processing unit 721), the registers 180 and 181 of the switch fabric 170 and 171 constituting the switch fabric group 105 are accessed, and the status information of all the ports constituting the switch fabric group 105 is obtained. To get. After acquiring the status information of these ports, a response to the switch information acquisition request is sent to the global QoS management unit 122 of the transmission source.

  In step 1008, register information acquisition processing is performed. In the register information acquisition process, the status information (adapter information) of the physical I / O port set in the register 160 is acquired by accessing the register 160 of the I / O adapter 113 of the self-processing node. Specifically, the operating status of the physical I / O port and the data transfer amount are acquired.

  In step 1009, logical I / O band calculation processing (processing of the logical I / O band calculation unit 712) is performed. In the logical I / O bandwidth calculation processing, the bandwidth (transfer rate) for each virtual I / O path is calculated from the current time and the data transfer amount corresponding to the virtual I / O path recorded in the logical I / O management table 150. Is calculated. Details will be described later (FIG. 12).

  Subsequently, when local QoS information transmission processing (processing of the local QoS information transmission unit 704) is performed and information managed by the local QoS management table 130 is updated, the information is transferred to all local QoS management units. 151 (step 1010). In the local QoS information update process, the update information is reflected in the local QoS management table 130 (step 1011).

  The local bandwidth monitoring process continues from step 1002 to step 1011 until interrupted (step 1012). In the first embodiment of the present invention, the global QoS information reception process (step 1003) and the local QoS information transmission process (step 1010) are described separately, but the response of the global QoS information reception process is described. The update information of the local QoS management table sent in the local QoS information transmission process may be included. In this case, the local QoS information transmission process in step 1010 can be omitted.

  Through the above processing, the local QoS management unit 111 of the processing node 101 acquires the information of the global QoS management table 140 from the management node 103 every predetermined time, and also acquires the adapter information of the own processing node, and acquires these. Of the information, information related to the own processing node is reflected in the local QoS management table 130. In addition, the local QoS management unit 111 notifies the updated information to other processing nodes when the local QoS management table 130 of the processing node 101 is updated.

  In addition, when there is a request from the management node 103, the local QoS management unit 111 of the processing node 101 acquires switch information of the switch fabric group 105 and transmits it to the management node 103.

  FIG. 12 is a diagram showing a flow of logical I / O bandwidth calculation processing performed in step 1009 of FIG. In the logical I / O bandwidth calculation processing executed by the bandwidth monitoring unit 132 of the processing node 101, first, the current time is acquired, the difference from the time when the previous bandwidth calculation processing was executed is calculated, and the current time is pre-monitored. Reflected in time 714 (step 1101).

  Next, the logical I / O management table 150 is accessed, and the data transfer amount for each registered virtual I / O path identifier is acquired (step 1102). Subsequently, a bandwidth (transfer rate) per unit time for each virtual I / O path identifier is calculated from the time difference calculated in step 1101 and the data transfer amount acquired in step 1102 (step 1103). The calculated bandwidth of each virtual I / O port is compared with the maximum bandwidth stored in the port information 204 of the local QoS management table 130. If the calculated bandwidth value is large, the port information 204 is updated. . Then, the data transfer amount for the virtual I / O path identifier registered in the logical I / O management table 150 is reset (step 1104). Through the above steps, the bandwidth (transfer rate) for each virtual I / O path (virtual I / O port) is calculated.

  As described above, according to the present invention, a plurality of virtualized I / Os are provided to the information processing system 100 that connects a plurality of processing nodes 101 and 102 via the switch fabric group 105 (I / O switch). Since paths (virtual I / O paths) can be configured and the transfer rate (bandwidth) can be controlled for each virtual I / O path, end-to-end QoS can be guaranteed for each virtualized I / O path . In addition, even an I / O switch such as the switch fabric group 105 without a CPU manages state information (switch information) for each I / O port virtually configured as a physical I / O port. Information can be collected. Furthermore, even when there are a large number of switch fabric groups 105 and a large number of virtual I / O ports per physical I / O port, the switch fabric group 105 can perform a process of reading the state information of the switch fabric that is highly loaded. By sharing the processing nodes with a low operating rate among the plurality of processing nodes to be connected, the influence of other communications can be minimized and information collection of the switch fabric group 105 can be executed in a certain time.

<Modification 1>
In the first embodiment of the present invention, the virtual I / O path generation unit 510 selects a route with the shortest virtual I / O path by the virtual I / O path selection unit 510. However, when the shortest path is selected for the virtual I / O path, when a failure occurs in a part of the virtual I / O path or when many virtual I / O paths are set for a specific port If the virtual I / O path can be selected more flexibly, stable communication can be provided even in the event of a failure or route congestion.

  Therefore, in the first modification of the present invention, when the virtual I / O path is generated, not based on the shortest path but based on the total bandwidth set for the virtual I / O path and the actual used bandwidth of the virtual I / O path. Make it possible to select from multiple candidates. Even when a failure occurs in a part of the virtual I / O path, the virtual I / O path generation unit 510 can configure a bypassable route.

  FIG. 13 is a block diagram of the virtual I / O path generation unit 510 of the QoS allocation management unit 141 of the global QoS management unit 122 executed by the management node 103.

  The virtual I / O path generation unit 510 includes a virtual I / O path list creation unit 1201, a virtual I / O path utilization rate calculation unit 1202, a virtual I / O path selection unit 1203, a virtual I / O path registration unit 1204, a virtual An I / O path candidate list 1205 and a condition flag 1206 are provided.

  Upon receiving a virtual I / O path generation request from the management console 104, the virtual I / O path list creation unit 1201 can configure a virtual I / O path from the transmission source node and the destination node of the virtual I / O path. Create a list of routes. The virtual I / O path list creation unit 1201 refers to the global QoS management table 140, searches for a preset finite number of routes, and registers the searched routes in the virtual I / O path candidate list 1205.

  The virtual I / O path usage rate calculation unit 1202 refers to the global QoS management table 140 for each route searched by the virtual I / O path candidate list 1205, and determines the maximum value of the total set bandwidth and the used bandwidth. The maximum value of the total value is calculated and reflected in the virtual I / O path candidate list 1205.

  The virtual I / O path selection unit 1203 selects a virtual I / O path from the virtual I / O path candidate list 1205 according to the value set in the condition flag 1206.

  The virtual I / O path candidate list 1205 is a list created corresponding to the logical I / O identifier. The virtual I / O path candidate list 1205 includes a candidate number 1210, a route length 1211, a maximum set bandwidth 1212, and a maximum use bandwidth 1213. The path length 1211 is the number of stages of the switch fabrics 170 and 171 between the transmission source node and the transmission destination node constituting the virtual I / O path. The maximum set bandwidth 1212 indicates the maximum set bandwidth out of the total value of the bandwidths already set for each related port constituting the virtual I / O path. The total value of the set bandwidth is obtained by referring to the global QoS management table 140. The maximum used bandwidth 1213 indicates the maximum used bandwidth among the total values of the used bandwidths of the related ports constituting the virtual I / O path. The total value of the used bandwidth is also obtained by referring to the global QoS management table 140.

  The condition flag 1206 holds a condition for selecting a virtual I / O path route. The selection condition used in the condition flag 1206 is included in the selection condition parameter of the virtual I / O path generation request. For example, the selection condition parameter is “1” when a route with the shortest path length is selected, “2” when a route with the minimum set bandwidth is selected, and a route with the minimum use bandwidth is selected. Is set to “3”. When “1” is set in the selection condition parameter of the virtual I / O path generation request, the virtual I / O path selection unit 1203 selects the candidate with the smallest available bandwidth from the virtual I / O path candidate list 1205. A virtual I / O path having a number is selected. In the first modification, three conditions are shown, but a virtual I / O path route may be selected based on other conditions. In Modification 1, the condition flag 1206 is included as a parameter for the virtual I / O path generation request. However, the condition flag 1206 is uniquely set in advance by the management program or the like instead of being included in the parameter. May be.

  As described above, in the first modification, the route selection of the virtual I / O path can be changed according to the selection condition parameter, and the virtual I / O path can be made more flexible even when a failure occurs or the route is congested. Setting is possible, and stable communication can be provided using a virtual I / O path.

<Modification 2>
In the first embodiment of the present invention, the logical I / O management unit 112 is incorporated in the I / O driver of the OS 350, and the status information of each virtual I / O path is managed by the I / O driver. The second modification of the present invention is different from the first embodiment in that the logical I / O management unit 112 is incorporated in a virtual machine monitor such as a hypervisor. By incorporating the logical I / O management unit 112 into the hypervisor, it is possible to manage the status information of the virtual I / O path without changing the I / O driver in the OS. In the second modification of the present invention, the local QoS management unit is also incorporated in the hypervisor. Therefore, end-to-end between the virtual host (guest OS on the LPAR) or between the virtual host and another processing node (I / O device) without degrading the performance of the application installed in the logical partition (LPAR). The virtual I / O path can be configured and managed. Hereinafter, a different part from the 1st Embodiment of this invention is demonstrated.

  FIG. 14 is a block diagram of the information processing system 100 in the second modification of the present invention. The information processing system 100 includes processing nodes 1301 and 1302, a management node 1303, a management console 1304, and a switch fabric group 1309. In the second modification, not only a management request for QoS from an administrator, for example, a request for assigning a QoS to a virtual I / O path to be configured, is sent from the management console 3104 to the management node 1303 but also the LPAR. Is sent from the management console 3104 to the hypervisors 1310 and 1311. In the second modification, the management console 1304 directly sends a management request to the hypervisors 1310 and 1311. However, the management request may be sent to the hypervisors 1310 and 1311 via the management node 1303 and the switch fabric group 1309. .

  The processing node 1301 includes a hypervisor 1310, an I / O adapter 1312, and LPARs 1320 to 1321. The hypervisor 1310 is virtualization software for logically dividing and managing hardware resources of the processing node 1301, and the LPAR management unit 1330 is a known technique for managing logical partitions generated on the hypervisor 1310.

  The second modification includes a local QoS management unit 1332 and a logical I / O management unit 1334 in addition to the LPAR management unit 1330. Here, in the second modification, the hypervisor 1310 includes a local QoS management unit 1332 and a logical I / O management unit 1334, which is different from the first embodiment. On the other hand, the functions of the local QoS management unit 1332 and the logical I / O management unit 1334 are equivalent to the functions described in the first embodiment. However, in the first embodiment, in addition to associating the processing node state information with the logical I / O information, in the second modification, the LPAR-virtual I / O path information correspondence table 1336 and the mapping unit 1337 is further different.

  The LPAR-virtual I / O path information correspondence table 1336 holds the correspondence between the LPAR status information managed by the LPAR management unit 1330 and the virtual I / O path information. The mapping unit 1337 has an interface for accessing the LPAR management unit 1330. The mapping unit 1337 accesses the LPAR management unit 1330, refers to an LPAR management table (not shown) that is a correspondence table between LPAR identifiers and resources managed by the LPAR management unit 1330, and refers to LPAR identifiers and virtual I / O path information. Associate.

In the second modification, the structure of the source node identifier and the transfer destination node identifier included in the virtual I / O path configuration request (generation request, deletion request, change request) from the management console 3104 is the first embodiment. And different. In the second modification, a virtual I / O path is configured between virtual hosts or between a virtual host and a processing node (or I / O node). Contains an identifier. Alternatively, when the hypervisor 1310 has an identifier, a combination of the hypervisor identifier and the LPAR identifier is treated as a node identifier so that a virtual host can be specified.
The function of the global QoS management unit 1322 is also the same as that of the first embodiment, except that the hypervisor 1310 includes a communication interface for exchanging information with the local QoS management unit 1332 installed in the hypervisor.

  As described above, when the computer resources are virtualized by the processing nodes 1301 and 1302, the local QoS management units 1332 and 1333 and the logical I / O management units 1334 and 1335 are incorporated in a virtualization system such as a hypervisor. Thus, virtual I / O paths can be set between virtual hosts (guest OSs) or between virtual hosts and other processing nodes and managed in the same manner as in the above embodiment.

<Modification 3>
In the second modification, the hypervisor 1310 includes the local QoS management unit 1332 and the logical I / O management unit 1334. The third modification is different in that the hypervisor 1310 does not include the local QoS management unit 1332 but the LPAR 1321 includes the local QoS management unit 1332. By providing the local QoS management unit 1332 in the LPAR 1321, hardware resources such as a CPU and a memory used by the local QoS management unit 1332 for bandwidth monitoring can be limited to a certain utilization rate.

  In the second and third modification examples, the computer resource is virtualized by the hypervisor 1310. However, the present invention can be applied to a processing node that performs virtualization using a virtual machine monitor. .

<Modification 4>
In the first embodiment of the present invention, only the global QoS management unit 122 has the state information of the entire information processing system 100, the state information of the processing node including the management node 103, the state information of the switch fabric group, the processing node and the switch fabric. The status information of virtual I / O paths configured in groups is managed. Therefore, when a failure occurs in the global QoS management unit 122, the contents of the global QoS management table may be lost. Thus, in the fourth modification, management nodes on which the global QoS management unit is mounted are provided at a plurality of locations so that the reliability can be improved.

  FIG. 15 is a block diagram of the information processing system 100 in Modification 4 of the present invention. The information processing system 100 includes processing nodes 101 and 102, a master management node 1401, a sub management node 1402, a management console 1405, and a switch fabric group 1409.

  In the fourth modification, unlike the first embodiment, a plurality of management nodes 1401 and 1402 each equipped with a global QoS management unit are provided, one of which functions as a master management node 1401, and the other functions as a sub management node 1402. To do. Further, some of the functions of the local QoS management units 111 and 116 and the global QoS management units 1414 and 1415 mounted on each processing node and each management node are different from those of the first embodiment. Hereinafter, a description will be given focusing on differences from the first embodiment.

  In the fourth modification, the local QoS management unit 111 of the processing node 101 has management node information 1452 newly. The management node information 1452 holds node identification information of the master management node 1401 and the sub management node 1402, respectively. The management node information 1450, 1451, 1453 has the same configuration as the management node information 1452.

  Next, the basic configuration of the bandwidth monitoring unit 1460 of the processing node 101 is the same as that of the bandwidth monitoring unit 132 shown in the first embodiment, but the transmission destination of the local QoS management information (local QoS management table 130). The node and the target node for checking the global QoS operation status are different from those in the first embodiment. Specifically, the bandwidth monitoring unit 142 of the processing node 101 transmits local QoS management information to the global QoS management unit 1414 of the master management node 1401 and confirms the operation status of the global QoS management unit 1415 of the sub management node 1402. .

  The bandwidth monitoring unit 1445 of the global QoS management unit 1414 of the master management node 1401 refers to the management node information 1450, collects local QoS information (information of the local QoS management tables 130 and 133) of the processing nodes 101 and 102, and performs local processing. The operating status of the QoS management units 111 and 116 is confirmed. Then, the bandwidth monitoring unit 1445 updates the global QoS management table 1443 with the collected local QoS information, and then sends a copy of the global QoS management table 1443 to the global QoS management unit 1415 of the sub management node 1402.

  The bandwidth monitoring unit 1446 of the global QoS management unit 1415 of the sub management node 1402 refers to the management node information 1451, collects update information of the global QoS management table 1443 of the master management node 1401, and acquires the global QoS management unit 1414 on the master side. And the update information is reflected in the global QoS management table 1448. Then, the bandwidth monitoring unit 1446 sends the updated contents of the global QoS management table 1448 to the local QoS management units 111 and 116 of the processing nodes 101 and 102.

  FIG. 16 is a diagram showing a flow of bandwidth monitoring processing of the global QoS management units 1414 and 1415 according to the fourth modification of the present invention. Since the flow in the case where the own node is the master management node 1401 and all the local QoS management units 111 and 116 are operating normally is the same as that in FIG. 10, in the fourth modification, the flow is different from that in FIG. explain.

  In step 1503, it is determined with reference to the management node information 1450 whether the own node is the master management node 1401 or the sub management node 1402. When the own node is the sub management node 1402, the update information of the global QoS management table 1443 is received from the master management node 1401 (step 1520).

  Then, the operating status of the global QoS management unit 1414 of the master management node 1401 is confirmed (step 1521). If the master management node 1401 is operating normally, the contents of the global QoS management table 1448 of the sub management node 1402 are updated (step 1508). If the operation cannot be confirmed, it is determined that a failure has occurred in the master management node 1401, and the master management node 1401 is changed (step 1523). In the change process of the master management node 1401, the identification information of the sub management node 1402 is held in the master management node of the management node information 1451 of the sub management node 1402.

  Subsequently, the sub management node 1402 is selected, the global QoS management table 1448 is initialized, and the global QoS management unit 1415 is initialized, and the sub management node 1402 takes over the master management node 1401 to each processing node. The management node information 1452 and 1453 of each processing node 101 and 102 is updated (step 1524).

  If the local node is the master management node 1401 and the local QoS operation status is confirmed (step 1505), and if the operation of some of the local QoS managers 111 and 116 cannot be confirmed, a preset failure Processing (for example, restarting the local QoS management units 111 and 116, etc.) is performed (step 1530).

  The processing flow of the bandwidth monitoring unit 1460 of the local QoS management unit 111 is the same as that in FIG. 11, but global QoS information is received from the sub management node 1402 in step 1003, and the operating status of the sub management node 1402 is confirmed in step 1004. This is different from the first embodiment.

  As described above, a plurality of management nodes that manage state information of the entire information processing system 100 are provided, and when a failure occurs in the master management node 1401, the sub management node 1402 takes over the master management node 1401, thereby the master management node 1401. The contents of the global QoS management table 1443 can be prevented from being lost, and the reliability of the information processing system using the virtual I / O path can be improved.

  The present invention can be used for a computer system that uses a virtual I / O path by connecting a plurality of servers and I / O devices with an I / O switch such as a switch fabric. End-to-end performance can be guaranteed for each virtual I / O path generated between processing nodes.

It is a block diagram which shows an example of the hardware of the information processing system in the 1st Embodiment of this invention. It is a block diagram which shows an example of the functional element of the information processing system in the 1st Embodiment of this invention. It is a block diagram which shows an example of the local QoS management table in the 1st Embodiment of this invention. It is a block diagram which shows an example of the global QoS management table in the 1st Embodiment of this invention. It is a block diagram which shows an example of the logical I / O management table in the 1st Embodiment of this invention. It is a block diagram which shows an example of the QoS allocation management part in the 1st Embodiment of this invention. It is a block diagram which shows an example of the bandwidth monitoring part of the global QoS management part in the 1st Embodiment of this invention. It is a block diagram which shows an example of the zone | band monitoring part of the local QoS management part in the 1st Embodiment of this invention. It is a flowchart which shows an example of the virtual I / O path | pass production | generation process in the 1st Embodiment of this invention. It is a flowchart which shows an example of the global bandwidth monitoring process in the global QoS management part in the 1st Embodiment of this invention. It is a flowchart which shows an example of the local bandwidth monitoring process in the local QoS management part in the 1st Embodiment of this invention. It is a flowchart which shows an example of the logical I / O band calculation process in the local band monitoring process in the 1st Embodiment of this invention. It is a block diagram which shows an example of the virtual I / O path generation part of the QoS allocation management part in the modification 1 of this invention. It is a block diagram which shows an example of the information processing system in the modification 2 of this invention. It is a block diagram which shows an example of the information processing system in the modification 4 of this invention. It is a flowchart which shows an example of the global band monitoring process in the global QoS management part in the modification 4 of this invention.

Explanation of symbols

100 Information processing system 101, 102 Processing node 103 Management node 104 Management console 105 Switch fabric group 111, 116 Local QoS management unit 112, 117 Logical I / O management unit 122 Global QoS management unit 130, 133 Local QoS management table 131, 134 QoS allocation management unit 132, 135, 142 Bandwidth monitoring unit 140 Global QoS management table 150, 151, 152 Logical I / O management table 701 Logical I / O bandwidth calculation unit 707 Switch information acquisition unit

Claims (16)

A plurality of nodes having I / O adapters capable of setting virtual I / O ports as physical I / O ports; connecting the I / O adapters of the nodes to I / O switch ports; and the plurality of nodes A method of managing a virtual I / O path in an information processing system that transfers data via the virtual I / O port between
Any one of the plurality of nodes collecting information of a port of the I / O switch;
Each node obtaining a bandwidth for each virtual I / O port of an I / O adapter provided in each node;
Any one of the plurality of nodes acquires the information of the port of the I / O switch and the bandwidth;
Receiving a request for generating a virtual I / O path including information on a port of the I / O switch and the node that has acquired the bandwidth, including information on a source node, information on a destination node, and a bandwidth;
A node that has received the virtual I / O path generation request selects a path of the virtual I / O path from the source node to the destination node from the port information of the I / O switch;
The node that has received the virtual I / O path generation request extracts a virtual I / O port on the route of the selected virtual I / O path, and the acquired bandwidth and the extracted virtual I / O port Determining whether the virtual I / O path can be set from the bandwidth included in the generation request;
A step of setting the virtual I / O path to a port on the virtual I / O path when the node that has received the virtual I / O path generation request determines that the virtual I / O path can be set When,
A method of managing a virtual I / O path. The step of setting the virtual I / O path includes:
The node that has received the virtual I / O path generation request gives a virtual I / O path identifier to the virtual I / O port corresponding to the virtual I / O path set as the physical port of the I / O adapter;
Holding a virtual I / O path identifier associated with a physical port of an I / O adapter that has set the virtual I / O path, and status information of a virtual I / O port corresponding to the virtual I / O path identifier; The virtual I / O path management method according to claim 1, further comprising: The step of collecting port information of the I / O switch includes:
The virtual I / O according to claim 1, wherein information on the I / O switch, information on a port of the I / O switch, and information on a virtual I / O path set in the port are collected. Path management method. The step of setting the virtual I / O path includes:
A virtual I / O path identifier is assigned to the virtual I / O port corresponding to the virtual I / O path set in the physical port of the I / O adapter;
The step of acquiring a bandwidth for each virtual I / O port includes:
A virtual I / O path identifier associated with a physical port of the I / O adapter and a bandwidth transmitted / received by a virtual I / O port configured corresponding to the virtual I / O path identifier are acquired. The virtual I / O path management method according to claim 1. The step in which any one of the plurality of nodes collects information of the port of the I / O switch includes:
Selecting one node based on operating states of the plurality of nodes;
The step of acquiring a bandwidth for each virtual I / O port includes:
Obtaining a bandwidth of a port of the I / O switch in which the virtual I / O path is set,
The step of determining whether or not the virtual I / O path can be set includes:
The node that has received the virtual I / O path generation request extracts the port on the route of the selected virtual I / O path, the bandwidth acquired from the extracted port, the bandwidth of the virtual I / O port, and Determining whether the virtual I / O path can be set from the bandwidth included in the generation request;
2. The virtual I / O according to claim 1, wherein the selected node transmits the collected I / O switch port information to a node that accepts the virtual I / O path generation request. Path management method. A plurality of nodes having I / O adapters capable of setting virtual I / O ports as physical I / O ports, and an I / O switch having ports for connecting the I / O adapters of the respective nodes. An information processing system for transferring data between the plurality of nodes via the virtual I / O port,
The node is
A switch information acquisition unit that collects information of the ports of the I / O switch;
A local management unit that obtains a bandwidth for each virtual I / O port of the I / O adapter,
At least one of the plurality of nodes includes a global QoS management unit that sets a virtual I / O path that connects virtual I / O ports of the I / O adapters of the plurality of nodes,
The global QoS management unit
A local information acquisition unit that acquires information on the port of the I / O switch and a bandwidth for each virtual I / O port from each node from the switch information acquisition unit and the local management unit;
Accepts a request for generating a virtual I / O path including information on a source node, information on a destination node, and a bandwidth, from the port information of the I / O switch to the destination node from the source node A virtual I / O path selection unit for selecting a virtual I / O path route;
A virtual I / O port on the route of the selected virtual I / O path is extracted, and the virtual I / O path is extracted from the acquired bandwidth and the bandwidth included in the generation request in the extracted virtual I / O port. It is determined whether or not setting is possible, and when the virtual I / O path can be set, a virtual I / O path setting unit that sets the virtual I / O path to a port on the virtual I / O path;
An information processing system comprising: The virtual I / O path setting unit
A virtual I / O path identifier is assigned to a virtual I / O port corresponding to a virtual I / O path set in the physical port of the I / O adapter, and the physical of the I / O adapter in which the virtual I / O path is set The information processing system according to claim 6, wherein a virtual I / O path identifier associated with a port and status information of a virtual I / O port corresponding to the virtual I / O path identifier are retained. The switch information acquisition unit
The information processing system according to claim 6, wherein the information on the I / O switch, the information on the port of the I / O switch, and the information on the virtual I / O path set in the port are collected. The virtual I / O path setting unit
A virtual I / O path identifier is assigned to the virtual I / O port corresponding to the virtual I / O path set in the physical port of the I / O adapter;
The local management unit
A virtual I / O path identifier associated with a physical port of the I / O adapter and a bandwidth transmitted / received by a virtual I / O port configured corresponding to the virtual I / O path identifier are acquired. The information processing system according to claim 6. The local management unit
Obtain the bandwidth for each virtual I / O port of the I / O adapter and the bandwidth of the port of the I / O switch,
The virtual I / O path setting unit
The port on the route of the selected virtual I / O path is extracted as a related port, the virtual I / O port is acquired from the extracted related port, the virtual I / O port bandwidth, and the bandwidth included in the generation request. It is determined whether or not an / O path can be set, and when the virtual I / O path can be set, the virtual I / O path is set to a port on the virtual I / O path. Item 7. The information processing system according to Item 6. The virtual I / O path selection unit
The information processing system according to claim 6, wherein a virtual I / O path is selected from preset virtual I / O path candidates according to a preset condition. The virtual I / O path selection unit
A virtual set in advance as a selection condition using either the path length of the virtual I / O path, the bandwidth set for each port constituting the virtual I / O path or the bandwidth used for each port constituting the virtual I / O path The information processing system according to claim 11, wherein a virtual I / O path is selected from I / O path candidates. The global QoS management unit
A virtual I / O path deleting unit for deleting the set virtual I / O path;
The information processing system according to claim 6, further comprising: a virtual I / O path changing unit that changes the set virtual I / O path. The node is
A virtualization unit that divides computer resources into logical partitions;
A logical partition management unit for managing the logical partition;
The local management unit
A mapping unit that retains correspondence with identifiers of logical partitions managed by the logical partition management unit;
The virtual I / O path selection unit
The information processing system according to claim 6, wherein a virtual I / O path from the logical partition to another node is selected. The step of receiving a request for generating a virtual I / O path including information on the port of the I / O switch and the node that has acquired the bandwidth includes information on a source node, information on a destination node, and a bandwidth.
The virtual I / O path generation request includes the priority of the virtual I / O path to be newly set,
The step of setting the virtual I / O path to a port on the virtual I / O path includes:
2. The virtual I / O path management method according to claim 1, wherein the virtual I / O path is set to a port on the virtual I / O path based on the bandwidth and priority. A plurality of nodes having I / O adapters capable of setting virtual I / O ports as physical I / O ports; connecting the I / O adapters of the nodes to I / O switch ports; and the plurality of nodes A program for setting a virtual I / O path for transferring data to and from the virtual I / O port,
A procedure for acquiring port information of the I / O switch;
Obtaining a bandwidth for each virtual I / O port of the I / O adapter from each node;
Information on the port of the I / O switch, and a procedure for acquiring the bandwidth;
A procedure for receiving a virtual I / O path generation request including information on a source node, information on a destination node, and a bandwidth;
A procedure for selecting a route of a virtual I / O path from the source node to the destination node from the information of the port of the I / O switch;
The virtual I / O port on the route of the selected virtual I / O path is extracted, and the virtual I / O path is set to the extracted virtual I / O port from the acquired bandwidth and the bandwidth included in the generation request A procedure for determining whether it is possible;
Setting the virtual I / O path to a port on the virtual I / O path when it is determined that the virtual I / O path can be set;
A program characterized by causing a computer to function.

Images