WO2010064661A1 - 並列計算システム、その方法及びそのプログラム - Google Patents
並列計算システム、その方法及びそのプログラム Download PDFInfo
- Publication number
- WO2010064661A1 WO2010064661A1 PCT/JP2009/070251 JP2009070251W WO2010064661A1 WO 2010064661 A1 WO2010064661 A1 WO 2010064661A1 JP 2009070251 W JP2009070251 W JP 2009070251W WO 2010064661 A1 WO2010064661 A1 WO 2010064661A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- functional
- communication request
- communication
- input
- node
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
- G06F15/173—Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
- G06F15/17356—Indirect interconnection networks
- G06F15/17368—Indirect interconnection networks non hierarchical topologies
- G06F15/17375—One dimensional, e.g. linear array, ring
Definitions
- the present invention relates to a parallel computing system, and more particularly to a parallel computing system capable of assigning and managing hardware resources to a plurality of applications and processes, a method thereof, and a program thereof.
- Non-Patent Document 1 An example of a parallel processing system related to the present invention is described in Non-Patent Document 1 and Non-Patent Document 2.
- the parallel computing system 2500 is configured by a combination of hardware and software.
- a plurality of functional nodes 2400 that perform functions such as computation, IO (input / output), and storage are provided as hardware.
- IO input / output
- storage storage
- a communication path for communication between these function nodes 2400 is provided.
- a software it has a virtualization layer 2300, a virtual machine 2200, and a process 2100 on the virtual machine.
- the parallel computing system 2500 related to the present invention having such a configuration operates as follows.
- the virtualization layer 2300 allocates a function node 2400 to each virtual machine 2200 dynamically or statically.
- the virtual machine 2200 can operate as a virtual parallel computing system 2500 composed of independent function nodes.
- Each virtual machine 2200 can execute the process 2100 on the virtual parallel machine 2200 assigned to the virtual machine 2200 without sensing the assignment of the physical function node 2400 and the execution state of the other virtual machines 2200. It is.
- the process 2100 operating on the virtual machine 2200 is a normal OS (Operating System) or an application itself.
- the functional nodes 2400 constituting the physical hardware perform processing while communicating with each other via a communication path.
- Patent Document 1 Another example of a parallel computing system having a plurality of function nodes is described in Patent Document 1.
- an object is configured by software, and functions such as calculation, IO (input / output), and storage can be realized in the object.
- each virtual machine 2200 is assigned with a physical function node 2400 or another virtual machine 2200. It becomes possible to operate without sensing the execution state of.
- the parallel processing system 2500 described in Non-Patent Document 1 and Non-Patent Document 2 has the following problems.
- the first problem is that the processing overhead is large.
- the reason is that a software layer called a virtualization layer and a virtual machine exists between physical hardware and a process for performing actual processing.
- the second problem is that the communication bandwidth between the function nodes allocated to each virtual machine and the process running on it is not guaranteed.
- the reason is that there is no function for guaranteeing the communication bandwidth for each virtual machine or between functional nodes in the virtual machine in a communication path that mediates communication between physical functional nodes.
- the third problem is that the separation between the virtual machine and the processes running on it is not perfect. This is because the mapping of each virtual machine to physical hardware is performed by software called a virtualization layer.
- an object of the present invention is to provide a parallel computing system, a method thereof, and a program thereof that can execute a plurality of processes with a small software layer overhead on one parallel computing system.
- Another object of the present invention is to provide a parallel computing system, a method thereof, and a program thereof that can guarantee a communication bandwidth between functional nodes assigned to a process.
- Another object of the present invention is to provide a parallel computing system, a method thereof, and a program thereof that can realize separation between a plurality of processes by hardware.
- communication is performed between a plurality of functional nodes having any one of an arithmetic function, an input / output function, and a storage function or a combination thereof for performing information processing.
- the functional node performs an arithmetic function, an input / output function, and a storage function for performing information processing.
- a function node input / output port which is a port for inputting / outputting a communication request transferred between the function nodes in order to perform communication between the function elements and any one or a combination of the function elements.
- a functional node group composed of a part or all of the functional nodes of the plurality of functional nodes included in the parallel computing system as one group
- a plurality of input / output ports which are ports for inputting and outputting the communication request by the network switch, and the input / output port receives the communication request input from the input / output port.
- a parallel computing system comprising: a switch for outputting to a network; and bandwidth management means for managing a communication bandwidth for each communication path between the functional nodes and controlling the switch.
- communication is performed between a plurality of functional nodes having any one of an arithmetic function, an input / output function, and a storage function or a combination thereof for performing information processing.
- a management method in a parallel computing system having a plurality of network switches functioning as a communication path that is a part of a communication path for performing processing, wherein the functional node performs an information processing, an input / output function A step of preparing a functional element having any one of a function and a storage function or a combination thereof; and a communication request transferred between the functional nodes so that the functional nodes communicate with each other between the functional nodes.
- a functional node group management step for managing a functional node group composed of a part or all of the functional nodes as one group, and a plurality of ports that are ports through which the network switch inputs and outputs the communication request.
- communication is performed between a plurality of functional nodes having a function of any of an arithmetic function, an input / output function and a storage function or a combination thereof for performing information processing.
- the function node performs an arithmetic function, an input / output function
- a functional node which is a port for inputting / outputting a communication element transferred between the functional nodes in order to perform communication between the functional elements and a functional element having any of the storage functions or a combination thereof
- a functional node group composed of an input / output port and a part or all of the functional nodes of the plurality of functional nodes of the parallel computing system,
- Functional node group management means for managing as a group, a plurality of input / output ports that are ports for the network switch to input / output the communication request, and the communication request input from the input / output
- the functional node group management unit can directly communicate between functional nodes without the intervention of the OS or the virtualization layer, it is possible to reduce the overhead of communication between functional nodes constituting the process. .
- FIG. 1 shows a first embodiment of the present invention.
- the parallel computing system 1000 according to the present embodiment includes a plurality of functional nodes 300 and a communication path 500 that is a communication path for performing communication between the functional nodes 300.
- the function node 300 is a node that provides an arithmetic function, an input / output function, a storage function, and the like.
- the functional node 300 includes a functional element 310 that provides an arithmetic function, an input / output function, a storage function, and the like, and a functional node group management unit 320 that manages a plurality of functional node groups as one unit.
- the communication path 500 includes a plurality of network switches 400.
- the network switch 400 includes a switch 420 that functions as a communication path between the functional nodes 300 and a bandwidth management unit 410 that manages the bandwidth of communication between the functional nodes 300.
- a plurality of processes 100 are executed.
- the process 100 may be a normal OS or an application itself.
- Each of the processes 100 is assigned a “functional node group” composed of a plurality of functional nodes 300 and a “bandwidth” of communication between the functional nodes 300.
- the process 200 and the function node group correspondence 200 are the process 100 and the function node group assigned to the process 100. This association is performed by the functional node group management unit 320 included in the functional node 300 and the bandwidth management unit 410 included in the network switch 400.
- FIG. 2 is an example of a process 100 and a group of functional nodes assigned to configure the process 100.
- the functional node group includes a plurality of functional nodes 300 and a communication path 510 for performing communication between the functional nodes 300.
- n1 to n9 are function nodes. Specifically, n1 and n6 are input / output nodes, n2, n3, n4 and n5 are operation nodes, and n7, n8 and n9 are storage nodes.
- the configuration of the communication path 510 differs depending on the processing performed in the process 100.
- the process 100 is a general-purpose process such as an OS
- a complete coupling may be required between the functional nodes 300 in some cases.
- the process 100 is a process for performing a specific process, it is not necessary to be completely coupled, and generally a communication path 510 that is far fewer than fully coupled is sufficient.
- This embodiment can cope with a case where the communication path between the functional nodes 300 requires a complete connection, but is particularly effective when a specific communication path 510 required for the process 100 is requested.
- the functional node 300 constituting the process 100 is identified by a “local node ID” that is valid only within the process 100.
- the communication path 510 is identified by a “local communication path ID” that is valid only within the process 100.
- n1 to n9 are local node IDs
- lnk1 to lnk19 are local communication path IDs.
- the functional node 300 has a “global node ID” in addition to the above-mentioned local node ID.
- the global node ID is an ID for identifying each functional node 300 in the parallel computing system 1000.
- FIG. 1 shows the global node IDs N1 to Nn.
- the global communication path ID is an ID for identifying a communication path between the functional nodes 300.
- the global communication path ID is not an ID corresponding to the individual network switch 400.
- a communication path that communicates from a certain transfer source function node 300 to a certain transfer destination function node 300 can be designated as a connection state of a plurality of network switches 400 constituting the communication path and the network switch 400.
- the global communication path ID is an ID used for identifying this connection state.
- the association 200 between the process and the functional node group is realized by managing the association between the local node ID of each functional node 300 constituting the process 100 and the global node ID.
- the functional element 310 of the transfer source functional node 300 When communicating between the functional nodes 300, first, the functional element 310 of the transfer source functional node 300 creates a local communication request 600 using "local identification information" such as a local node ID and a local communication path ID. Then, the created local communication request 600 is sent to the functional node group management unit 320 as a communication request to the transfer destination functional node 300.
- local identification information such as a local node ID and a local communication path ID.
- the functional node group management unit 320 creates the global communication request 700 by converting the local identification information of the received local communication request 600 into “global identification information” such as a global node ID and a global communication path ID.
- the functional node group management unit 320 sends the created global communication request 700 to the network switch 400 that configures the communication path 500.
- the network switch 400 controls the switch 420 based on the global identification information included in the global communication request 700, whereby the global communication request 700 is transferred to the transfer destination functional node 300.
- bandwidth management unit 410 of the network switch 400 a communication path and a bandwidth allocated to the communication path are registered in advance. For example, bandwidth information to be assigned to the communication path corresponding to the transfer source global node ID and the transfer destination global node ID is registered in the bandwidth management unit 410. As the registration information, it is also possible to register bandwidth information to be allocated to the communication path corresponding to the global communication path ID.
- the network switch 400 controls the switch 420 based on the global identification information included in the global communication request 700. At this time, the network switch 400 is necessary according to the bandwidth information registered in advance in the bandwidth management unit 410 of the network switch 400. By allocating the bandwidth, the bandwidth requested by the communication request is guaranteed.
- FIG. 3 shows an example of the parallel computing system 1000 of this embodiment in the case of having a communication path 500 having a two-dimensional mesh configuration.
- the functional nodes 300 are in a two-dimensional array, and one network switch 400 is connected to each functional node 300.
- Each network switch 400 is connected to the other four network switches 400 that are adjacent vertically and horizontally on the two-dimensional array.
- the function node 300 provides functions such as calculation, input / output, and storage.
- the function node 300 may include a memory node 300-A with an MMU (Memory Management Unit) for sharing the memory function among a plurality of processes 100.
- MMU Memory Management Unit
- a power supply fv control node 300-B for controlling the power supply, operating frequency, and operating voltage of the network switch 400 may be included.
- FIG. 4 is a diagram illustrating a specific example of a data structure of a communication request and identification information for performing communication between the functional nodes 300.
- FIG. 4 a is a configuration example of the global communication request 700 and includes global identification information 710 and a payload 740.
- Payload 740 is data transferred by communication.
- the global communication request 700 may further include a function node group ID 720 and privilege control information 730.
- the function node group ID 720 is an ID for identifying the function node group.
- the privilege control information 730 is information for performing privileged control and other communication control.
- the privilege control information 730 can also hold information for controlling the transfer destination of the global communication request 700.
- the global communication request 700 can be transferred not only to the functional node 300 but also to the network switch 400 using the privilege control information 730.
- the function nodes 300 and the network switches 400 correspond one-to-one. Therefore, when the global communication request 700 is transferred to the network switch 400 corresponding to the functional node 300, the same global identification information 710 is set in the global communication request 700.
- the privilege control information 730 with a flag for identifying either the functional node 300 or the network switch 400 as a transfer destination, it becomes possible to transfer the data in a distinguished manner.
- FIG. 4 b shows a configuration example of the local communication request 600 and has local identification information 610 and a payload 740. Similar to the global communication request 700, the local communication request 600 can also be configured to include a functional node group ID 720 and privilege control information 730.
- FIG. 4 c 1 is a configuration example of the global identification information 710 and includes a transfer destination global node ID 810 and a transfer source global node ID 820.
- FIG. 4c2 is a configuration example of the local identification information 610, and includes a transfer destination local node ID 910 and a transfer source local node ID 920.
- FIG. 4d1 is another configuration example of the global identification information 710, and has a global communication path ID 830.
- FIG. 4 d 2 is another configuration example of the local identification information 610 and has a local communication path ID 930.
- FIG. 5 is a diagram illustrating a configuration example of the function node 300.
- the functional node 300 includes a functional element 310, a functional node group management unit 320, a power supply fv control unit 350, a privilege level storage unit 360, a functional node input port 370, and a functional node output port 380.
- the functional node group management unit 320 includes a communication request conversion unit 330 and a functional node group management information holding unit 340.
- the functional element 310 provides any of a calculation function, an input / output function, and a storage function. Further, the functional element 310 may provide other functions instead of these functions or together with these functions.
- the functional node group management unit 320 functions as an interface unit with the network switch 400.
- the functional node 300 has a functional node input port 370 and a functional node output port 380 in order to exchange a global communication request 700 with the network switch 400.
- the functional node group management information holding unit 340 holds information necessary for mutually converting the local communication request 600 and the global communication request 700.
- a correspondence table between the local node ID and the global node ID of the transfer destination functional node 300 when the functional node 300 performs communication as the transfer source functional node 300 is mentioned.
- the functional element 310 generates a local communication request 600 using the local identification information 610 illustrated in FIG. 4c2 (step S11).
- the communication request conversion unit 330 converts the transfer destination local node ID 910 constituting the local identification information 610 of the local communication request 600 into a global node ID 810 with reference to the functional node group management information holding unit 340. Similarly, the function node group management information holding unit 340 is referred to, and the transfer source local node ID 920 is converted into the transfer source global node ID 820. Then, the global identification information 710 (see FIG. 4c1) is set using these converted IDs (step S13).
- the communication request conversion unit 330 transfers the communication request to the transfer destination functional node 300 by sending the global communication request 700 including the global identification information 710 of FIG. 4c1 from the functional node output port 380 to the network switch 400 (step S15). ).
- the global communication request 700 transferred from the network switch 400 to the functional node 300 via the functional node input port 370 is converted into a local communication request 600 by the communication request conversion unit 330 and the functional node group management information holding unit 340, (Step S17).
- the communication request conversion unit 330 extracts the payload 740 from the global communication request 700 and transfers only the payload 740 to the functional element 310. You can also In this case, it is not necessary to refer to the functional node group management information holding unit 340, and the communication request conversion unit 330 and the functional node group management information holding unit 340 can be simplified.
- the global identification information 710 included in the global communication request 700 and the local identification information 610 of the local communication request 600 are configured using the global communication path ID 830 (see FIG. 4d1) and the local communication path ID 930 (see FIG. 4d2), respectively.
- the local communication request 600 and the global communication request 700 can be converted into each other in the same procedure.
- the functional node group configuration can be changed by rewriting and updating the functional node group management information holding unit 340.
- the functional node group management information holding unit 340 is rewritten by a global communication request 700 from another functional node 300.
- the privilege control information 730 included in the global communication request 700 and the payload 740 are used to instruct rewriting and rewriting contents of the functional node group management information holding unit 340.
- the function node group management information holding unit 340 is erroneously rewritten, the process 100 being executed in another function node group is affected. Therefore, it is preferable that only the privileged global communication request 700 can be rewritten so as not to be rewritten accidentally or rewritten by the malicious process 100.
- the privilege control information 730 is added to the global communication request 700, and the privilege control information 730 is used to identify whether or not the communication request can request a privileged operation.
- the privilege level storage unit 360 of the functional node 300 holds the privilege level of the functional node 300.
- the communication request conversion unit 330 controls transmission of a global communication request 700 that instructs a privileged operation.
- the functional element 310 of the functional node 300 having a low privilege level (non-privileged level) sends a local communication request 600 instructing a privileged operation
- the communication request conversion unit 330 transmits the global communication request 700 of the local communication request 600. And conversion to the network switch 400 is blocked.
- Rewriting of the privilege level held in the privilege level storage unit 360 is also performed by the privileged global communication request 700. On the other hand, rewriting by the global communication request 700 having a low privilege level (or a non-privileged level) is blocked.
- the power fv control unit 350 of the functional node 300 performs power ON / OFF switching of the functional node 300, and control of the operating frequency and operating voltage.
- the power supply fv control of the functional node 300 can be designated as the global communication request 700. Whether to perform power ON / OFF switching, operation frequency control, or operation voltage control can be specified using the privilege control information 730 and the payload 740 constituting the global communication request 700.
- the network switch 400 includes a switch 410, a bandwidth management unit 420, a switch control unit 430, a bandwidth management information holding unit 440, and a network switch power supply fv control unit 450.
- the network switch 400 has a connection with other network switches 400 that are adjacent vertically and horizontally.
- the network switch 400 is connected to one functional node 300.
- the network switch 400 has input / output ports corresponding to these connections. Therefore, the network switch 400 includes four network switches 400 that are vertically and horizontally adjacent to each other and five input ports for input from one adjacent functional node 300. In FIG. 6 and the following description, these five input ports are denoted as 460U, 460D, 460L, 460R, 460N.
- four network switches 400 that are vertically and horizontally adjacent to each other and five output ports for output to one adjacent functional node 300 are provided. In FIG. 6 and the following description, these five ports are represented as 470U, 470D, 470L, 470R, and 470N.
- the global communication request 700 is input to a single, plural or all of the five input ports 460U, 460D, 460L, 460R, 460N (step S21).
- the switch controller 430 of the bandwidth 420 determines from which input port 460 to which output port 470 the global communication request 700 is transferred (step S23).
- the switch control unit 430 controls the switch 410, the global communication request 700 from the five input ports 460U, 460D, 460L, 460R, and 460N is transmitted to the five output ports 470U via the 5 ⁇ 5 switch 420.
- 470D, 470L, 470R, and 470N are transferred and output (step S25).
- the transfer destination is not limited to one, and there may be a plurality of transfer destinations. For example, if the same communication request is transferred to the entire node group, that is, all nodes belonging to the node group, or if the same communication request is transferred to some of the nodes belonging to the node group, there are a plurality of transfer destinations. It becomes.
- the switch 410 is controlled by the switch control unit 430 of the bandwidth management unit 420 to which input port 460 to which output port 470 the global communication request 700 is transferred.
- Each global communication request 700 defines a required transfer bandwidth.
- the bandwidth management information holding unit 440 stores information for controlling the switch 410 with respect to the plurality of communication paths 510 passing through the network switch 5.
- the output port 470 to be transferred for each communication path 510 identified by the global identification information 710 included in the global communication request 700 is associated with the requested bandwidth.
- the attached table can be illustrated.
- the switch control unit 430 controls the switch 410 based on the bandwidth management information holding unit 440.
- the bandwidth management information holding unit 440 can be rewritten by an instruction of the global communication request 700 sent with the network switch 400 as a transfer destination.
- FIG. 7 is a configuration example of the 5 ⁇ 5 switch 410.
- the transfer route having the network switch 400 itself as the transfer destination is omitted.
- the transfer port number having the network switch 400 itself as the transfer destination is changed from 5 to 6. It can be easily expanded by increasing the number.
- the transfer operation of the switch 410 is designated by a path control signal 414 and a bandwidth control signal 415 that constitute a control signal from the switch control unit 430 shown in FIG.
- a multiplexer (denoted as “MUX” in the drawing and in the following description) MUX 411 performs path control for appropriately transferring the global communication request 700 received from the input port 460 to the output port 470.
- the bandwidth control unit 413 performs control for performing transfer that guarantees the bandwidth requested by the global communication request 700.
- a buffer (buffer: expressed as “BUF” in the figure and the following description) is a BUF for storing a global communication request 700 between the MUX 411 and the bandwidth control unit 413.
- the MUX 411 is set in advance with an appropriate output destination for each communication path 510 based on the path control signal 414.
- the MUX 411 identifies the communication path 510 from the global identification information 710 included in the global communication request 700, and transfers the global communication request 700 to the preset output destination BUF 412.
- FIG. 8 is a diagram illustrating an example of bandwidth control.
- the output port 470 is shared by four communication paths 510L1, 510L2, 510L3, and 510L4. If the bandwidth of the output port 470 is B, and the bandwidths required by the four communication paths 510L1, 5102, 5103, and 5104 are B1, B2, B3, and B4, respectively,
- the relationship needs to hold.
- the path from the BUF 412U to the output port 470 is shared by the two communication paths 510L1 and 510L2, and the bandwidth required for this path is (B1 + B2).
- the communication path 510 is not set as the path from the BUF 412D and the BUF 412R to the output port 470.
- the communication path 510L3 is set as the path from the BUF 412L to the output port 470.
- the communication path 510L4 is set as the path from the BUF 412N to the output port 470.
- the bandwidth management unit 413 allocates B1 + B2, 0, B3, 0, and B4 as bandwidths to each of the four BUF412U, BUF412D, BUF412L, BUF412R, and BUF412N.
- the BUF 412U is shared by the two communication paths 510L1 and 510L2, but if each source function node 300 sends out a global communication request 700 within a preset bandwidth range, the respective communication paths The bandwidth (B1 + B2) for the BUF 412U may be guaranteed without distinguishing between them.
- the bandwidth control unit 413 confirms whether the global communication request 700 exists in each BUF 412 by, for example, round robin. If there is a global communication request 700, one global communication request 700 is taken out from the BUF 412 and transferred to the output port 470. Since the bandwidth allocated to each BUF 412 is different, communication that guarantees the bandwidth required by each communication path 510 can be realized by performing round robin at a frequency proportional to the allocated bandwidth. In the example of FIG. 8, for five BUF412U, BUF412D, BUF412L, BUF412R, and BUF412N
- the round robin may be performed at a frequency proportional to.
- the frequency of round robin is set by the bandwidth control signal 415.
- the maximum length of the global communication request 700 may be set, and round robin may be performed so as to guarantee the communication of the maximum length.
- round robin may be performed so as to guarantee the communication of the maximum length.
- the network switch 400 in FIG. 6 further includes a network switch power fv control unit 450.
- the network switch power supply fv control unit 450 controls the power supply ON / OFF switching, the operating voltage, and the operating frequency according to the instruction of the global communication request 700 sent with the network switch 400 as a transfer destination.
- FIG. 9 is a configuration example of the storage node 300-A with MMU.
- the function node group management unit 320, the communication request conversion unit 330, the function node group management information holding unit 340, the power supply fv control unit 350, and the privilege level storage unit 360 are the same as the function node 300 in FIG. is there.
- the storage node 300-A with MMU has a functional node group ID extraction unit 390.
- the functional element 310 includes a storage element 310-A.
- the storage element 310-A includes a storage element 311-A, an MMU 312-A, a storage access control unit 313-A, and storage management information 314-A.
- the functional node group ID extraction unit 390 extracts the functional node group ID from the global communication request 700.
- the storage element 3101 functions as a storage element with an MMU.
- the global communication request 700 received from the function node input port 370 is converted into a local communication request 600 and transferred to the storage access control unit 313-A.
- the payload 740 of the local communication request 600 is an instruction to access the storage element.
- the payload 740 of the local communication request 600 has an access type that indicates whether the access is write or read, and a logical address. When the access type is “write”, it has write-in data. On the other hand, when the access type is read, it has the local node ID of the functional node 300 to which the read data is to be transferred.
- the storage access control unit 313-A extracts the access type and logical address from the local communication request 600. Here, if the access type is writing, data is also extracted. On the other hand, if the access type is read, the local node ID of the functional node 300 to which the read data is to be transferred is also extracted. The storage access control unit 313-A performs a read / write operation on the storage element based on the extracted information.
- the storage management information 314-A is information necessary for conversion of the functional node group accessing the storage node 300-A with MMU, a logical address unique to each functional node group, and a physical address of the storage element 311-A. Hold.
- the MMU 312 -A performs conversion from a logical address to a physical address.
- An access request for the storage node 300-A with MMU is transferred as a global communication request 700 from a plurality of functional node groups.
- the MMU 312 -A converts the logical address into a physical address based on the functional node group ID extracted by the functional node group ID extraction unit 390 and the storage management information 314 -A.
- the global communication request 700 received from the functional node input port 370 is converted into a local communication request 600 by the functional node management unit 320 and transferred to the storage access control unit 313-A.
- the storage access control unit 313-A performs access control on the storage element 311-A according to the access type.
- the logical address is converted into a physical address by the MMU 312 -A and used for addressing the storage element 311 -A.
- the storage access control unit 313-A writes the extracted data to the storage element 311-A when the access type is write.
- the access type is read, the data read from the storage element 311 -A, the local node ID of the functional node 300 to which the data is to be transferred, and the functional node group extracted by the functional node group ID extracting unit 390
- a local communication request 600 is generated from the ID. Then, the generated local communication request 600 is transferred to the communication request conversion unit 330.
- the communication request conversion unit 330 converts the local communication request 600 into a global communication request 700 and sends it to the function node output port 380.
- the storage management information 314-A can be rewritten by a privileged global communication request 700 from the function node input port 370.
- FIG. 10 is a configuration example of the power supply fv control node 300-B.
- the power supply fv control node 300-B controls ON / OFF switching of the power supply of the network switch 400, the operating voltage, and the operating frequency via the dedicated communication path.
- the point that the power supply fv control node 300-B includes a functional node group management unit 320, a communication request conversion unit 330, a functional node group management information holding unit 340, a power supply fv control unit 350, and a privilege level storage unit 360. This is the same as the function node 300 of FIG.
- the power supply fv control node 300-B has a power supply fv control input / output element 310-B as the functional element 310.
- the power supply fv control input / output element 310-B is connected to the power supply fv control unit 450 of the network switch 400 according to the instruction of the payload 740 included in the global communication request 700 from the function node input port 370.
- the power supply fv control of the network switch 400 is performed via the path.
- the first effect is that the overhead of communication between function nodes constituting the process can be reduced. This is because the functional node group management unit can directly communicate between functional nodes without the intervention of the OS or the virtualization layer.
- the second effect is that the communication bandwidth between functional nodes assigned to the process can be guaranteed.
- the reason is that communication control according to the bandwidth of communication between the function nodes set in advance by the bandwidth management unit of the network switch can be performed.
- the third effect is that a plurality of processes can be executed on one parallel processing system without interfering with each other.
- the reason is that the function node group management unit, the function node assigned to each process by the bandwidth management unit, and the communication bandwidth between the function nodes can be operated completely separated from other processes.
- the fourth effect is that power consumption can be reduced.
- the reason is that each functional node 300 is configured to have the power supply fv control unit 350. Therefore, the power consumption of the functional node 300 not assigned to the process is turned off or the clock is stopped. This is because reduction can be realized. When there is a margin in processing performance of the functional node 300, power consumption can be reduced by lowering the clock frequency or lowering the power supply voltage.
- each network switch 400 is configured to have a network switch power supply fv control unit 450, power consumption can be reduced by turning off the power of the network switch 400 to which the communication path 510 is not assigned or stopping the clock. Reduction is possible.
- the bandwidth of the network switch 400 has a margin with respect to the required bandwidth of the communication path 510 assigned to the network switch 400, the power consumption can be reduced by lowering the clock frequency or reducing the power supply voltage. Is possible.
- the fifth effect is that setting of management information can be performed only by the process 100 having a high privilege level.
- the global communication request 700 and the local communication request 600 can be configured to have the privilege control information 730, so that the function node group management information holding unit 340, the privilege level storage unit 360, the power supply fv control This is because the operation of management information such as the unit 350, the network switch power fv control unit 450, and the storage management information 314-A can be set so as to be performed only by the process 100 having a high privilege level.
- an OS process for allocating resources such as the function node 300 and the communication path 510 to other processes 100 can be configured.
- the sixth effect is that communication between different processes 100 is also possible.
- the reason is that a storage node 300-A with an MMU can be provided as a storage node, and the storage node 300-A with an MMU has a function of a shared memory in which a plurality of processes 100 share the same storage element 311-A. This is because it can be realized safely.
- parallel computing system can be realized by hardware, software, or a combination thereof.
- the parallel computing system described above can be realized by hardware, but can also be realized by a computer reading a program for causing the computer to function as the system from a recording medium and executing the program.
- the above parallel calculation method can also be realized by hardware.
- a program for causing a computer to execute the method can be read from a computer-readable recording medium and executed. Can be realized.
- the hardware and software configurations described above are not particularly limited, and any configuration can be applied as long as the functions of the respective units described above can be realized.
- it may be configured individually for each function of each unit described above, or may be configured integrally with the function of each unit.
- the present invention can be applied to applications such as embedded control in which real-time performance and low power performance are important for multi-core systems. Also, the present invention can be applied to a case where a plurality of processes are optimized for a plurality of cores for each process by an automatic parallelizing compiler and executed without interfering with each other on one parallel computing system.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multi Processors (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
Description
200 プロセスと機能ノード群の対応付け仮想マシン 2200
300、2400 機能ノード
300-A MMU付き記憶ノード
300-B 電源fv制御ノード
310 機能要素
310-A 記憶要素
311-A 記憶素子
312-A MMU
313-A 記憶アクセス制御部
314-A 記憶管理情報 216 電源fv制御用入出力要素
320 機能ノード群管理部
330 通信要求変換部
340 機能ノード群管理情報保持部
350 電源fv制御部
360 特権レベル記憶部
370 機能ノード入力ポート
380 機能ノード出力ポート
390 機能ノード群ID抽出部400 ネットワークスイッチ
410 スイッチ
411 MUX
412、412U、412D、412L、412R、412N BUF
413 帯域幅制御部
414 経路制御信号
415 帯域幅制御信号
420 帯域幅管理部
430 スイッチ制御部
440 帯域幅管理情報保持部
450 ネットワークスイッチ電源fv制御部
460、460U、460D、460L、460R、460N 入力ポート
470、470U、470D、470L、470R、470N 出力ポート
500 通信路
510、510L1、510L2、510L3、510L4 通信経路
600 ローカル通信要求
610 ローカル識別情報
700 グローバル通信要求
710 グローバル識別情報
720 機能ノード群ID
730 特権制御情報
740 ペイロード
810 転送先グローバルノードID
820 転送元グローバルノードID
830 グローバル通信経路ID
910 転送先ローカルノードID
920 転送元ローカルノードID
930 ローカル通信経路ID
1000、2500 並列計算システム
2300 仮想化層
Claims (24)
- 情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する複数の機能ノードと、前記機能ノード間で通信を行うための通信路の一部であり通信経路として機能する複数のネットワークスイッチとを有する並列計算システムにおいて、
前記機能ノードが、
情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する機能要素と、
前記機能ノード間で相互に通信を行うために前記機能ノード間で転送される通信要求を入出力するためのポートである機能ノード入出力ポートと、
前記並列計算システムが有する複数の前記機能ノードの一部又は全部の前記機能ノードから構成される機能ノード群を、1つの群として管理する機能ノード群管理手段と、
を備え、
前記ネットワークスイッチが、
前記通信要求を入出力するためのポートである複数の入出力ポートと、
前記入出力ポートから入力された前記通信要求を前記入出力ポートへ出力するスイッチと、
前記機能ノード間の前記通信経路毎に通信帯域幅を管理し前記スイッチを制御する帯域幅管理手段と、
を備えていることを特徴とする並列計算システム。 - 前記機能ノード群を構成する複数の前記機能ノード間の通信においては、前記機能ノード群を構成する複数の前記機能ノード間のみで相互に通信を行うための通信要求であるローカル通信要求を用い、
前記通信路における通信では、前記並列計算システムを構成する全ての前記機能ノード間で相互に通信を行うためのグローバル通信要求を用い、
前記機能ノード群管理手段が、
前記ローカル通信要求と前記グローバル通信要求の相互の変換のための情報を管理する機能ノード管理情報保持手段と、
前記機能ノード管理情報保持手段が管理する前記情報を用いて、前記機能要素が入出力する前記ローカル通信要求と前記機能ノード入出力ポートから入出力される前記グローバル通信要求とを相互に変換する通信要求変換手段と、
を備えることを特徴とする請求項1に記載の並列計算システム。 - 前記機能ノードが、前記機能ノードの電源の投入及び遮断の制御、前記機能ノードの動作周波数の制御、前記機能ノードの動作電圧の制御の何れか又はこれらを組み合わせた制御を行う電源fv制御手段を、更に備えることを特徴とする請求項2に記載の並列計算システム。
- 前記ローカル通信要求及び前記グローバル通信要求は、当該ローカル通信要求又は当該前記グローバル通信要求が特権的な通信要求であるか否かを示す情報である特権制御情報を有し、
前記機能ノードが、前記機能ノードが前記機能ノード入出力ポートから送出できる前記グローバル通信要求の前記特権制御情報の特権レベルの上限値を保持する特権レベル記憶手段を、
更に備えることを特徴とする請求項2又は3に記載の並列計算システム。 - 前記帯域幅管理手段が、
前記ネットワークスイッチを前記通信経路の一部として含む前記通信経路と前記通信経路に割り当てられている通信帯域幅との対応付けを保持する帯域幅管理情報保持手段と、
前記帯域幅管理情報に従って前記ネットワークスイッチの前記スイッチを制御するスイッチ制御手段と、
を備えることを特徴とする請求項2乃至4の何れか1項に記載の並列計算システム。 - 前記ネットワークスイッチは、
前記ネットワークスイッチの前記入出力ポートに対応する複数の入力ポート及び複数の出力ポートと、
複数の前記入力ポート毎に前記通信経路を制御し、前記複数の出力ポートに対応した出力を持つマルチプレクサと、
前記マルチプレクサの前記出力毎にグローバル通信要求を蓄積するためのバッファと、
複数の前記マルチプレク毎の前記出力ポートを出力先とする複数の前記バッファに蓄積されている前記グローバル通信要求を前記出力ポートに出力する帯域幅制御手段と、
を備えることを特徴とする請求項2乃至5の何れか1項に記載の並列計算システム。 - 前記帯域幅制御手段は、当該帯域幅制御手段が前記グローバル通信要求を受け取る複数の前記バッファに対する処理を、前記通信経路毎に割り当てられている通信帯域幅に比例した頻度でラウンドロビン方式によって行うことを特徴とする請求項6に記載の並列計算システム。
- 前記ネットワークスイッチが、前記ネットワークスイッチの電源の投入及び遮断の制御、前記ネットワークスイッチの動作周波数の制御、前記ネットワークスイッチの動作電圧の制御の何れか又はこれらを組み合わせた制御を行うネットワークスイッチ電源fv制御手段を、
更に備えることを特徴とする請求項2乃至7の何れか1項に記載の並列計算システム。 - 前記機能ノードとして、前記通信路を構成する複数の前記ネットワークスイッチ電源fv制御手段を専用通信路を介して制御する電源fv制御ノードを備えることを特徴とする請求項8に記載の並列計算システム。
- 前記機能ノードとして複数の前記機能ノード群から共通にアクセスすることができ、記憶機能を共有することができるMMU(Memory Management Unit)付き記憶ノードを持つことを特徴とする請求項2乃至9の何れか1項に記載の並列計算システム。
- 前記MMU付き記憶ノードは、
前記機能ノード群管理手段と、前記特権レベル記憶手段と、前記電源fv制御手段と、前記機能ノード入出力ポートと、に加え、
前記機能ノード入出力ポートから入力された前記グローバル通信要求から機能ノード群IDを抽出するための機能ノード群ID抽出手段と、
物理アドレスと読み書き制御信号によってデータを読み書きできる記憶素子と、
前記機能ノード群管理手段によって前記グローバル通信要求から変換された前記ローカル通信要求から前記記憶要素に対するアクセス制御情報を抽出し、抽出された前記アクセス制御情報の論理アドレスによって前記記憶素子に対してデータの読み書き制御を行う記憶アクセス制御手段と、
前記機能ノード群毎に前記論理アドレスと前記記憶素子にアクセスするための物理アドレスの変換情報を保持する記憶管理情報保持手段と、
前記機能ノード群ID抽出手段で抽出された前記機能ノード群IDと前記記憶管理情報を用いて前記記憶アクセス手段が出力する論理アドレスを物理アドレスに変換して前記記憶素子に入力するMMUと、を備え、
前記記憶アクセス制御手段は、前記ローカル通信要求から抽出された前記アクセス制御情報のアクセスタイプが読み出しの場合、前記記憶素子から読み出されたデータと前記機能ノード群IDとから前記読み出されたデータを転送すべき前記機能ノードへの前記ローカル通信要求を構成し前記機能ノード群管理手段に転送し、
前記機能ノード管理手段は、前記機能アクセス制御手段から転送された前記ローカル通信要求を前記グローバル通信要求に変換して前記機能ノード入出力ポートから送出することを特徴とする請求項10に記載の並列計算システム。 - 情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する複数の機能ノードと、前記機能ノード間で通信を行うための通信路の一部であり通信経路として機能する複数のネットワークスイッチとを有する並列計算システムにおける管理方法であって、
前記機能ノードが、情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する機能要素を用意するステップと、
前記機能ノードが、前記機能ノード間で相互に通信を行うために前記機能ノード間で転送される通信要求を入出力するためのポートである機能ノード入出力ポートを用意するステップと、
前記機能ノードが、前記並列計算方法が有する複数の前記機能ノードの一部又は全部の前記機能ノードから構成される機能ノード群を、1つの群として管理する機能ノード群管理ステップと、
前記ネットワークスイッチが、前記通信要求を入出力するためのポートである複数の入出力ポートを用意するステップと、
前記ネットワークスイッチが、前記入出力ポートから入力された前記通信要求を前記入出力ポートへ出力するスイッチを用意するステップと、
前記ネットワークスイッチが、前記機能ノード間の前記通信経路毎に通信帯域幅を管理し前記スイッチを制御する帯域幅管理ステップと、
を備えていることを特徴とする並列計算方法。 - 前記機能ノード群を構成する複数の前記機能ノード間の通信においては、前記機能ノード群を構成する複数の前記機能ノード間のみで相互に通信を行うための通信要求であるローカル通信要求を用い、
前記通信路における通信では、前記並列計算システムを構成する全ての前記機能ノード間で相互に通信を行うためのグローバル通信要求を用い、
前記機能ノード群管理ステップにおいて、前記ローカル通信要求と前記グローバル通信要求の相互の変換のための情報を管理する機能ノード管理情報保持ステップと、
前記機能ノード群管理ステップにおいて、前記機能ノード管理情報保持ステップにおいて管理する前記情報を用いて、前記機能要素が入出力する前記ローカル通信要求と前記機能ノード入出力ポートから入出力される前記グローバル通信要求とを相互に変換する通信要求変換ステップと、
を更に備えることを特徴とする請求項12に記載の並列計算方法。 - 前記機能ノードが、前記機能ノードの電源の投入及び遮断の制御、前記機能ノードの動作周波数の制御、前記機能ノードの動作電圧の制御の何れか又はこれらを組み合わせた制御を行う電源fv制御ステップを、更に備えることを特徴とする請求項13に記載の並列計算方法。
- 前記ローカル通信要求及び前記グローバル通信要求は、当該ローカル通信要求又は当該前記グローバル通信要求が特権的な通信要求であるか否かを示す情報である特権制御情報を有しており、
前記機能ノードが、前記機能ノードが前記機能ノード入出力ポートから送出できる前記グローバル通信要求の前記特権制御情報の特権レベルの上限値を保持する特権レベル記憶ステップを、
更に備えることを特徴とする請求項13又は14に記載の並列計算方法。 - 前記帯域幅管理ステップにおいて、
前記ネットワークスイッチを前記通信経路の一部として含む前記通信経路と前記通信経路に割り当てられている通信帯域幅との対応付けを保持する帯域幅管理情報保持ステップと、
前記帯域幅管理情報に従って前記ネットワークスイッチの前記スイッチを制御するスイッチ制御ステップと、
を更に備えることを特徴とする請求項13乃至15の何れか1項に記載の並列計算方法。 - 前記ネットワークスイッチが、前記ネットワークスイッチの前記入出力ポートに対応する複数の入力ポート及び複数の出力ポートを用意するステップと、
前記ネットワークスイッチが、複数の前記入力ポート毎に前記通信経路を制御し、前記複数の出力ポートに対応した出力を持つマルチプレクサを用意するステップと、
前記ネットワークスイッチが、前記マルチプレクサの前記出力毎にグローバル通信要求を蓄積するためのバッファを用意するステップと、
複数の前記マルチプレク毎の前記出力ポートを出力先とする複数の前記バッファに蓄積されている前記グローバル通信要求を前記出力ポートに出力する帯域幅制御ステップと、
を更に備えることを特徴とする請求項13乃至16の何れか1項に記載の並列計算方法。 - 前記帯域幅制御ステップにおいて、当該帯域幅制御ステップが前記グローバル通信要求を受け取る複数の前記バッファに対する処理を、前記通信経路毎に割り当てられている通信帯域幅に比例した頻度でラウンドロビン方式によって行うことを特徴とする請求項17に記載の並列計算方法。
- 前記ネットワークスイッチが、前記ネットワークスイッチの電源の投入及び遮断の制御、前記ネットワークスイッチの動作周波数の制御、前記ネットワークスイッチの動作電圧の制御の何れか又はこれらを組み合わせた制御を行うネットワークスイッチ電源fv制御ステップを、
更に備えることを特徴とする請求項13乃至18の何れか1項に記載の並列計算方法。 - 前記機能ノードとして、前記通信路を構成する複数の前記ネットワークスイッチ電源fv制御ステップにおける動作を専用通信路を介して制御する電源fv制御ノードを備えることを特徴とする請求項19に記載の並列計算方法。
- 前記機能ノードとして、複数の前記機能ノード群から共通にアクセスすることができ、記憶機能を共有することができるMMU付き記憶ノードを持つことを特徴とする請求項13乃至20の何れか1項に記載の並列計算方法。
- 前記MMU付き記憶ノードが、前記機能ノード群管理ステップと、前記特権レベル記憶ステップと、前記電源fv制御ステップと、前記機能ノード入出力ポートを用意するステップと、に加え、
前記MMU付き記憶ノードが、前記機能ノード入出力ポートから入力された前記グローバル通信要求から機能ノード群IDを抽出するための機能ノード群ID抽出ステップと、
前記MMU付き記憶ノードが、物理アドレスと読み書き制御信号によってデータを読み書きできる記憶素子を用意するステップと、
前記MMU付き記憶ノードが、前記機能ノード群管理ステップによって前記グローバル通信要求から変換された前記ローカル通信要求から前記記憶要素に対するアクセス制御情報を抽出し、抽出された前記アクセス制御情報の論理アドレスによって前記記憶素子に対してデータの読み書き制御を行う記憶アクセス制御ステップと、
前記MMU付き記憶ノードが、前記機能ノード群毎に前記論理アドレスと前記記憶素子にアクセスするための物理アドレスの変換情報を保持する記憶管理情報保持ステップと、
前記MMU付き記憶ノードが、前記機能ノード群ID抽出ステップで抽出された前記機能ノード群IDと前記記憶管理情報を用いて前記記憶アクセスステップが出力する論理アドレスを物理アドレスに変換して前記記憶素子に入力するMMUと、を備え、
前記記憶アクセス制御ステップにおいて、前記ローカル通信要求から抽出された前記アクセス制御情報のアクセスタイプが読み出しの場合、前記記憶素子から読み出されたデータと前記機能ノード群IDとから前記読み出されたデータを転送すべき前記機能ノードへの前記ローカル通信要求を構成し前記機能ノード群管理ステップに転送し、
前記機能ノード管理ステップにおいて、前記機能アクセス制御ステップで転送された前記ローカル通信要求を前記グローバル通信要求に変換して前記機能ノード入出力ポートから送出することを特徴とする請求項21に記載の並列計算方法。 - 情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する複数の機能ノードと、前記機能ノード間で通信を行うための通信路の一部であり通信経路として機能する複数のネットワークスイッチとを有する並列計算システムにおける管理プログラムにおいて、
前記機能ノードが、
情報処理を行うための、演算機能、入出力機能及び記憶機能の何れか又はそれらを組合せた機能を有する機能要素と、
前記機能ノード間で相互に通信を行うために前記機能ノード間で転送される通信要求を入出力するためのポートである機能ノード入出力ポートと、
前記並列計算システムが有する複数の前記機能ノードの一部又は全部の前記機能ノードから構成される機能ノード群を、1つの群として管理する機能ノード群管理手段と、
を備え、
前記ネットワークスイッチが、
前記通信要求を入出力するためのポートである複数の入出力ポートと、
前記入出力ポートから入力された前記通信要求を前記入出力ポートへ出力するスイッチと、
前記機能ノード間の前記通信経路毎に通信帯域幅を管理し前記スイッチを制御する帯域幅管理手段と、
を備えている並列計算システムとしてコンピュータを機能させることを特徴とする管理プログラム。 - 請求項23に記載の管理プログラムにおいて、
前記並列通信システムを、
前記機能ノード群を構成する複数の前記機能ノード間の通信においては、前記機能ノード群を構成する複数の前記機能ノード間のみで相互に通信を行うための通信要求であるローカル通信要求を用い、
前記通信路における通信では、前記並列計算システムを構成する全ての前記機能ノード間で相互に通信を行うためのグローバル通信要求を用い、
前記機能ノード群管理手段が、
前記ローカル通信要求と前記グローバル通信要求の相互の変換のための情報を管理する機能ノード管理情報保持手段と、
前記機能ノード管理情報保持手段が管理する前記情報を用いて、前記機能要素が入出力する前記ローカル通信要求と前記機能ノード入出力ポートから入出力される前記グローバル通信要求とを相互に変換する通信要求変換手段と、
を備える並列計算システムとして機能させるための管理プログラム。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2010541339A JPWO2010064661A1 (ja) | 2008-12-04 | 2009-12-02 | 並列計算システム、その方法及びそのプログラム |
US13/133,031 US8824317B2 (en) | 2008-12-04 | 2009-12-02 | Parallel calculation system, and method and program thereof |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008-309813 | 2008-12-04 | ||
JP2008309813 | 2008-12-04 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010064661A1 true WO2010064661A1 (ja) | 2010-06-10 |
Family
ID=42233308
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/070251 WO2010064661A1 (ja) | 2008-12-04 | 2009-12-02 | 並列計算システム、その方法及びそのプログラム |
Country Status (3)
Country | Link |
---|---|
US (1) | US8824317B2 (ja) |
JP (1) | JPWO2010064661A1 (ja) |
WO (1) | WO2010064661A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2014211767A (ja) * | 2013-04-18 | 2014-11-13 | 富士通株式会社 | 情報処理システム、制御装置及び情報処理システムの制御方法 |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102522886B (zh) * | 2012-01-19 | 2013-10-30 | 浙江大学城市学院 | 一种开关电源模块并联供电装置 |
US9875204B2 (en) * | 2012-05-18 | 2018-01-23 | Dell Products, Lp | System and method for providing a processing node with input/output functionality provided by an I/O complex switch |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH09231187A (ja) * | 1996-02-22 | 1997-09-05 | Hitachi Ltd | 並列計算機のデータ転送方法 |
JP2002229963A (ja) * | 2001-02-07 | 2002-08-16 | Hitachi Ltd | 並列計算機システムおよびクロスバスイッチ |
JP2003216565A (ja) * | 2002-01-18 | 2003-07-31 | Hitachi Ltd | コンピュータシステムおよびコンフィグレーションアクセスルーティング方法 |
JP2005018753A (ja) * | 2003-06-25 | 2005-01-20 | Internatl Business Mach Corp <Ibm> | サーバ・システム内のスイッチ・モジュールの負荷を平衡させる方法およびシステムならびにこれらを利用したコンピュータ・システム |
JP2006065812A (ja) * | 2004-08-30 | 2006-03-09 | Renesas Technology Corp | データ処理装置 |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3560662B2 (ja) * | 1994-12-06 | 2004-09-02 | 富士通株式会社 | 並列プロセッサ装置 |
DE19611520A1 (de) * | 1996-03-23 | 1997-09-25 | Bosch Gmbh Robert | System zum Test eines in einem Steuergerät eingebauten Rechners |
JPH1185547A (ja) | 1997-09-12 | 1999-03-30 | Hitachi Ltd | 仮想クラスタ構成方法 |
JP2000155692A (ja) | 1998-11-18 | 2000-06-06 | Toshiba Corp | 計算機システム及びネットワークシステム並びに記録媒体 |
JP3584873B2 (ja) | 2000-10-31 | 2004-11-04 | ヤマハ株式会社 | 通信制御装置及び通信システム |
JP2002374286A (ja) | 2001-06-13 | 2002-12-26 | Hitachi Ltd | 通信品質制御システムおよび通信品質制御方法 |
US20060259733A1 (en) | 2005-05-13 | 2006-11-16 | Sony Computer Entertainment Inc. | Methods and apparatus for resource management in a logically partitioned processing environment |
JP2006319849A (ja) | 2005-05-16 | 2006-11-24 | Kddi Corp | エンドユーザ間帯域保証通信システム |
JP4606404B2 (ja) | 2006-12-01 | 2011-01-05 | 富士通株式会社 | 計算資源管理プログラムおよび計算資源管理装置 |
US9847953B2 (en) * | 2008-09-11 | 2017-12-19 | Juniper Networks, Inc. | Methods and apparatus related to virtualization of data center resources |
-
2009
- 2009-12-02 JP JP2010541339A patent/JPWO2010064661A1/ja not_active Withdrawn
- 2009-12-02 WO PCT/JP2009/070251 patent/WO2010064661A1/ja active Application Filing
- 2009-12-02 US US13/133,031 patent/US8824317B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH09231187A (ja) * | 1996-02-22 | 1997-09-05 | Hitachi Ltd | 並列計算機のデータ転送方法 |
JP2002229963A (ja) * | 2001-02-07 | 2002-08-16 | Hitachi Ltd | 並列計算機システムおよびクロスバスイッチ |
JP2003216565A (ja) * | 2002-01-18 | 2003-07-31 | Hitachi Ltd | コンピュータシステムおよびコンフィグレーションアクセスルーティング方法 |
JP2005018753A (ja) * | 2003-06-25 | 2005-01-20 | Internatl Business Mach Corp <Ibm> | サーバ・システム内のスイッチ・モジュールの負荷を平衡させる方法およびシステムならびにこれらを利用したコンピュータ・システム |
JP2006065812A (ja) * | 2004-08-30 | 2006-03-09 | Renesas Technology Corp | データ処理装置 |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2014211767A (ja) * | 2013-04-18 | 2014-11-13 | 富士通株式会社 | 情報処理システム、制御装置及び情報処理システムの制御方法 |
Also Published As
Publication number | Publication date |
---|---|
US8824317B2 (en) | 2014-09-02 |
US20110261830A1 (en) | 2011-10-27 |
JPWO2010064661A1 (ja) | 2012-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8526422B2 (en) | Network on chip with partitions | |
JP6328134B2 (ja) | クラスタ化されたコンピュータ・システムで通信チャネルのフェイルオーバを実行する方法、装置、およびプログラム | |
CN112148421B (zh) | 虚拟机迁移的方法以及装置 | |
US10768972B2 (en) | Managing virtual machine instances utilizing a virtual offload device | |
US8615623B2 (en) | Internet connection switch and internet connection system | |
US20170329625A1 (en) | Method, apparatus, and system for accessing storage device | |
US8464019B2 (en) | Apparatus for dynamically migrating LPARS with pass-through I/O devices, its method, and its program | |
US20090282211A1 (en) | Network On Chip With Partitions | |
US20090138567A1 (en) | Network on chip with partitions | |
US7849260B2 (en) | Storage controller and control method thereof | |
US20070288921A1 (en) | Emulating a network-like communication connection between virtual machines on a physical device | |
US20080052432A1 (en) | Data buffer allocation in a non-blocking data services platform using input/output switching fabric | |
JP5373893B2 (ja) | 異なるサイズを有するデータのブロックを格納し、取り出すための構成 | |
US20090228541A1 (en) | Network virtualization in a multi-node system with multiple networks | |
JP2016541072A (ja) | リソース処理方法、オペレーティング・システム、およびデバイス | |
US9904639B2 (en) | Interconnection fabric switching apparatus capable of dynamically allocating resources according to workload and method therefor | |
US20120183001A1 (en) | Network apparatus, network configuration method and program recording medium which records a network apparatus program | |
KR20140018187A (ko) | 계산 작업을 처리하기 위한 컴퓨터 클러스터 장치 및 이를 작동시키기 위한 방법 | |
JP2005309553A (ja) | 計算機 | |
US20100064301A1 (en) | Information processing device having load sharing function | |
WO2010064661A1 (ja) | 並列計算システム、その方法及びそのプログラム | |
JP2012226471A (ja) | 通信方法および通信サーバ | |
JP2019164661A (ja) | 情報処理装置、情報処理方法及びプログラム | |
JP7281528B2 (ja) | 情報処理装置 | |
JP6694007B2 (ja) | 情報処理装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09830427 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2010541339 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13133031 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09830427 Country of ref document: EP Kind code of ref document: A1 |