JP2010102523A - Blade server system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To continue an operation even if a trouble occurs in a blade server while a symmetrical multiprocessor blade system is in operation. <P>SOLUTION: The blade server system includes a plurality of blade servers and a management device for managing the whole system. Whether each of the blade servers operates as a symmetrical multiprocessor blade server system or an independent blade server system is set depending on system configuration information stored in the management device. An auxiliary blade server stands by for the symmetrical multiprocessor blade server system. If any of the blade servers constituting the symmetrical multiprocessor blade server system fails, the management device cuts off supply of power to the defective blade server, and turns on the auxiliary blade server. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  A blade server (hereinafter referred to as a blade) is a circuit board on which server functions such as a CPU and a memory are mounted as one of the components of a server. A blade server system operates by attaching a plurality of substrates to a casing (chassis, enclosure). The present invention relates to a recovery method from a failure of a blade server system having a plurality of symmetric multiprocessors in the blade server field.

  In the symmetric multiprocessor system apparatus, a plurality of processors are mounted on one server to improve the processing capability and enhance the fault tolerance. In addition, as a requirement of the symmetric multiprocessor system apparatus, the processor type and the amount of installed memory must be the same. In the blade server field, two or more blades equipped with symmetric multiprocessors are used to form one system. However, by preparing a small number (M) of spare blades for a plurality (N) of business blades in a symmetric multiprocessor blade system, a technology that improves fault tolerance while reducing costs (hereinafter referred to as N + M) However, there is no configuration that moves only the blade in which the failure has occurred, although the system can move from the blade in which the failure has occurred to another blade that is on standby.

JP 2006-190240 A

  If multiple blades with a symmetric multiprocessor are configured and operated as a single blade server system, if one blade fails, replace the failed blade without stopping the system. Can not do it. Further, it takes some time to move each system such as N + M, and the system operation (transaction or job processing) must be temporarily stopped during the movement.

  In the present invention, even if a failure occurs in a blade while the system is operating, it is possible to switch to a standby blade that is in a standby state without stopping or moving the system and to continue operations. To do. Further, by replacing the failed blade, it can be made to stand by as a new spare blade.

  In an environment where multiple blades can be installed, a spare blade is installed in a symmetric multiprocessor blade server system, and if a failure occurs, the failed blade goes offline and the spare blade goes from standby to online. Changed and switched dynamically.

  According to the present invention, in a blade system having a symmetric multiprocessor configuration, operation can be continued without stopping the entire system when a specific blade failure occurs. In addition, it is possible to perform maintenance that can replace a failed blade while operating.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a configuration diagram of a general blade server system. At least two or more blades 10 (# 0 to #n), a service processor 11 that manages the entire blade server system device, an external storage device, and a PCI module 12 on which an interface card for connection for communication is mounted. , A backplane 13 for realizing signal path transmission between units or modules, and a case 14 (also referred to as a chassis or an enclosure) as a device for mounting them. At least one processor 15 can be mounted on the blade 10, the processor 15, the memory 16, the node controller 18 that controls the I / O access 17, and power control, configuration management, environmental monitoring, etc. inside the blade 10. By providing the baseboard management controller 19 having the above functions, it has a function as one blade server system apparatus.

  FIG. 2 shows an example of a symmetric multiprocessor system 26 according to the present invention. A plurality of blades 20 (# 0 to #n), a service processor 21 that manages the entire blade server system apparatus, an external storage device, a PCI module 22 on which an interface card is connected for communication to the outside, a unit or The node controller 24 in each blade 20 includes a symmetric multiprocessor coupling interface 25, and includes a plurality of blades 20 via the backplane 23. Has a function of realizing a symmetric multiprocessor system with one.

  FIG. 3 shows the configuration of the node controller 31. The node controller firmware 32 in the node controller 31 manages transactions and the shared memory. The node controller 31 has a memory space management table 34 for managing a memory space shared with a transaction management information table 33 necessary for executing transaction processing. The transaction management information table 33 manages transaction queue information, the status of a transaction being executed, the address of a processing destination processor, and the like. The node controller firmware 32 assigns a transaction to each blade. After the process is completed, the queue is deleted from the transaction management information table 33. The memory space management table 34 can be accessed from each processor. To access the memory space management table 34, the memory space management table 34 is referred to, it is confirmed that no other processor is accessing, and the area is reserved and used. . After use, the reserved area is released by accessing the memory space management table 34. By exclusively controlling the memory space, it is possible to prevent a failure caused by reference or overwriting by another processor.

  FIG. 4 shows a typical example of the system configuration of the blade server system 48 independent of the symmetric multiprocessor configuration system, and will be specifically described. In this system configuration example, four blades 40 (# 0 to # 3) are mounted on the backplane 41, and a single symmetric multiprocessor server system in which # 0, # 1, and # 2 are connected is provided. In the configuration used as an independent blade server system, first, the user sets the configuration information of the system device in the management program 43 operating on the service processor 42. The configuration information set here is stored in the memory 44 in the service processor 42, and the configuration information is not erased even when the power of all system devices including the service processor 42 is shut off. Each time the system is activated, the information is transmitted from the service processor 42 to the baseboard management controller 45 in each blade 40. Each baseboard management controller 45 instructs a form to be used according to the configuration information transmitted from the service processor 42. In the configuration example of this figure, the blade servers # 0, # 1, and # 2 serve as the symmetric multiprocessor blade server system 47, and the blade # 3 serves as an independent blade server system. Even when the blade 40 is replaced due to a failure by holding the system configuration information in the memory 44 in the service processor 42, a procedure for taking over the system configuration information becomes unnecessary. Also, two service processors (42, 46) for centrally managing system configuration information are installed, the service processor 42 is mainly operated as system device management, and the other service processor 46 is made to stand by to be put on the service processor 42. Even if a failure occurs, it is possible to improve the reliability of the entire system apparatus by taking over all information managed by the waiting service processor 46.

  The present invention will be described with reference to FIG. First, in the management program 58 operating on the service processor 51, the user determines the number of blades to be operated in the symmetric multiprocessor blade server system 52 and one blade to be standby. The blade that performs the main processing is called a primary server, the standby blade is called a standby server, and the other blades are called secondary servers. In the embodiment, the blade 54 is a primary server, the blade 55 is a secondary server, and the blade 56 is a standby server. The system configuration information of the set system is stored in the memory 57 in the service processor 51. At the time of starting the symmetric multiprocessor blade server system, the processor type and the number of processors installed in each blade server (54, 55, 56) are installed based on the system configuration information held in the service processor 51. After confirming that all the memory amounts are the same, each server blade (54, 55, 56) starts to boot the system in synchronization.

  The baseboard management controller 59 of the blade 55 in the symmetric multiprocessor blade server system 52 detects the failure and notifies the service processor 51 of the occurrence of the failure. Simultaneously with the notification, the baseboard management controller 59 notifies the baseboard management controller 60 of the blade 54, which is the primary server, of the occurrence of a failure. The notified baseboard management controller 60 executes the detachment process of the blade 55 in which a failure has occurred with respect to the node controller 61. The transaction processing that cannot be executed due to the failure is re-queued from the transaction management information table 62, and the processing destination blade 55 is changed from the transaction management information table 62 to unusable. Further, the memory space management table 63 is changed so that the shared memory provided by the failed blade 55 cannot be used. The node controller 61 is notified to the service processor 51 via the baseboard management controller 60 that the disconnection has been completed. The notified service processor 51 instructs the baseboard management controller 59 of the blade 55 in which the failure has occurred to execute the power-off process. At the same time, the system configuration information instructs the baseboard management controller 64 of the blade 56 waiting as a standby to turn on the power. The service processor 51 receives the blade 56 activation completion notification from the baseboard management controller 64, and the standby blade server 56 becomes usable via the baseboard management controller 60 of the blade 54 as the primary server. The processing server blade 56 is added from the transaction management information table 62, and the memory space management table 63 is changed so that the shared memory can be provided. As a result, it is possible to prevent the symmetric multiprocessor blade server system 52 from being stopped or the processing due to movement of the entire system from being temporarily interrupted. This system is a blade server system having a symmetric multiprocessor configuration, but the blades 55 other than the blade 54 of the primary server only execute transaction processing, and therefore can be disconnected.

  Further, the failed blade 55 is subjected to maintenance and replacement, and then can be mounted as a standby blade 55.

It is a block diagram of a general blade server system It is a system configuration figure of an embodiment of the present invention. It is a configuration example of a node controller It is a configuration example in which the embodiment of the present invention and independent blade server systems are mixed. Examples of the present invention

Explanation of symbols

10: Blade 11: Service processor 12: PCI module 13: Backplane 14: Housing 15: Processor 16: Memory 17: I / O
18: Node controller 19: Baseboard management controller 20: Blade 21: Service processor 22: PCI module 23: Backplane 24: Node controller 25: Symmetric multiprocessor coupling interface 26: Symmetric multiprocessor blade server system 31: Node controller 32: Node controller firmware 33: Transaction management information table 34: Memory space management table 40: Blade 41: Backplane 42: Service processor 43: Management program 44: Memory 45: Baseboard management controller 46: Service processor 47: Symmetric multi Processor blade server system 48: Independent blade server system 51: Service processor 52: Symmetric multiprocessor Blade server system 53: Blade 54: Blade (primary server)
55: Blade (secondary server)
56: Blade (standby server)
57: Memory 58: Management program 59: Baseboard management controller 60: Baseboard management controller 61: Node controller 62: Transaction management information table 63: Memory space management table 64: Baseboard management controller

Claims (2)

  Whether the blade server operates as a symmetric multiprocessor blade server system or as an independent blade server system based on system configuration information stored in the management device. In the blade server system to be set, when the spare blade server is put on standby in the symmetric multiprocessor blade server system and a failure occurs in the blade server constituting the symmetric multiprocessor blade server system, the management device A blade server system characterized in that a blade server in which a failure has occurred is shut off and a spare blade server is turned on.   After the failed blade server is switched to a spare blade server, the failed blade server is replaced with a new blade server while the system is operating, and the new blade server is put on standby as a spare blade server. The blade server system according to claim 1.

Images