Method and device for quickly starting massively parallel computer system

Abstract

The invention discloses a method and a device for quickly starting a massively parallel computer system. The method comprises the steps that a configuration interface is arranged for each computer node, and the computer node reads characteristic parameters needed for starting the system and stored in the configuration interface in advance before starting; if the characteristic parameters in the configuration interface are usable, the computer node starts the system according to the read characteristic parameters; if the characteristic parameters in the configuration interface are unusable, characteristic parameters for the computer node are generated through a management controller and written into the corresponding configuration interface, then the computer node reads the characteristic parameters from the configuration interface and starts the system according to the read characteristic parameters; and the device comprises the management controller and the configuration interfaces, wherein the configuration interfaces are arranged in the computer nodes and connected with microprocessors of the computer nodes, and the management controller is connected with at least one configuration interface. The method provided by the invention has the advantages of being fast in starting speed, high in starting efficiency, good in flexibility and low in implementation cost.

Description

technical field [0001] The invention relates to the field of computer system management automation, in particular to a method and device for quickly starting a large-scale parallel computer system. Background technique [0002] A parallel computer system usually includes many computer nodes connected by a high-speed interconnection network. These computer nodes all have their own microprocessors, local memory, high-speed interconnection interfaces, and I / O interfaces. They have the same physical structure but are relatively independent. system. In order to distinguish, each computer node is assigned a set of globally unique identifiers, such as network identifier (NID-Network Identification), IP address and hostname, etc. These regular identifiers are called characteristic parameters. The node of the parallel computer system can be configured with a local hard disk and set to start the system locally. However, in a large-scale system, the installation, configuration and upd...

AI Technical Summary

Problems solved by technology

However, the development speed of the current computer system has exceeded people's imagination. Large-scale computer systems such as grid and cloud emerge in endlessly. More than 99% of the TOP500 computer systems have more than K-level nodes (statistics in June 2011). In such a system, when many computer nodes start at the same time, the performance and bandwidth of the startup server may not match the demand, resulting in slow system startup

Images