Distributed computing system and resource allocation method
In a distributed computing system, a bottleneck for performance of a network is avoided and a high-performance scalable resource management function is achieved. The distributed computing system includes a plurality of components connected to each other via a network. Each of the components includes a plurality of compute nodes, a plurality of drive casings, and a plurality of storage devices. The network includes a plurality of network switches and is configured in layers. When a storage region is to be allocated to a compute node among the compute nodes, a managing unit selects, from the storage devices, a storage device related to the storage region to be allocated or selects, from the drives, a drive casing related to the storage region to be allocated, based on a network distance between two of the compute node, the storage device, and the drive casing.
1. A distributed computing system comprising a plurality of components connected to each other via a network, wherein
each of the components includes
a plurality of compute nodes that activate an application and issue an input and output request to input and output data on the application,
a plurality of drives that physically store the data to be input and output in accordance with the input and output request, and
a plurality of storage devices that execute a process on the data to be input to and output from the drives,
wherein, the network has a plurality of network switches and is configured in layers, and
wherein, upon allocating a storage region to a compute node among the compute nodes, a managing unit selects one of a storage device related to the storage region and a drive related to the storage region, based on a network distance between the compute node and the storage device and the network distance between the storage device and the drive,
wherein the network distance is determined based on the number of network switches via which data is transferred between the compute node and the storage device and between the storage device and the drive.
2. The distributed computing system according to claim 1 , further comprising switch connection management information storing a connection relationship between the network switches, wherein
the managing unit uses the switch connection management information to determine the network distance.
3. The distributed computing system according to claim 1 , wherein
the drives have a mode in which the drives transmit and receive data to and from the compute nodes without the storage devices,
the managing unit selects another drive related to another storage region to be allocated, based on network distances between the compute nodes and the another drive.
4. A resource allocation method of a distributed computing system including a plurality of components connected to each other via a network, wherein
each of the components includes
a plurality of compute nodes that activate an application and issue an input and output request to input and output data on the application,
a plurality of drives that physically store the data to be input and output in accordance with the input and output request, and
a plurality of storage devices that execute a process on the data to be input to and output from the drives,
the network has a plurality of network switches and is configured in layers, and
upon allocating a storage region to a compute node among the compute nodes, a managing unit selects one of a storage device related to the storage region and a drive related to the storage region, based on a network distance between two of the compute node and the storage device and the network distance between the storage device and the drive,
wherein the network distance is determined based on the number of network switches via which data is transferred between the compute node and the storage device and between the storage device and the drive.