US 2018/0018196 A1 · Workload Management System and Process

Abstract

A high performance computing (HPC) system has an architecture that separates data paths used by compute nodes exchanging computational data from the data paths used by compute nodes to obtain computational work units and save completed computations. The system enables an improved method of saving checkpoint data, and an improved method of using an analysis of the saved data to assign particular computational work units to particular compute nodes. The system includes a compute fabric and compute nodes that cooperatively perform a computation by mutual communication using the compute fabric. The system also includes a local data fabric that is coupled to the compute nodes, a memory, and a data node. The data node is configured to retrieve data for the computation from an external bulk data storage, and to store its work units in the memory for access by the compute nodes.

Claims15

Dependent claims

A system for performing a computation that has instructions and data, the data being divisible into computational work units, the system comprising:

a compute fabric;
a plurality of compute nodes to cooperatively perform the computation by mutual communication between the compute nodes, according to the instructions, using the compute fabric;
a local data fabric simultaneously coupled to each compute node;
coupled to the local data fabric, a memory for storing computational work units, each of the plurality of compute nodes to obtain computational work units using the local data fabric;
and
coupled to the local data fabric, a data node to retrieve the data from an external bulk data storage, and to store the computational work units in the memory.

3 more elements

A tangible computer-readable medium in which is non-transitorily fixed a series of computer instructions that, when executed by a plurality of compute nodes, produces a method of performing a computation that has data, the data being divisible into computational work units, the method comprising:

configuring a plurality of compute nodes to cooperatively perform the computation by mutual communication between the compute nodes, according to the instructions, using a compute fabric;
configuring each of the plurality of compute nodes to obtain computational work units using a local data fabric simultaneously coupled to each compute node;
and
configuring a data node, coupled to the local data fabric, to retrieve the data from an external bulk data storage, and to store the computational work units in a memory coupled to the local data fabric.

1 more element

Dependencies are inferred from the claim text.

Workload Management System and Process

Life of this patent

Workload Management System and Process

Life of this patent

Abstract

Claims15