International Business Machines Corporation

Xingbo Wu

Li Feng Zhang

Yufei Ren

Wei Zhang

Michel H T Hack

Yandong Wang

Wei Tan

A computer-implemented topology-aware all-reduce method for an environment including a plurality of systems is provided. Each system of the systems includes a plurality of computing modules. The computer-implemented topology-aware all-reduce method according to aspects of the invention includes locally partitioning and scattering data slices among the computing modules of each system to produce local summation results. The local summation results are copied from the computing modules to corresponding host memories of the f systems. A cross system all-reduce operation is executed among the systems to cause an exchange of the local summation results across the host memories and a determination of final summation partitions from the local summation results. The final summation partitions are copied from the host memories to the corresponding computing modules of each system. The final summation partitions are forwarded to all graphical processing units (GPUs) to cause a determination of final summation results therefrom.

Graphic processor unit topology-aware all-reduce operation

Computational storage techniques for distribute computing are disclosed. The computational storage server receives input from multiple clients, which is used by the server when executing one or more computation functions. The computational storage server can aggregate multiple client inputs before applying one or more computation functions. The computational storage server sets up: a first memory area for storing input received from multiple clients; a second memory area designated for storing the computation functions to be executed by the computational storage server using the input data received from the multiple clients; a client specific memory management area for storing metadata related to computations performed by the computational storage server for specific clients; and a persistent storage area for storing checkpoints associated with aggregating computations performed by the computation functions.

Computational storage for distributed computing

A method for compressing a group of key-value pairs, the method including dividing the group of key-value pairs into a plurality of segments, creating a plurality of blocks, each block of the plurality of blocks corresponding to a segment of the plurality of segments, and compressing each block of the plurality of blocks.

Growing community of inventors

Chicago, IL, United States of America

Xingbo Wu