International Business Machines Corporation

Huawei Technologies Co., Limited

Zhaohui Ding

Chong Chen

Michael Spriggs

Qi Wang

Zhe Jin

Chun Liang Yuan

Ajith Shanmuganathan

Xiu Qiao Li

Xun Pan

Rong Song Shen

Jun Jiang

Bo Chen Zhu

Shutao Yuan

Da Xu

Sicong Li

Zhao Lu

Kai Huang

Rongsong Shen

Wang Qi

Wenjin Gou

Chun Lin Yang

The present disclosure relates to methods and apparatuses for processing a control packet in a collective communication system, where the collective communication system includes a switch network and multiple computing nodes, and the switch network includes a first switch. In one example method, the first switch forwards a query packet transmitted by a source node to a destination node, where the query packet is generated by the source node based on a context of the collective communication system. Then, the first switch forwards a notification packet transmitted by the destination node to the source node, where the notification packet carries an in-network computing capability of the switch network.

Method and apparatus for processing control packet in collective communication system, device, and system

A computer-implemented method, a computer system and a computer program product for releasable resource-based preemptive scheduling. One or more currently running workloads are determined to be preempted by a pending workload. Releasable resources from the one or more currently running workloads meet required resources of the pending workload. The pending workload is dispatched so that it uses at least part of the releasable resources from the one or more currently running workloads to run.

Releasable resource based preemptive scheduling

A method and system for process tracking is disclosed. A first process is associated with an identity of a job, wherein the first process is launched by an outsourcing process to process the job. A second process is created locally. The second process is associated with the identity of the job to enable the second process to manage the first process based on the identity.

Process tracking

A technique for allocating a global resource in a distributed grid environment includes a master resource manager (MRM), and a plurality of slave resource managers (SRMs) each configured to allocate a global resource within a respectively controlled region of a distributed grid computing environment. The MRM receives from each SRM use data and demand data associated with the global resource by each consumer in the respective region controlled by the respective SRM. The MRM then computes an allocation of the global resource for each consumer and for each region based on the use data and the demand data. The MRM then communicates to each SRM an amount of the global resource allocated to each consumer in each respective region controlled by a respective SRM and a total allocation of the global resource for the region controlled by the respective SRM.

Allocating a global resource in a distributed grid environment

A technique for allocating a global resource in a distributed grid environment includes receiving from slave resource managers use and demand data associated with the global resource by each consumer in a respective region. A master resource manager computes an allocation of the global resource for each consumer and for each region based on the use data and the demand data. The master resource manager then communicates to each slave resource manager an amount of the global resource allocated to each consumer in each respective region controlled by a respective slave resource manager and a total allocation of the global resource for the region controlled by the respective slave resource manager.

A computer-implemented method comprises identifying a computer task; determining a hardware exception source associated with the computer task; determining an exception unit associated with the hardware exception source; determining a parallelization factor associated with the hardware exception source; and determining a parallel execution scenario associated with the computer task based on the exception unit and the parallelization factor. A corresponding computer program product and computer system are also disclosed.

Managing execution of computer tasks under time constraints

A method for estimating job start times on a system may include: performing a first simulation of the system by a first process to generate a first estimate of the start time based a first snapshot of the computing system; and performing a second simulation of the system to generate a second estimate of the start time. The first and second estimates may each be either a distinct time estimate or a range of time estimate, depending on whether the simulations dispatches the job before the respective first and second processes end. The method may then include generating a fifth estimate of the time that the computing job will start by a third process based on estimates respectively generated by the first and second processes.

Estimating job start times on workload management systems

A technique for allocating a global resource in a distributed grid environment includes: a master resource manager (MRM) that receives from slave resource managers use and demand data associated with a global resource by each consumer in a region controlled by a respective slave resource manager. The MRM computes an allocation of the resource for each consumer and for each region by determining a total amount of the resource to allocate, estimating a need by each consumer for the resource based on the use and demand data, determining an allocation of the total amount of the resource among the consumers based on the estimated need; subdividing the allocation for each consumer among each region where the consumer utilizes the resource; and computing a total allocation of the resource for each region.

Allocating a global resource to consumers across different regions of a distributed grid environment based on use data and demand data of each consumer

A technique for allocating a global resource in a distributed grid environment includes: receiving from slave resource managers use and demand data associated with the global resource by each consumer in a respective region; compute an allocation of the resource for each consumer and for each region by determining a total amount of the resource to allocate, estimating a need by each consumer for the resource based on the use and demand data, determining an allocation of the total amount of the resource among the consumers based on the estimated need; subdividing the allocation for each consumer among each region where the consumer utilizes the resource; and computing a total allocation of the resource for each region; and communicating to each slave resource manager an amount of the resource allocated to each consumer in each respective region and the total allocation of the resource for the region.

According to one aspect of the present disclosure, a method and technique for job distribution within a grid environment is disclosed. The method includes: receiving jobs at a submission cluster for distribution of the jobs to at least one of a plurality of execution clusters, each execution cluster comprising one or more execution hosts; determining resource capacity corresponding to each execution cluster; determining resource requirements for the jobs; dynamically determining a pending job queue length for each execution cluster based on the resource capacity of the respective execution clusters and the resource requirements of the jobs; and forwarding jobs to the respective execution clusters according the determined pending job queue length for the respective execution cluster.

Growing community of inventors

Beijing, China

Zhaohui Ding