International Business Machines Corporation

Richard J Coppinger

Alexander Druyan

Enci Zhong

Hanhong Xue

Chulho Kim

William G Tuel, Jr

Christophe Fagiano

Christophe Lombard

David Paul Brelsford

Deryck X Hong

Gary J Mincher

Christophe Pierre Francois Quintard

This present invention provides a portable user space application release/reacquire of adapter resources for a given job on a node using information in a network resource table. The information in the network resource table is obtained when a user space application is loaded by some resource manager. The present invention provides a portable solution that will work for any interconnect where adapter resources need to be freed and reacquired without having to write a specific function in the device driver. In the present invention, the preemption request is done on a job basis using a key or 'job key' that was previously loaded when the user space application or job originally requested the adapter resources. This is done for each OS instance where the job is run.

Managing preemption in a parallel computing system

Jobs are scheduled such that resource reuse is maximized and deallocation of reusable resources is minimized. A multi-phase scheduling capability is provided in which during a first phase, weights are assigned to the reusable resources, and during a second phase, the jobs are scheduled. If reusable resources are available for the job, the job is assigned to those resources. Otherwise, a search is performed for a solution which minimizes the weight and number of reusable resources that are to be deallocated.

Job scheduling to maximize use of reusable resources and minimize resource deallocation

A hierarchical fanout connectivity infrastructure is built and used to start a parallel application within a parallel computing environment. The connectivity infrastructure is passed to a checkpoint library, which employs the infrastructure and a defined sequence of events, to perform checkpoint, restart and/or migration operations on the parallel application.

Fanout connectivity structure for use in facilitating processing within a parallel computing environment

A method is provided for prioritizing jobs in a computing environment having a plurality of nodes sharing one or more resources. Such method includes, given a job A to be scheduled having certain resource requirements, generating a list P of already scheduled jobs which utilize resources required by job A and which can be preempted by job A. For each resource requirement of job A, an ordered list of jobs which can be preempted by job A is created which is ranked by the applicability of each job in the list relative to resource requirements of job A, so as to provide a plurality of ordered lists, one for each resource requirement. An ordered list L is generated by selecting one of the ordered lists, adding the top-ranked job from that ordered list to list L, then adding the top-ranked job from another one of the ordered lists to list L and so on, by repetitively selecting the top-ranked job from individual ones of all the ordered lists. Finally, a set of jobs to be preempted by job A is selected based on the list L of jobs.

Job preempt set generation for resource management

A method, system and program product are provided for facilitating scheduling of computer jobs of a computing environment having a plurality of resources. The method includes decoupling a job scheduling algorithm from responsibility for maintaining data structures for tracking recorded usage of the plurality of resources by computer jobs. The decoupling includes (i) providing resource space data structures separate from the job scheduling algorithm for recording usage of the plurality of resources by computer jobs; and (ii) providing operators for facilitating manipulation of values within the resource space data structures by the job scheduling algorithm and by one or more functions external to and independent of the job scheduling algorithm.

Facilitating scheduling of jobs by decoupling job scheduling algorithm from recorded resource usage and allowing independent manipulation of recorded resource usage space

Disclosed are a method, information processing system, and computer readable medium for reserving resources in a massively parallel processing system comprising. The method includes receiving a reservation request for reserving at least one resource within a massively parallel processing system. The massively parallel processing system includes compute nodes, base partitions, switches, wires, and node cards. The reservation type associated with the reservation is determined. The reservation type is at least one of a number of compute nodes, a base partition list, a particular shape configuration, a particular job, and a particular partition. A set of requirements is generated based on the reservation type. A set of resources is identified for satisfying the set of requirements and the set of resources is reserved.

Resource reservation for massively parallel processing systems

A fault tolerant capability for the reliable collection and aggregation of disparate data from multiple processing units. Data of a processing unit is collected and locally aggregated at that processing unit. The locally aggregated data is stored in a local database, and periodically forwarded to a central manager. In response to receiving the data, the central manager determines whether the data should be included in a total aggregate of data for the processing environment.

Growing community of inventors

Milton, NY, United States of America

Richard J Coppinger