International Business Machines Corporation

Rajan Ravindran

Satish Kumar Sadasivam

Prathiba Kumar

Anbazhagan Mani

Dibyendu Das

Bhaskaran Venkatsubramaniam

Karthik Subbian

Sangram Alapati

Ramakrishnan Kannan

Nagarajan Kathiresan

Giridhar M Prabhakar

Varun Mallikarjunan

Chiranjib Sur

Gowri Shankar Palani

Manish Satyapal Gupta

Hubertus Franke

Pradipta Kumar Banerjee

Vaidyanathan Srinivasan

Madhavi Gopal Valluri

Shailabh Nagar

Jong Hyuk Choi

Kumar Prathiba

Described are a system and method for identifying variables which impact performance of software under development. Data is collected that is related to performance characteristics of the software under development. Performance change gradients are determined between previous builds of the software under development. A set of performance change factors are generated from the collected data that corresponds to each performance change gradient. Performance characteristic data corresponding to a current build of the software under development are compared to the performance change gradients. At least one fault component from the set of performance change factors that influences performance of the current build is output in response to the comparison between the performance characteristic data corresponding to the current build and the plurality of performance change gradients.

Systems and methods for identifying software performance influencers

A method of mapping processes to processors in a parallel computing environment where a parallel application is to be run on a cluster of nodes wherein at least one of the nodes has multiple processors sharing a common memory, the method comprising using compiler based communication analysis to map Message Passing Interface processes to processors on the nodes, whereby at least some more heavily communicating processes are mapped to processors within nodes. Other methods, apparatus, and computer readable media are also provided.

Process mapping parallel computing

The present disclosure includes, but is not limited to, a method, system and computer-usable medium for improving performance measurement by analyzing the various events in a multiplexing counting mode and configuring the sampling time accordingly to more effectively performing the sampling. In certain embodiments, when groups of operations are identified for sampling, the present disclosure generates a time sampling table for these groups of operations. The time sampling table is dynamically altered during the runtime of the application to alter the sampling interval of each group. The sampling interval of each group can be increased or decreased based on a threshold of occurrence of the event. This disclosure provides more accurate performance measurement of important events and facilitates a determination of how important events impact application performance.

Smart multiplexing of performance counters for performance measurement

Process mapping in parallel computing

A computer program product for identifying bottlenecks includes a computer readable storage medium with stored computer readable program instructions. The computer readable program instructions, when executed, provide a data collector module, a mapper module, and an analyzer module that are collectively configured to read mapped data and configuration files, and identify, based upon the mapped data and the configuration files, an undesirable bottleneck condition that causes a computer program to run inefficiently. A method includes reading a configuration file that includes data regarding processor components, and collecting data from hardware activity counters based upon the configuration file. The method also includes mapping the collected data to corresponding sections of code of a computer program, reading the mapped data and the configuration file, and identifying, based upon the reading of the mapped data and the configuration file, an undesirable bottleneck condition that causes the processor to run the computer program inefficiently.

Performance bottleneck identification tool

Embodiments of the invention broadly contemplate systems, methods, apparatuses and program products providing a power management technique for an HPC cluster with performance improvements for parallel applications. According to various embodiments of the invention, power usage of an HPC cluster is reduced by boosting the performance of one or more select nodes within the cluster so that the one or more nodes take less time to complete. Embodiments of the invention accomplish this by selectively identifying the appropriate node(s) (or core(s) within the appropriate node(s)) in the cluster and increasing the computing capacity of the selected node(s) (or core(s) within the appropriate node(s)).

Systems and methods for power management in a high performance computing (HPC) cluster

A system, and computer program product for cloud optimization using workload analysis are provided in the illustrative embodiments. An architecture of a workload received for execution in a cloud computing environment is identified. The cloud computing environment includes a set of cloud computing resources. A section of the workload is identified and marked for static analysis. Static analysis is performed on the section to determine a characteristic of the workload. A subset of the set of cloud computing resources is selected such that a cloud computing resource in the subset is available for allocating to the workload and has a characteristic that matches the characteristic of the workload as determined from the static analysis. The subset of cloud computing resources is suggested to a job scheduler for scheduling the workload for execution.

Cloud optimization using workload analysis

A method for dynamic optimization of thread assignments for application workloads in an simultaneous multi-threading (SMT) computing environment includes monitoring and periodically recording an operational status of different processor cores each supporting a number of threads of the thread pool of the SMT computing environment and also operational characteristics of different workloads of a computing application executing in the SMT computing environment. The method further can include identifying by way of the recorded operational characteristics a particular one of the workloads demonstrating a threshold level of activity. Finally, the method can include matching a recorded operational characteristic of the particular one of the workloads to a recorded status of a processor core best able amongst the different processor cores to host execution in one or more threads of the particular one of the workloads and directing the matched processor core to host execution of the particular one of the workloads.

Dynamic optimization of thread assignments for application workloads in parallel computing

An energy-aware backfill scheduling method combines overestimation of job run-times and processor adjustments, such as dynamic voltage and frequency scaling, to balance overall schedule performance and energy consumption. Accordingly, some scheduled jobs are executed in a manner reducing energy consumption. A computer-implemented method comprises identifying job performance data for a plurality of representative jobs and running a simulation of backfill-based job scheduling of the jobs at various combinations of run-time over-estimation values and processor adjustment values. The simulation generates data including energy consumption and job delay. The method further identifies one of the combinations of values that optimizes the mathematical product of an energy consumption parameter and a job delay parameter using the simulation generated data for the plurality of jobs. Jobs submitted to a processor are then scheduled using the identified combination of a run-time over-estimation value and a processor adjustment value.

Job scheduling to balance energy consumption and schedule performance

A method for cloud optimization using workload analysis is provided in the illustrative embodiments. An architecture of a workload received for execution in a cloud computing environment is identified. The cloud computing environment includes a set of cloud computing resources. A section of the workload is identified and marked for static analysis. Static analysis is performed on the section to determine a characteristic of the workload. A subset of the set of cloud computing resources is selected such that a cloud computing resource in the subset is available for allocating to the workload and has a characteristic that matches the characteristic of the workload as determined from the static analysis. The subset of cloud computing resources is suggested to a job scheduler for scheduling the workload for execution.

A computer program product including computer usable program code embodied on a computer usable medium, the computer program product comprising: computer usable program code for identifying job performance data for a plurality of representative jobs; computer usable program code for running a simulation of backfill-based job scheduling of the plurality of jobs at various combinations of a run-time over-estimation value and a processor adjustment value, wherein the simulation generates data including energy consumption and job delay; computer usable program code for identifying one of the combinations of a run-time over-estimation value and a processor adjustment value that optimize the mathematical product of an energy consumption parameter and a job delay parameter using the simulation generated data for the plurality of jobs; and computer usable program code for scheduling jobs submitted to a processor using the identified combination of a run-time over-estimation value and a processor adjustment value.

Growing community of inventors

Bengaluru, India

Rajan Ravindran