International Business Machines Corporation

Dinesh Kumar

Zon-Yin Shae

Hani Talal Jamjoom

Li Feng Zhang

Gary A Van Huben

Asser Nasreldin Tantawi

Colleen Ann Weller

Margaret Annabelle Allen

Addison Daniel Ferrari

According to a technique of code development, a plurality of subsystem power profiles for a corresponding plurality of subsystems of a host platform in a production data processing environment are received. Each of the subsystem power profiles specifies power consumption of a corresponding one of the plurality of subsystems across a range of workload traffic intensities. Based on the plurality of power profiles, source code and/or software architecture of the source code is power-optimized in a development data processing environment such that, in the production data processing environment, power consumption per unit of performance of an executable program derived from the source code is improved.

Cross-layer power optimization of program code and/or software architecture

Systems and methods are provided for scheduling homogeneous workloads including batch jobs, and heterogeneous workloads including batch and dedicated jobs, with run-time elasticity wherein resource requirements for a given job can change during run-time execution of the job.

Scheduling homogeneous and heterogeneous workloads with runtime elasticity in a parallel processing environment

Accepting a job having a job size representing a number or quantity of processors; computing an expected size, and a standard deviation in size, for the accepted job; adding the expected size to the standard deviation in size to determine a sum; comparing the sum to a number or quantity of available clusters at each of a plurality of non-leaf nodes of a tree representing a high-performance computing environment; and when the number or quantity of available clusters is more than the sum at a sub-tree of the tree and, going down one level further in the sub-tree, the number of available clusters is less than the sum, selecting the sub-tree for the accepted job such that the accepted job is placed on one or more clusters associated with the selected sub-tree.

Strategic placement of jobs for spatial elasticity in a high-performance computing environment

Methods and systems for determining a real-time performance model of systems that process multi-class workloads. The methods can include collecting measurement data, selecting a series of prior time windows, processing the measurement data to compute a set of constraints based on the measurement data from the series of prior time windows, augmenting the set of constraints to a measurement model to obtain an augmented measurement model and running a modified extended Kalman filter with the augmented measurement model to obtain a state estimate.

Real-time performance modeling of software systems with multi-class workload

A method for modeling performance of an information technology system having one or more servers for serving a number of types of transactions includes modeling a service time of each transaction type at each server and a processor overhead at each server as one of a polynomial, exponential, or logarithmic function of the average arrival rate of each transaction type at the corresponding server to generate service time and processor overhead functions and inferring optimal values of coefficients in the service time and processor overhead functions to generate a performance model of the information technology system.

Growing community of inventors

Sleepy Hollow, NY, United States of America

Dinesh Kumar