Browse Prior Art Database

Grid QoS Job Equalizer

IP.com Disclosure Number: IPCOM000012586D
Original Publication Date: 2003-May-16
Included in the Prior Art Database: 2003-May-16
Document File: 2 page(s) / 40K

Publishing Venue

IBM

Abstract

Grid QoS Job Equalizer is based around grid customers wanting to optimize performance on the grid and attain some level of quality of service. Grid QoS Job Equalizer can determine if the same job on one computer is far out pacing the same job on another computer and automatically bring the later computer's job progress equal to the progressive computers level. The idea being that the ebb and flow of grid performance can lead to any given computer performing better throughout the life cycle of the job. The Grid QoS Job Equalizer accelerates all computers running this job to the level of the best performing computer at that moment in time.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

Grid QoS Job Equalizer

Grid QoS Job Equalizer is based around grid customers wanting to optimize performance on the grid and attain some level of quality of service. The current state of the art in grid computing, a job (program) can be executed on the grid and directed to run on a specific resource, e.g. a fast host. A fast host may not yield fast results. Grids can be geographically vast and databases servers and replica servers may respond quicker serving one area of the grid, and even this may vary with the time of day, server demand, and network traffic. All of this leads to unpredictable performance of a job, even when running on a fast grid resource.

Also common in grid technology is that grid jobs (programs) run for a long time, and are often very math/cpu intensive (cancer research, physics, etc).

The above characteristics of grid computing means that if the exact same job was submitted on grid computer A and grid computer B, one might yield better performance over the other on any given day. Moreover, one computer may perform better and far out pace the other for a portion of the total job, only to slow and have the other computer perform better for the second half of the job.

Grid QoS Job Equalizer can determine if the same job on one computer is far out pacing the same job on another computer and automatically bring the later computer's job progress equal to the progressive computers level. The idea being that the ebb and flow of grid performance can lead to any given computer performing better throughout the life cycle of the job. The Grid QoS Job Equalizer accelerates all computers running this job to the level of the best performing computer at that moment in time.

Grid QoS Job Equalizer combines existing technologies to make its' job accelerator process work.

First Grid QoS Job Equalizer uses cpu time as a fairly accurate measure of job progression. If the exact same job running on different computers of equal processing power will always finish in approximately the same amount of cpu time. Real time may vary, but cpu time will consistently remain close to equal for the same job running on any given computer. Basically, whenever the processor blocks (must wait) its' cpu clock stops and it is move...