Browse Prior Art Database

Predicting speed-up effect by slowing down (P)

IP.com Disclosure Number: IPCOM000248936D
Publication Date: 2017-Jan-23
Document File: 2 page(s) / 51K

Publishing Venue

The IP.com Prior Art Database

Abstract

Predicting speed-up effect by slowing down

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 57% of the total text.

1

Predicting speed -up effect by slowing down (P)

It is often hard to predict system response for change of certain component to better performing one. For example - "If I change my networking topology from 1GbE to 10GbE will my cluster run better in terms of job execution times ?" This can of course be checked by changing the networking hardware and observing the results - but this approach is costly - when it proves no gain after hardware change . Similar problem exist with all resources that may perform better when upgraded (disk arrays, CPUs etc.). This proposal shows how to, with more certainty (but not perfectly), check if a change is viable without significant costs .

The idea is: Since we cannot (usually, without additional costs) speed system up to check what happens, we usually can check how it will behave when the resource is slowed down. And looking at system response draw certain conclusions leading to better business decisions (investing in hardware etc.). Such approach is especially important as complex system's behavior upon such change (ie resource speed modification) is very hard to predict, but not that hard to measure.

The core idea behind this disclosure is to slow given resource down - be it network, disk etc. and check system reaction to change. We sample system performance at several slow down points - say 10%, 30%, 50% and we usually see some relation between the two. So what we are trying to get is to establish function overall_perfrormance(resource...