TheEffect onThroughput ofMultiprocessing inaMultiprogramming Environment
John Hogarth · 1973
This paper investigates someoftheeffects ofvarying the Model4:Multiprogrammingwith multitasking andwithjobs numberofcentral processing units (CPU's) available toamultiprogram- differentiated. mingsystem bothwhenparallel processing ofasingle task(multitask- Forthefirst twomodels, bothsingle andmultiple CPUcon- ing) isallowed andisnotallowed. Thevariables investigated arethose f that would beexpected tocontrol theCPUqueue length distribution figurations areconsidered. Thebasic goal istoestablish quanti- andinclude thedegree ofmultiprogramming, CPUservice discipline, tative relations between throughput andthenumberofCPU's CPUservice distribution, thedegree ofcooperation between processors inthevarious system models andtoevaluate theeffect ofsec- engaged inmultitasking, andtheratio ofCPUtoI/Oservice capacity. ondary variables onthroughput inthevarious models. The Thecomputer systems aremodeled byqueueing networks. Analytic results aredisplayed inmanycaseswithsupplementation bynumericsysemodels hosen aredesignedetotbegrepresentive andsimulation solutions whenconvenient ornecessary. somecurrent problems insystem design relating toeffective Itisdemonstrated that under properly defined conditions doubling utilization ofmultiprocessor systems andsignificant software thenumber ofCPU's inasystem canmorethandouble throughput. andlanguage constructs suchasmultitasking ofindividual Theeffectiveness ofmultitasking inimproving throughput isfound to jobs. befairly small ifthedegree ofmultiprogramming ranges frommodest Theprincipal results include a definition ofcircumstances, tohigh. whereincreasing thenumberofprocessors inasystemdispro- Index Terms-CPU, multiprocessing, multiprogramming, multitasking, portionately increases itsthroughput, andtheconclusion that queueing models, throughput. multitasking isaneffective waytoincrease throughput onlyif increasing thedegree ofmultiprogramming isnotfeasible.