Optimal allocationofserversandprocessingtimeinaloadbalancingsystem

We considertheproblemofallocatingprocessingtimeinamulti-channelloadbalancingsystemby focusingonsystemswhereprocessingtimeshavedistributionscharacterizedbyhighvariability.Our objectiveistoreducecongestionbyroutingjobstoserversbasedontheirworkload.Specifically,we arrangeserversintwostationsinseries,andrequirethattheloadbebalancedbetweenthetwo stations.Allarrivalsjointhefirstservicecenterwheretheyreceiveamaximumof T unitsofservice. Arrivalswithservicerequirementsthatexceedthevalue T join thesecondstationwheretheyreceive their remainingservice.Foravarietyofheavytailservicetimedistributions,characterizedbyhigh variability,analyticalandnumericalcomparisonsshowthatourschemeprovidesbettersystem performancethanthe standardparallelmulti-servermodel in thesenseofreducingthemeandelayper customerwhenthetrafficintensityisnottoolow.Inparticular,wedeveloplowerboundsonthetraffic intensityandtheservicetimecoefficientofvariationbeyondwhichthebalancedseriessystem outperformstheparallelsystem.