RED PERFORMANCE EVALUATIONUSINGSTOCHASTICMODELLING AND FLUID-BASEDANALYSIS

RandomEarly Detection (RED)isapowerful mechanism used forqueue management.Manystudies showed that RED hasadvantages over Tail Drop(TD). However, mostofthese studies were based on simulation anddoesnotreflect a conclusive performance evaluation ofRED.Theneedfor an analytical evaluation that allows better understanding ofRED was addressed byseveral researchers. Theresult was several approaches, eachpartially characterizes RED performance. Thisworkaimstoprovide a better understanding oftheRED algorithm andtoquantify thebenefits andlimitations ofusing RED queue managementbyusingtwo analytic models, namely. stochastic-based model, usingqueuing theory and stochastic modeling andfluid-based model, usingstochastic diferential equations. Thefluid-based model was modified to incorporate smooth nonconforming traffic (e.g. UDP)as well as TCP.Themodel was verified using simulation results. Ouranalysis showed that REDoutperforms TDmostofthe time.Thestochastic basedmodelshowedthat RED removes thebias against bursty traffic andhelps control the queue size andthedelay. Ontheother hand, itincreases thevariability of thequeue sizeandhencethejitter. Theresults showedthat RED queue can handletheaddedUDP traffic while maintaining itsnormal operation tosome extent. However, the increase intheUDP traffic sharecreated oscillation inthe queue sizeandresulted ininstability. WhentheUDP traffic reached 500ofthetotal capacity the queue startsoscillating wildly which can causebuffer overflow andjitter. Theeffect of theUDPportion oftheoverall link capacity on theTCPtraffic intheRED queuewhere studied Itseems that ahighUDPrate will notstarvetheTCPtraffic under RED.Another observation was that for higher linkcapacities theRED performance becomes highly dependent on thesampling rate(a). Simulation confirms theaboveconclusions andmatches wellwiththe findings obtainedfrom analysis.