Mo deling TCP Throughput: A Simple Mo del and its Empirical V alidation � Jitendra P adh y e Victor Firoiu Don T o wsley Jim Kurose f jitu, v�roiu, to wsley , kurose g @cs.umass.edu Departmen t of Computer Science Univ ersit y of Massac h usetts Amherst, MA 01003 USA Abstract In this pap er w e dev elop a simple analytic c haracteriza- tion of the steady state throughput of a bulk transfer TCP In this pap er w e dev elop a simple analytic c haracterization �o w (i.e., a �o w with a large amoun t of data to send, suc h of the steady state throughput, as a function of loss rate as FTP transfers) as a function of loss rate and round trip and round trip time for a bulk transfer TCP �o w, i.e., a time. Unlik e the recen t w ork of [6 , 7, 10 ], our mo del captures �o w with an unlimited amoun t of data to send. Unlik e the not only the b eha vior of TCP's fast retransmit mec hanism mo dels in [6 , 7, 10 ], our mo del captures not only the b e- (whic h is also considered in [6 , 7, 10 ]) but also the e�ect ha vior of TCP's fast retransmit mec hanism (whic h is also of TCP's timeout mec hanism on throughput. The measure- considered in [6 , 7, 10]) but also the e�ect of TCP's timeout men ts w e presen t in Section 3 indicate that this latter b eha v- mec hanism on throughput. Our measuremen ts suggest that ior is imp ortan t from a mo deling p ersp ectiv e, as w e observ e this latter b eha vior is imp ortan t from a mo deling p ersp ec- more timeout ev en ts than fast retransmit ev en ts in almost tiv e, as almost all of our TCP traces con tained more time- all of our TCP traces. Another imp ortan t di�erence b et w een out ev en ts than fast retransmit ev en ts. Our measuremen ts ours and previous w ork is the abilit y of our mo del to accu- demonstrate that our mo del is able to more accurately pre- rately predict throughput o v er a signi�can tly wider range dict TCP throughput and is accurate o v er a wider range of of loss rates than b efore; measuremen ts presen ted in [7 ] as loss rates. w ell the measuremen ts presen ted in this pap er, indicate that this to o is imp ortan t. W e also explicitly mo del the e�ects of small receiv er-side windo ws. By comparing our mo del's 1 Intro duction predictions with a n um b er of TCP measuremen ts made b e- t w een v arious In ternet hosts, w e demonstrate that our mo del A signi�can t amoun t of to da y's In ternet tra�c, including is able to more accurately predict TCP throughput and is WWW (HTTP), �le transfer (FTP), email (SMTP), and re- able to do so o v er a wider range of loss rates. mote access (T elnet) tra�c, is carried b y the TCP transp ort The remainder of the pap er is organized as follo ws. In proto col [18 ]. TCP together with UDP form the v ery core Section 2 w e describ e our mo del of TCP congestion con trol of to da y's In ternet transp ort la y er. T raditionally , sim ula- in detail and deriv e a new analytic c haracterization of TCP tion and implemen tation/measuremen t ha v e b een the to ols throughput as a function of loss rate and a v erage round trip of c hoice for examining the p erformance of v arious asp ects of time. In Section 3 w e compare the predictions of our mo del TCP . Recen tly , ho w ev er, sev eral e�orts ha v e b een directed with a set of measured TCP �o ws o v er the In ternet, ha ving at analytically c haracterizing the throughput of TCP's con- as their endp oin ts sites in b oth United States and Europ e. gestion con trol mec hanism, as a function of pac k et loss and Section 4 discusses the assumptions underlying the mo del round trip dela y [6, 10 , 7]. One reason for this recen t in- and a n um b er of related issues in more detail. Section 5 terest is that a simple quan titativ e c haracterization of TCP concludes the pap er. throughput under giv en op erating conditions o�ers the p os- sibilit y of de�ning a \fair share" or \TCP-friendly" [6 ] through- put for a non-TCP �o w that in teracts with a TCP connec- 2 A Mo del fo r TCP Congestion Control tion. Indeed, this notion has already b een adopted in the design and dev elopmen t of sev eral m ulticast congestion con- In this section w e dev elop a sto c hastic mo del of TCP conges- trol proto cols [19 , 20 ]. tion con trol that yields a relativ ely simple analytic expres- sion for the throughput of a saturated TCP sender, i.e., a � This material is based up on w ork supp orted b y the National �o w with an unlimited amoun t of data to send, as a function Science F oundation under gran ts NCR-95-08274, NCR-95-23807 and of loss rate and a v erage round trip time (R TT). CD A-95-02639. An y opinions, �ndings, and conclusions or recommen- TCP is a proto col that can exhibit complex b eha vior, dations expressed in this material are those of the authors and do not necessarily re�ect the views of the National Science F oundation. esp ecially when considered in the con text of the curren t In- ternet, where the tra�c conditions themselv es can b e quite complicated and subtle [14 ]. In this pap er, w e fo cus our at- ten tion on the congestion a v oidance b eha vior of TCP and its impact on throughput, taking in to accoun t the dep en- dence of congestion a v oidance on A CK b eha vior, the manner in whic h pac k et loss is inferred (e.g., whether b y duplicate A CK detection and fast retransmit, or b y timeout), limited
Recommend
More recommend
