Internet Engineering Task Force Mukul Goyal, Padmini Misra INTERNET-DRAFT Raj Jain draft-goyal-dpstdy-diffserv-00.txt The Ohio State University March, 1999 Expires: September, 1999 Effect of Number of Drop Precedences in Assured Forwarding Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. NOTE This document is not to be taken as a finished product. Some of the sections are rough and are included in order to obtain comments from the community that will benefit future iterations of this document. This is simply a step in the ongoing conversation about this document. Finally, all the authors of this draft do not necessarily agree with and/or advocate all the mechanisms outlined in this document. Abstract This informational draft presents a simulation study of the effect of having different number of drop precedences in an Assured Forwarding [juha1] traffic class. The simulations involve both TCP and UDP traffic. The results show that discrimination between UDP and TCP Goyal, Misra & Jain Drop Precedence Study [Page 1] draft-goyal-dpstdy-diffserv-00.txt- 2 - March, 1998 flows can be achieved with just 2 drop precedences. 1. Introduction The Assured Forwarding PHB [juha1] specifies 4 traffic classes with 3 drop precedence each to provide differentiated services to the customers. One of the reasons given for having 3 drop precedences is that in case of mixed traffic (both congestion-sensitive and congestion-insensitive) in the same traffic class, 3 drop precedences can be used to effectively control congestion-insensitive sources (e.g. UDP sources) from getting more than their fair share of network resources. In this study, we simulate a mixed traffic scenario with 0, 2, and 3 drop precedences. 2. Simulation Configuration and Parameters: In our simulations, we used the network configuration shown in Figure 1. It consists of 50 sources, labeled Src_1 through Src_50, arranged in 5 groups of 10 sources each. Of these, Src_10, Src_20, Src_30, Src_40 and Src_50 are UDP sources and the rest are TCP sources. All TCP sources implement slow start and the Reno version of fast retransmit and recovery algorithm. There are 5 access routers labeled as AR_1 through AR_5. These represent ingress to a DS domain for 5 customers belonging to same AF traffic class. These routers implement a drop precedence marker. This marker can be either Three Color Marker (TCM) as specified in [juha2] or a Rate Based Marker (RBM). In TCM, there are two token buckets with different bucket sizes called committed burst size (CBS) and excess burst size (EBS). The tokens are generated for the two buckets at a combined rate. This rate is called committed information rate (CIR). The tokens are added to the second bucket when the first bucket is full. The RBM is similar to TCM except that separate token generation rates (TGR0 and TGR1) are specified for the two buckets. Moreover, in RBM, token generation in two buckets is independent of each other. To get similar behavior with either TCM or RBM, we set CIR of TCM to be sum of token genration rates of two buckets in RBM (CIR = TGR0 + TGR1). Also, the same bucket sizes are set for both TCM and RBM. In all our simulations, the color markers work in the 'color aware' mode [juha2], meaning that packets pre-marked with a drop precedence can not use tokens of higher precedence buckets and that they cannot Goyal, Misra & Jain Drop Precedence Study [Page 2] draft-goyal-dpstdy-diffserv-00.txt- 3 - March, 1998 be upgraded to a higher precedence. However, a marker can degrade the drop precedence of a packet. Src_1 . \ . ----AR_1------------ . / | Src_10 | | Src_11 | . \ | Snk_1 . ----AR_2 -------- | / . . / | | / . Src_20 | | / . | | / . Src_21 | | / . . \ / . . ----AR_3-------- R_1 -----------------R_2 . . / \ . Src_30 | | \ . | | \ . Src_31 | | \ . . \ | | \ . . ----AR_4 -------- | \ . . / | Snk_50 Src_40 | | Src_41 | . \ | . ----AR_5 ----------- . / Src_50 DATA ----> <----Ack Src_i's are Senders Snk_i's are Receivers AR_i's and R_i's are Routers Figure 1: Simulation Configuration Goyal, Misra & Jain Drop Precedence Study [Page 3] draft-goyal-dpstdy-diffserv-00.txt- 4 - March, 1998 The traffic arriving at each of color markers consists of both UDP and TCP packets. If the number of drop precedences is 3, TCP packets are marked with dp0 by their sources whereas UDP packets are marked with dp2. If the number of drop precedences is 2, TCP packets are marked with dp0 and UDP packets are marked with dp1. In the configuration with n drop precedences, router R_1 implements RED drop policy [floy1] with n different drop probabilities. Here onwards, we call such a drop policy RED_n. To implement RED_n, we extended the RIO algorithm used in [ibanez] to have different thresholds and maximum drop probability for packets of precedence 0, 1, 2, and so on. As in [ibanez], to calculate RED average queue length for packets with precedence 'i', current queue occupancy of all packets of precedence 0 to 'i' is used. We have used NS simulator version 2.1b4a [NS] for these simulations. New code has been added to the simulator to implement TCM, RBM and RED_n. Values of various simulation parameters are shown in Table 1. Table 1: Simulation Parameters -------------------------------------------------------------------------- Simulation Time 100 seconds TCP Window 64 packets IP Packet Size 576 UDP Rate 128Kb, 1.28Mb Maximum queue size 60 packets (for all queues) RED_n parameters: Queue Weight 0.002 (for all queues) No D.P. 2 D.P. 3 D.P. Min Thresh(dp0) 20 20 20 Max Thresh(dp0) 40 40 40 Min Thresh(dp1) N/A 20 20 Max Thresh(dp1) N/A 40 40 Min Thresh(dp2) N/A N/A 20 Max Thresh(dp2) N/A N/A 40 Drop prob(dp0) 1/30 1/30 1/30 Drop prob(dp1) N/A 1/20 1/20 Drop prob(dp2) N/A N/A 1/10 Goyal, Misra & Jain Drop Precedence Study [Page 4] draft-goyal-dpstdy-diffserv-00.txt- 5 - March, 1998 (Table 1: Simulation Parameters: Continued) RBM Parameters : No D.P. 2 D.P. 3 D.P. TGR0 N/A 192000, 256000 64000, 128000 TGR1 N/A N/A 128000,128000 BucketSize dp0 N/A 64000 32000 BucketSize dp1 N/A N/A 32000 TCM Parameters : No D.P. 2 D.P. 3 D.P. CIR N/A 192000, 256000 192000, 256000 BucketSize dp0 N/A 64000 32000 BucketSize dp1 N/A N/A 32000 Link between Src_i's & AR_i's : Link Bandwidth : 10 Mbps Link Delay : 1 microsecond Drop Policy : DropTail Link between AR_i's & R_1 : From AR_i To AR_i Link Bandwidth : 1.5 Mbps 1.5 Mbps Link Delay : 5 microseconds 5 microseconds Drop Policy : DropTail(with marker) DropTail Link between R_1 & R_2 : From R_1 To R_2 Link Bandwidth : 1.5 Mbps 1.5 Mbps Link Delay : 30 miliseconds 30 miliseconds Drop Policy : RED_n DropTail Link between R_2 & Snk_i's : Link Bandwidth : 1.5 Mbps Link Delay : 5 microseconds Drop Policy : Droptail -------------------------------------------------------------------------- 4. Simulation Results Tables 2 and 3 summarize the simulation results for both TCM and RBM. In each case, we compare the TCP and UDP throughputs and fairness for different number of drop precedences. Fairness is computed using the following formula [jain]: Goyal, Misra & Jain Drop Precedence Study [Page 5] draft-goyal-dpstdy-diffserv-00.txt- 6 - March, 1998 Fairness = [{sum(xi)}**2]/[n*sum(xi**2)] Where xi is the throughput of the ith flow and n is the number of flows. Notice that, in case of no drop precedence, average TCP throughput and fairness values are very low. However, with 2 drop precedences, TCP throughputs and fairness values improve radically. Having 3 drop precedence does not help to improve TCP throughputs or fairness. However, as the token generation rate is increased (CIR for TCM: from 192 to 256kbps, TGR0 for RBM:from 64 to 128kbps), both TCP throughputs and fairness improves and UDP throughputs decrease. This happens irrespective of whether there are 2 drop precedences or 3. In all cases, results achieved with 2 drop precedences are almost same as with 3 drop precedences. Table 2 ( For TCM [juha2] as Color Marker) ------------------------------------------------------------------------ UDP # of CIR Max TCP Min TCP Avg TCP Max UDP Min UDP Fairness Rate DP's (kbps) Thruput Thruput Thruput Thruput Thruput (kbps) (kbps) (kbps) (kbps) (kbps) ----------------------------------------------------------------------- 1.28M No DP N/A 0.64 0.05 0.21 299.40 296.92 .101 1.28M 2 192 26.06 14.27 19.90 123.74 118.81 .494 1.28M 3 192 24.67 12.11 20.15 122.54 115.91 .507 1.28M 2 256 26.52 15.97 22.81 98.14 91.61 .658 1.28M 3 256 25.83 20.39 22.84 98.79 89.44 .661 128k No DP N/A 28.36 10.17 21.36 108.69 106.06 .570 128k 2 192 31.76 7.46 25.04 74.62 73.42 .796 128k 3 192 30.84 13.81 24.87 76.23 74.07 .788 128k 2 256 32.27 16.76 27.91 49.03 46.95 .951 128k 3 256 33.28 21.64 27.96 48.70 46.95 .957 ----------------------------------------------------------------------- Goyal, Misra & Jain Drop Precedence Study [Page 6] draft-goyal-dpstdy-diffserv-00.txt- 7 - March, 1998 Table 3 (For RBM as Color Marker) (For 2 DP: TGR = TGR0; For 3 DP: TGR = TGR0{64,128kbps} + TGR1{128kbps}) ------------------------------------------------------------------------ UDP # of TGR Max TCP Min TCP Avg TCP Max UDP Min UDP Fairness Rate DP's (kbps) Thruput Thruput Thruput Thruput Thruput (kbps) (kbps) (kbps) (kbps) (kbps) ----------------------------------------------------------------------- 1.28M No DP N/A 0.64 0.05 0.21 299.40 296.92 .101 1.28M 2 192 24.81 11.00 19.89 126.82 117.57 .493 1.28M 3 192 23.94 10.96 18.54 135.43 130.14 .431 1.28M 2 256 26.38 19.29 22.76 97.82 93.13 .656 1.28M 3 256 25.83 13.35 23.08 95.01 89.44 .675 128k No DP N/A 28.36 10.17 21.36 108.69 106.06 .570 128k 2 192 34.66 19.33 25.04 74.71 73.56 .799 128k 3 192 29.51 20.07 24.92 76.88 74.11 .794 128k 2 256 33.79 19.84 28.07 48.38 46.26 .960 128k 3 256 34.57 22.42 28.18 46.40 45.48 .964 ----------------------------------------------------------------------- 6. Summary Simulations presented here show that having 3 drop precedences in each AF traffic class does not give any better results than 2 drop precedences. 7. References [juha1] J. Heinanen, et al., Assured Forwarding PHB Group. Internet draft draft-ietf-diffserv-af-05.txt, February 1999. [juha2] J. Heinanen, et al., A Three Color Marker. Internet draft draft-heinanen-diffserv-tcm-00.txt, February 1999. [floy1] S. Floyd, V. Jacobson, "Random Early Detection Gateways for Congestion Avoidance," IEEE/ACM Transactions on Networking, August 1993. [ibanez] J. Ibanez, K. Nichols, "Preliminary Simulation Evaluation of an Assured Service" , August, 1998. Goyal, Misra & Jain Drop Precedence Study [Page 7] draft-goyal-dpstdy-diffserv-00.txt- 8 - March, 1998 [jain] Raj Jain, "The Art of Computer Systems Performance Analysis," John Wiley and Sons Inc., 1991. [RFC2475] S. Blake, et al., An Architecture for Differentiated Services. RFC 2475, December 1998. [NS] NS simulator, Version 2.1 Available from http://www- mash.cs.berkeley.edu/ns Authors's address Mukul Goyal, Padmini Misra, Raj Jain Computer and Information Science Dept, The Ohio State University, 2015, Neil Avenue,Room 395, Columbus, OH 43201 Phone: +1 (614) 292-3989 Fax: +1 (614) 292 2911 Email: {mukul,misra,jain}@cse.wustl.edu This internet draft expires on September 1999 Goyal, Misra & Jain Drop Precedence Study [Page 8]