Internet Engineering Task Force Mukul Goyal, Arian Durresi INTERNET-DRAFT Raj Jain draft-goyal-dpstdy-diffserv-01.txt The Ohio State University June, 1999 Expires: December, 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 to compare the performance of two and three drop precedence levels in an Assured Goyal, Durresi & Jain Drop Precedence Study [Page 1] draft-goyal-dpstdy-diffserv-01.txt- 2 - June, 1999 Forwarding traffic class for mixed TCP/UDP traffic. This is a revised and comprehensive version of our previous study presented at the March 99 IETF Meeting at Minneapolis. For this study, we consulted the principal authors of the Assured Forwarding Class to determine the parameter set and configurations under which it would be beneficial to have 3-drop preferences. The results presented here are based on 1296 simulations (different parameter/traffic pattern combinations). The results of the simulation study suggest that optimal number of drop precedence levels in a traffic class depends on the traffic load and amount of reserved traffic. In a mixed TCP- UDP traffic scenario, best performance can be achieved with either 2 or 3 levels of drop precedence if most of the TCP packets have 'better' drop precedence than excess UDP packets. 1. Introduction The Assured Forwarding PHB [juha] specifies 4 traffic classes with 3 colors (drop precedence levels) each, in order to provide differentiated services to the customers on IP networks. One of the specified reasons for having 3 levels of drop precedence is that in case of mixed TCP-UDP traffic, 3 levels of drop precedence can be used to enforce a fair distribution of network resources among congestion sensitive and insensitive flows. In this study, we perform a wide range of simulations with 2 and 3 levels of drop precedence and compare their best performance for different traffic scenarios. 2. Simulation Configuration and Parameters: The network configuration used in the simulation study is shown in Figure 1. It consists of 10 customers, labeled CST_1 through CST_10. All customers except CST_10 carry an aggregated traffic coming from 5 Reno TCP sources. CST_10 gets its traffic from a single UDP source sending data at a rate of 1.28 Mbps. All the customers belong to the same AF traffic class. Each customer is assigned the same amount of reserved (green) traffic rate (CIR). All the traffic passes through a traffic conditioner at the customer's site before being forwarded to router R_1. The traffic conditioners associated with CST_1 through CST_9 consist of two token buckets and mark packets on the basis of reserved traffic rate and a yellow traffic rate. Yellow traffic rate for customers CST_1 through CST_9 is same. Yellow traffic rate associated with CST_10 equals zero. Thus, TCP packets coming from CST_1 through CST_9 are marked green, yellow or red but UDP packets coming from CST_10 are marked either green or red. All simulation parameters are listed in Tables 1,2 and 3. Goyal, Durresi & Jain Drop Precedence Study [Page 2] draft-goyal-dpstdy-diffserv-01.txt- 3 - June, 1999 TCP_1 . \ . ----CST_1----------- . / | TCP_5 | | TCP_6 | . \ | . ----CST_2-------- | . / | | TCP_10 | | | | SNK_1 TCP_11 | | / . . \ | | / . . ----CST_3----- | | / . . / | | | / . TCP_15 | | | / . R_1 -----------------R_2 . . . . . . | | \ . . . | | \ . . . | | \ . . . | | \ . | | \ . TCP_40 | | SNK_46 . \ | | . ----CST_9-------- | . / | TCP_45 | | | | UDP_46---CST_10---------- DATA ----> <----Ack TCP_i's are TCP Senders UDP_46 is the only UDP Sender SNK_i's are Receivers R_1 and R_2 are Routers CST_i's are customer sites with Three Color Markers Figure 1: Simulation Configuration As shown in Table 2, the simulations are repeated for four different values of total reserved traffic load with the purpose of observing Goyal, Durresi & Jain Drop Precedence Study [Page 3] draft-goyal-dpstdy-diffserv-01.txt- 4 - June, 1999 change in the performance achieved with 2 and 3 colors with increase in reserved traffic load. Each customer is assigned a green packet rate equal to 1/10th of total reserved traffic load. Customers CST_1 through CST_9 carrying TCP traffic are also assigned a yellow packet rate. This rate becomes zero for two color simulations. All the simulations are repeated for 2 values of yellow packet rates per customer - 12800 bps and 128000bps. With these values, the combined yellow traffic rate equals 8.5% and 85% of the link capacity. The yellow traffic rate is used only for generating tokens in a bucket and does not represent the actual load of yellow packets in the network. Router R_1 implements an RED queue [floyd] that has different drop thresholds and drop probabilities for packets of different colors. The average queue length for RED queue can be calculated based on total buffer occupancy and used for packets of all the colors. Alternately, average queue length can be calculated differently for different colors by counting only packets of equal or lower colors [ibanez]. Thus average queue for red packets will be based on buffer occupancy of red,yellow and green packets. Average queue for yellow packets will be based on buffer occupancy of yellow and green packets and that for green packets will be based just on the number of green packets in the queue. We conducted simulations with both ways of calculating the average queue. The observed behavior of 2 and 3 colors was similar in both cases. In order to save space, we show results of only the first case where the average queue length is calculated based on total buffer occupancy and applies to packets of all colors. We have repeated our simulations for a wide range of drop thresholds and maximum drop probabilities in order to determine close to best performance achievable with 2 and 3 colors in a traffic class. The RED parameters used in these simulations are listed in Table 3. We have used NS simulator version 2.1b4a [NS] for these simulations. The code has been modified to implement the traffic conditioner and multi-color RED (RED_n). Goyal, Durresi & Jain Drop Precedence Study [Page 4] draft-goyal-dpstdy-diffserv-01.txt- 5 - June, 1999 Table 1: Simulation Parameters Common for 2 and 3 Color Simulations -------------------------------------------------------------------------- Simulation Time : 100 seconds TCP Window : 64 packets IP Packet Size : 576 UDP Rate : 1.28Mb Maximum queue size (for all queues - RED/DropTail) : 60 packets Link between UDP_46/TCP__i's & CST__i's : Link Bandwidth : 10 Mbps Link Delay : 1 microsecond Drop Policy : DropTail Link between CST_i's & R_1 : From CST_i To CST_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_1 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 -------------------------------------------------------------------------- Goyal, Durresi & Jain Drop Precedence Study [Page 5] draft-goyal-dpstdy-diffserv-01.txt- 6 - June, 1999 Table 2: Traffic Conditioner Parameters for 2 and 3 Color Simulations -------------------------------------------------------------------------- Common Parameters: ----------------- Reserved(Green) Traffic (Token Generation Rate for Green Bucket-CIR) : 12800,25600,38400,76800bps Per Customer (8.5%,17%,25.6%,51.2% of Available Bandwidth) Green Bucket Size : 8 kB Traffic Conditioner Mode: Color Aware 2/3 Color-Specifc Parameters: ---------------------------- 2-Color Simulations 3-Color Simulations Yellow Traffic (Token Generation Rate for Yellow Bucket) : 0 12800 bps per customer (= 8.5% of link bandwidth) 128000 bps per customer (= 85% of link bandwidth) Yellow Bucket Size : 0 8 kB -------------------------------------------------------------------------- Goyal, Durresi & Jain Drop Precedence Study [Page 6] draft-goyal-dpstdy-diffserv-01.txt- 7 - June, 1999 Table 3: RED_n Parameters for 2 and 3 Color Simulations -------------------------------------------------------------------------- Common Parameters: ----------------- Queue Weight for Green Packets : 0.002 Queue Weight for Yellow Packets : 0.002 Queue Weight for Red Packets : 0.002 2/3 Color Specific Parameters: ----------------------------- 2-Color Simulations 3-Color Simulations RED_n Thresholds {0/60, - ,0/60} {0/60,0/40,0/20} {minth_green/maxth_green, {0/60, - ,0/40} {0/60,0/60,0/60} minth_yellow/maxth_yellow, {0/60, - ,0/20} {40/60,20/60,0/60} minth_red/maxth_red} : {0/60, - ,0/5} {40/60,20/60,0/20} {0/60, - ,0/10} {40/60,20/40,0/20} {40/60, - ,0/20} {40/60, - ,0/60} {40/60, - ,0/40} {40/60, - ,20/40} {40/60, - ,20/60} {40/60, - ,0/5} {40/60, - ,0/10} Max Drop Probability {green,yellow,red} : {1,1,1} {1,1,1} {0.5,1,1} {0.5,1,1} {0.5,1,0.5} {0.5,0.5,1} {0.1,1,0.5} {0.5,0.5,0.5} {0.1,1,1} {0.1,0.5,0.5} {0.1,1,0.1} {0.1,0.5,1} {0.1,1,1} {0.1,0.1,0.1} {0.1,0.1,0.5} -------------------------------------------------------------------------- 4. Performance Metrics In this simulation study, our goal is to compare the best performance achieved by 2 and 3 colors. We have used two criteria for such a performance evaluation - fairness index and UDP throughput. For n flows sharing a link, the fairness index can be computed using the following formula [jain]: Fairness = [{sum(xi)}**2]/[n*sum(xi**2)] Goyal, Durresi & Jain Drop Precedence Study [Page 7] draft-goyal-dpstdy-diffserv-01.txt- 8 - June, 1999 Where xi is the throughput of the ith flow. According to fairness criterion, both TCP and UDP customers should get a fair allocation of network resources. In other words, allocation of network resources should not be biased by congestion control mechanism used/not used by a flow. The other criterion for evaluating performance is how much a scheme can restrict UDP throughput close to its reserved rate. According to this criterion, the best scheme will restrict UDP throughput to its reserved rate while achieving fair allocation of excess resources among TCP flows. In our calculations with this criterion, we ignore the fairness among TCP sources as long as it is reasonably high. Note that a scheme that performs well by the fairness test may not pass UDP throughput criterion if the fair share of a UDP source is much more than its reserved bandwidth. 5. Simulation Results This simulation study consists of a total of 1296 simulations. In order to save space, we present only the most significant results. Tables 4 and 5 show the best results obtained by the two performance criteria for the case when total reserved traffic is around 8.5% of link capacity. For such a traffic scenario, 3 colors clearly result in better performance than 2 colors by both performance criteria. Also in both cases, best performance is achieved with high yellow packet rate. With low level of reserved traffic and 2 colors, excess TCP and UDP packets have the same color. Congestion insensitive UDP flow is able to grab the fair share of TCP flows that reduce their data rate in response to packet drop. This results in both very low fairness and very high UDP throughput. However, with 3 colors and sufficient yelow packet rate for TCP flows, there is clear demarkation between excess TCP and UDP packets. Harsh treatment applied to red UDP packets does not affect TCP flows as most of excess TCP packets are marked yellow. Thus, with 3 colors in this case, one can set RED parameters appropriately to achieve either high fairness or restrict UDP throughput. Goyal, Durresi & Jain Drop Precedence Study [Page 8] draft-goyal-dpstdy-diffserv-01.txt- 9 - June, 1999 Table 4: Best 20 Fairness Results for Green Rate = 8.5% of 1.5Mbps (12800bps per customer) (+ Best 2-Color Result) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 3 0.5 0.5 1 40/60 20/60 0/60 85% 175012 147111 1499012 0.9953 3 0.1 0.5 1 40/60 20/60 0/60 85% 175012 147111 1499012 0.9953 3 0.1 0.1 0.1 0/60 0/40 0/20 85% 90685 156436 1498605 0.9745 3 1 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.5 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.1 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.5 0.5 0.5 40/60 20/40 0/20 85% 68060 158836 1497580 0.9672 3 0.1 0.5 0.5 40/60 20/40 0/20 85% 68060 158836 1497580 0.9672 3 1 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.5 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.1 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.5 0.5 1 40/60 20/40 0/20 85% 56771 160116 1497811 0.9569 3 0.1 0.5 1 40/60 20/40 0/20 85% 56771 160116 1497811 0.9569 3 0.1 0.1 0.5 0/60 0/40 0/20 85% 50043 160978 1498843 0.9507 3 0.1 0.1 0.1 40/60 20/60 0/20 8.5% 253348 138462 1499508 0.9498 3 0.1 0.1 0.1 40/60 20/40 0/20 8.5% 260076 137733 1499676 0.9434 3 0.5 0.5 0.5 40/60 20/60 0/20 85% 24561 163787 1498641 0.9268 3 0.1 0.5 0.5 40/60 20/60 0/20 85% 24561 163787 1498641 0.9268 3 0.5 0.5 1 40/60 20/60 0/20 85% 21842 164044 1498242 0.9242 3 0.1 0.5 1 40/60 20/60 0/20 85% 21842 164044 1498242 0.9242 2 0.1 1 0.1 40/60 - 0/20 0% 626181 97076 1499861 0.4716 ------------------------------------------------------------------------------ Tables 6 and 7 show the best performances achieved for both performance criteria when total amount of reserved traffic is 17% of the link capacity. According to fairness criterion, in this case, 3 colors do not perform substantially better than 2 colors. Infact, best fairness performance is achieved by a configuration where red packets get the same treatment as the yellow ones. In this case, reserved traffic is large enough to ensure that most of the TCP packets will be green. This observation is confirmed by same results obtained with different yellow packet rates (8.5% and 85%) and actual color profile of TCP packets in this configuration (not shown here). We expect simulation results to change if we increase the amount of TCP traffic entering the network so that a reserved rate of 17% is no longer sufficient to keep most of the packets green. Also Table 7 shows that it is difficult to control the UDP throughput in this case with just 2 colors. This is because in order to control UDP throughput, treatment given to red packets needs to be very harsh (drop thresholds 0/5). Though the reserved rate is quite high, many of the TCP packets still overflow the bucket and get marked as red. Goyal, Durresi & Jain Drop Precedence Study [Page 9] draft-goyal-dpstdy-diffserv-01.txt- 10 - June, 1999 Table 5: Best 20 UDP Control Results for Green Rate = 8.5% of 1.5Mbps (12800bps per customer) (+ Best 2-Color Result) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 3 0.1 0.1 0.5 40/60 20/60 0/20 85% 14239 164880 1498159 0.9151 3 0.1 0.1 0.5 40/60 20/40 0/20 85% 15160 164862 1498920 0.9173 3 0.1 0.1 0.1 40/60 20/60 0/20 85% 16082 164791 1499202 0.9173 3 0.1 0.1 0.1 40/60 20/40 0/20 85% 16727 164631 1498407 0.9185 3 0.5 0.5 1 40/60 20/60 0/20 85% 21842 164044 1498242 0.9242 3 0.1 0.5 1 40/60 20/60 0/20 85% 21842 164044 1498242 0.9242 3 0.5 0.5 0.5 40/60 20/60 0/20 85% 24561 163787 1498641 0.9268 3 0.1 0.5 0.5 40/60 20/60 0/20 85% 24561 163787 1498641 0.9268 3 0.1 0.1 0.5 0/60 0/40 0/20 85% 50043 160978 1498843 0.9507 3 1 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.5 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.1 1 1 40/60 20/60 0/20 85% 56771 160116 1497811 0.9569 3 0.5 0.5 1 40/60 20/40 0/20 85% 56771 160116 1497811 0.9569 3 0.1 0.5 1 40/60 20/40 0/20 85% 56771 160116 1497811 0.9569 3 0.5 0.5 0.5 40/60 20/40 0/20 85% 68060 158836 1497580 0.9672 3 0.1 0.5 0.5 40/60 20/40 0/20 85% 68060 158836 1497580 0.9672 3 1 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.5 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.1 1 1 40/60 20/40 0/20 85% 73359 158418 1499119 0.9705 3 0.1 0.1 0.1 0/60 0/40 0/20 85% 90685 156436 1498605 0.9745 2 0.1 1 0.1 40/60 - 0/20 0% 626181 97076 1499861 0.4716 ------------------------------------------------------------------------------ Thus harsh treatment applied to red UDP packets also affects TCP flows as they reduce their data rate in response to packet drops. Reduced data rate of TCP flows enables UDP to grab much more than its fair share of network resources even with very harsh RED parameters. However, with 3 colors and appropriate RED parameters, the UDP throughput can be restricted to a rate very close to its reserved rate. Goyal, Durresi & Jain Drop Precedence Study [Page 10] draft-goyal-dpstdy-diffserv-01.txt- 11 - June, 1999 Table 6: Best 20 Fairness Results for Green Rate = 17% of 1.5Mbps (25600bps per customer) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 3 0.1 0.5 0.5 0/60 0/60 0/60 8.5% 202291 144018 1498451 0.9786 3 0.1 0.5 0.5 0/60 0/60 0/60 85% 202291 144018 1498451 0.9786 3 0.1 0.5 0.5 40/60 20/60 0/60 8.5% 202291 144018 1498451 0.9786 3 0.1 0.5 0.5 40/60 20/60 0/60 85% 202291 144018 1498451 0.9786 3 0.1 0.1 0.5 0/60 0/60 0/60 8.5% 186301 145769 1498221 0.9784 3 0.1 0.1 0.5 0/60 0/60 0/60 85% 186301 145769 1498221 0.9784 3 0.1 0.1 0.5 40/60 20/60 0/60 8.5% 186301 145769 1498221 0.9784 3 0.1 0.1 0.5 40/60 20/60 0/60 85% 186301 145769 1498221 0.9784 2 0.1 1 0.1 0/60 - 0/5 0% 245192 139227 1498232 0.9568 2 0.1 1 0.1 40/60 - 0/5 0% 245192 139227 1498232 0.9568 2 0.1 1 0.1 0/60 - 0/10 0% 246804 139129 1498964 0.9555 2 0.1 1 0.1 40/60 - 0/10 0% 246804 139129 1498964 0.9555 2 0.1 1 0.5 0/60 - 0/5 0% 252242 138373 1497602 0.9505 2 0.1 1 0.5 40/60 - 0/5 0% 252242 138373 1497602 0.9505 2 0.1 1 0.5 0/60 - 0/10 0% 255882 138133 1499082 0.9473 2 0.1 1 0.5 40/60 - 0/10 0% 255882 138133 1499082 0.9473 2 0.1 1 1 0/60 - 0/10 0% 256067 138018 1498227 0.947 2 0.1 1 1 40/60 - 0/10 0% 256067 138018 1498227 0.947 2 0.1 1 0.1 0/60 - 0/20 0% 256389 138133 1499589 0.9466 2 0.1 1 0.1 40/60 - 0/20 0% 256389 138133 1499589 0.9466 ------------------------------------------------------------------------------ Table 7: Best 20 UDP Control Results for Green Rate = 17% of 1.5Mbps (25600bps per customer) (+ Best 2 Color Result) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 3 0.1 0.5 1 40/60 20/60 0/20 8.5% 25759 163636 1498479 0.9112 3 0.1 0.5 1 40/60 20/60 0/20 85% 25759 163636 1498479 0.9112 3 0.1 1 1 40/60 20/60 0/20 8.5% 25989 163582 1498229 0.9134 3 0.1 1 1 40/60 20/60 0/20 85% 25989 163582 1498229 0.9134 3 0.1 0.1 0.5 40/60 20/60 0/20 8.5% 26220 163556 1498220 0.9205 3 0.1 0.1 0.5 40/60 20/60 0/20 85% 26220 163556 1498220 0.9205 3 0.5 0.5 0.5 40/60 20/60 0/20 8.5% 26266 163538 1498106 0.9182 3 0.5 0.5 0.5 40/60 20/60 0/20 85% 26266 163538 1498106 0.9182 3 0.5 0.5 1 40/60 20/60 0/20 8.5% 26266 163538 1498106 0.9182 3 0.5 0.5 1 40/60 20/60 0/20 85% 26266 163538 1498106 0.9182 3 0.1 0.5 1 0/60 0/40 0/20 8.5% 26588 163600 1498988 0.9269 3 0.1 0.5 1 0/60 0/40 0/20 85% 26588 163600 1498988 0.9269 3 0.1 0.5 1 40/60 20/40 0/20 8.5% 26588 163600 1498988 0.9269 3 0.1 0.5 1 40/60 20/40 0/20 85% 26588 163600 1498988 0.9269 Goyal, Durresi & Jain Drop Precedence Study [Page 11] draft-goyal-dpstdy-diffserv-01.txt- 12 - June, 1999 3 0.1 1 1 0/60 0/40 0/20 8.5% 26634 163564 1498714 0.9285 3 0.1 1 1 0/60 0/40 0/20 85% 26634 163564 1498714 0.9285 3 0.1 1 1 40/60 20/40 0/20 8.5% 26634 163564 1498714 0.9285 3 0.1 1 1 40/60 20/40 0/20 85% 26634 163564 1498714 0.9285 3 1 1 1 40/60 20/60 0/20 8.5% 26726 163227 1495766 0.9218 3 1 1 1 40/60 20/60 0/20 85% 26726 163227 1495766 0.9218 2 0.1 1 0.1 40/60 - 0/5 0% 245192 139227 1498232 0.9568 ------------------------------------------------------------------------------ Tables 8 and 9 show the best performing simulation configurations when the total reserved traffic is around 25.6% of link capacity. In this case, reserved traffic rate is high enough to keep most of the TCP packets green. Thus a clear demarkation between TCP and excess UDP packets can be achieved just with two colors. Now, by approriately setting the RED parameters, one can achieve either good fairness or restrict the UDP flow close to its reserved rate or both. However, as in case of 17% reserved traffic, with substantially increased TCP traffic, 2 colors may fail to maintain good performance. Notice that, with harsh drop thresholds for red packets (0/5), UDP throughput actually falls short of its reserved rate. This can be attributed to loss of almost all of excess red and a few green UDP packets. Table 8: Best 20 Fairness Results for Green Rate = 25.6% of 1.5Mbps (38400bps per customer) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 2 0.5 1 1 0/60 - 0/5 0% 205194 143538 1497034 0.9809 2 0.5 1 1 40/60 - 0/5 0% 205194 143538 1497034 0.9809 2 0.1 1 0.1 0/60 - 0/10 0% 86630 156907 1498790 0.9775 2 0.1 1 0.5 0/60 - 0/10 0% 79304 157724 1498824 0.9738 2 0.1 1 0.5 40/60 - 0/10 0% 79304 157724 1498824 0.9738 2 0.1 1 1 0/60 - 0/10 0% 76124 158027 1498364 0.9731 2 0.1 1 1 40/60 - 0/10 0% 76124 158027 1498364 0.9731 2 0.1 1 0.5 0/60 - 0/5 0% 74051 158124 1497171 0.9714 2 0.1 1 0.5 40/60 - 0/5 0% 74051 158124 1497171 0.9714 2 0.1 1 1 0/60 - 0/5 0% 74051 158124 1497171 0.9714 2 0.1 1 1 40/60 - 0/5 0% 74051 158124 1497171 0.9714 2 0.1 1 0.1 0/60 - 0/5 0% 72714 158516 1499354 0.9704 2 1 1 1 0/60 - 0/20 0% 227912 141227 1498952 0.9692 2 1 1 1 40/60 - 0/20 0% 227912 141227 1498952 0.9692 3 0.1 0.5 0.5 0/60 0/60 0/60 8.5% 202199 144053 1498679 0.9644 3 0.1 0.5 0.5 0/60 0/60 0/60 85% 202199 144053 1498679 0.9644 3 0.1 0.5 0.5 40/60 20/60 0/60 8.5% 202199 144053 1498679 0.9644 3 0.1 0.5 0.5 40/60 20/60 0/60 85% 202199 144053 1498679 0.9644 2 0.5 1 0.5 0/60 - 0/20 0% 61839 159716 1499279 0.9625 Goyal, Durresi & Jain Drop Precedence Study [Page 12] draft-goyal-dpstdy-diffserv-01.txt- 13 - June, 1999 2 0.5 1 0.5 40/60 - 0/20 0% 61839 159716 1499279 0.9625 ------------------------------------------------------------------------------ Table 9: Best 20 UDP Control Results for Green Rate = 25.6% of 1.5Mbps (38400bps per customer) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 2 0.1 1 0.1 40/60 - 0/5 0% 37647 162284 1498207 0.9408 3 1 1 1 40/60 20/60 0/20 8.5% 38431 162178 1498031 0.9375 3 1 1 1 40/60 20/60 0/20 85% 38431 162178 1498031 0.9375 3 0.5 0.5 0.5 40/60 20/60 0/20 8.5% 38569 162231 1498649 0.9365 3 0.5 0.5 0.5 40/60 20/60 0/20 85% 38569 162231 1498649 0.9365 3 0.5 0.5 1 40/60 20/60 0/20 8.5% 38569 162231 1498649 0.9365 3 0.5 0.5 1 40/60 20/60 0/20 85% 38569 162231 1498649 0.9365 2 0.1 1 0.1 40/60 - 0/10 0% 38615 162213 1498535 0.9411 3 0.1 0.5 1 40/60 20/60 0/20 8.5% 38661 162187 1498341 0.9226 3 0.1 0.5 1 40/60 20/60 0/20 85% 38661 162187 1498341 0.9226 3 1 1 1 0/60 0/40 0/20 8.5% 38707 162196 1498467 0.942 3 1 1 1 0/60 0/40 0/20 85% 38707 162196 1498467 0.942 3 1 1 1 40/60 20/40 0/20 8.5% 38707 162196 1498467 0.942 3 1 1 1 40/60 20/40 0/20 85% 38707 162196 1498467 0.942 3 0.1 1 1 0/60 0/40 0/20 8.5% 38845 162204 1498685 0.9414 3 0.1 1 1 0/60 0/40 0/20 85% 38845 162204 1498685 0.9414 3 0.1 1 1 40/60 20/40 0/20 8.5% 38845 162204 1498685 0.9414 3 0.1 1 1 40/60 20/40 0/20 85% 38845 162204 1498685 0.9414 3 0.5 1 1 40/60 20/60 0/20 8.5% 38845 162133 1498045 0.9327 3 0.5 1 1 40/60 20/60 0/20 85% 38845 162133 1498045 0.9327 ------------------------------------------------------------------------------ Tables 10 and 11 show the best performing simulation configurations when reserved traffic is 51.2% of link bandwidth. The results follow the same trends as in case of 25.6% reserved traffic. Again, reserved packet rate is large enough to clearly demarkate TCP packets from excess UDP packets. For 3 color simulations, both yellow packet rates (8.5% and 85%) give same results as not many TCP packets need to get demoted to yellow color. Very large number of green packets means that red packets get very harsh treatment even with not so harsh drop thresholds. With either 2 or 3 colors, RED parameters can be set appropriately in order to achieve good fairness or restrict UDP throughput or both. Table 10: Best 20 Fairness Results for Green Rate = 51.2% of 1.5Mbps (76800bps per customer) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total Goyal, Durresi & Jain Drop Precedence Study [Page 13] draft-goyal-dpstdy-diffserv-01.txt- 14 - June, 1999 ------------------------------------------------------------------------------- 2 0.5 1 1 40/60 - 20/60 0% 146120 150320 1499000 0.9934 2 0.5 1 0.5 40/60 - 20/40 0% 142802 150711 1499202 0.9919 2 0.5 1 0.5 40/60 - 0/40 0% 118103 153511 1499703 0.9914 2 0.5 1 1 0/60 - 0/5 0% 149253 149760 1497093 0.9879 2 0.5 1 1 40/60 - 0/5 0% 149253 149760 1497093 0.9879 2 1 1 1 40/60 - 20/60 0% 171372 147422 1498172 0.9867 2 0.5 1 1 0/60 - 0/10 0% 179251 146533 1498051 0.9862 2 0.5 1 1 40/60 - 0/10 0% 179251 146533 1498051 0.9862 2 0.5 1 0.5 0/60 - 0/5 0% 149944 149698 1497224 0.9859 2 0.5 1 0.5 40/60 - 0/5 0% 149944 149698 1497224 0.9859 2 0.1 1 0.1 40/60 - 20/40 0% 90225 156524 1498945 0.9790 3 0.1 0.5 1 0/60 0/60 0/60 8.5% 209802 143156 1498202 0.9790 3 0.1 0.5 1 0/60 0/60 0/60 85% 209802 143156 1498202 0.9790 2 0.1 1 1 0/60 - 0/60 0% 204088 143876 1498968 0.9786 2 0.1 1 0.1 40/60 - 0/40 0% 88105 156809 1499385 0.9758 2 1 1 1 40/60 - 20/40 0% 78797 157858 1499517 0.9726 2 0.1 1 0.5 40/60 - 0/40 0% 76815 158018 1498975 0.9709 3 0.1 0.1 0.1 40/60 20/40 0/20 8.5% 76447 158089 1499247 0.9704 3 0.1 0.1 0.1 40/60 20/40 0/20 85% 76447 158089 1499247 0.9704 3 0.1 0.1 0.1 40/60 20/60 0/20 8.5% 76447 158089 1499247 0.9704 ------------------------------------------------------------------------------ Table 11: Best 20 UDP Control Results for Green Rate = 51.2% of 1.5Mbps (76800bps per customer) ------------------------------------------------------------------------------- #Colors Max_Prob Min_Th/Max_Th Yellow Throughputs(bps) Fairness G Y R G Y R Rate UDP Avg TCP Total ------------------------------------------------------------------------------- 3 0.5 0.5 1 40/60 20/60 0/20 85% 70963 158613 1498483 0.9602 2 0.1 1 0.5 0/60 - 0/10 0% 71378 158560 1498418 0.9667 2 0.1 1 0.5 40/60 - 0/10 0% 71378 158560 1498418 0.9667 2 0.1 1 0.1 0/60 - 0/5 0% 71424 158524 1498144 0.9585 2 0.1 1 0.1 40/60 - 0/5 0% 71424 158524 1498144 0.9585 2 0.1 1 0.5 0/60 - 0/5 0% 71470 158524 1498190 0.9598 2 0.1 1 0.5 40/60 - 0/5 0% 71470 158524 1498190 0.9598 2 0.1 1 1 0/60 - 0/10 0% 71470 158524 1498190 0.9598 2 0.1 1 1 0/60 - 0/5 0% 71470 158524 1498190 0.9598 2 0.1 1 1 40/60 - 0/10 0% 71470 158524 1498190 0.9598 2 0.1 1 1 40/60 - 0/5 0% 71470 158524 1498190 0.9598 3 0.5 1 1 40/60 20/60 0/20 8.5% 71747 158489 1498147 0.9671 3 0.5 1 1 40/60 20/60 0/20 85% 71747 158489 1498147 0.9671 3 0.5 0.5 1 40/60 20/40 0/20 8.5% 71747 158489 1498147 0.9671 3 0.5 0.5 1 40/60 20/40 0/20 85% 71747 158489 1498147 0.9671 3 0.1 1 1 0/60 0/40 0/20 8.5% 71931 158524 1498651 0.9502 3 0.1 1 1 0/60 0/40 0/20 85% 71931 158524 1498651 0.9502 3 0.1 0.5 1 0/60 0/40 0/20 8.5% 72161 158533 1498961 0.968 2 0.5 1 1 40/60 - 0/20 0% 72161 158551 1499121 0.9511 Goyal, Durresi & Jain Drop Precedence Study [Page 14] draft-goyal-dpstdy-diffserv-01.txt- 15 - June, 1999 3 0.5 1 1 40/60 20/40 0/20 8.5% 72161 158551 1499121 0.9511 ------------------------------------------------------------------------------ 6. Summary The simulations presented here suggest that the optimal number of colors or drop precedence levels in an Assured Forwarding traffic class depends on the traffic load and the amount of reserved (green) traffic. In a mixed TCP-UDP traffic scenario, best performance in terms of either fairness or restricting UDP throughput can be achieved with either 2 or 3 levels of drop precedence if most of the TCP packets have different (and 'better') color than excess UDP packets. With a clear distinction between TCP and excess UDP packets, one can use appropriate RED parameters so as to achieve desired results in terms of fairness or restricting UDP throughput. With 3 colors in a traffic class, TCP packets can be easily differentiated from excess UDP packets in most of the traffic scenarios. Note: This draft is being continuously updated. For the latest version, please see: http://www.cse.wustl.edu/~jain/ietf 7. References [juha] J. Heinanen, et al., Assured Forwarding PHB Group. RFC 2597, June 1999. [floyd] 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. Internet draft-Work in Progress , August, 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, Arian Durresi, Raj Jain Computer and Information Science Dept, Goyal, Durresi & Jain Drop Precedence Study [Page 15] draft-goyal-dpstdy-diffserv-01.txt- 16 - June, 1999 The Ohio State University, 2015, Neil Avenue,Room 395, Columbus, OH 43210 Phone: +1 (614) 292-3989 Fax: +1 (614) 292 2911 Email: {mukul,durresi,jain}@cse.wustl.edu This internet draft expires on December 1999 Goyal, Durresi & Jain Drop Precedence Study [Page 16]