NetWare MultiProtocol Router Configuring for Performance September 25, 1992 The NetWare MultiProtocol Router is an NLM based software product capable of full 32 bit processing and can be installed on a variety of 386 or 486 PCs using many types of ODI certified Network Interface Cards. By considering the amount of traffic that travels across the router and choosing the right PC and Network Interface Cards, you can build a router that is just right for your application. Choosing a PC based Router Before you purchase a PC or a Network Interface Card for a PC based router, you should evaluate the performance and cost trade offs of each available configuration. The major considerations are: - PC bus - CPU - Network Interface Cards. PC Bus There are three choices available: Industry Standard Architecture (ISA), the Enhanced Industry Standard Architecture (EISA), and the Micro Channel Architecture (MCA). The following table represents the instantaneous transfer rates for a burst of words at the bus speed of 8 MHZ (using BCLK). You should use an MCA or EISA PC for high performance routing. Bus Type Peak Transfer Rate -------- ------------------ ISA 8-bit 1 MBps ISA 16-bit 2 Mbps EISA 32-bit burst 33 Mbps MCA 32-bit burst 40 Mbps CPU A 80386 or 80486 processor is a good choice for PC based routers. Common clock speeds are 16MHz, 25MHz, 33MHz, and 40Mhz for 80386; and 25 Mhz, 33 Mhz, and 50 Mhz for 80486 platforms. You should use a 80486/50MHz machine for high performance routing. When NetWare v3.11 is loaded, it performs a system speed test. This test ranks a file server's or router's performance with respect to CPU types, clock speeds, memory, and cache. A higher rating indicates a faster system. PC Platforms Table PC Platform Speed Rating 80386/16 Mhz 121 80386/25 MHz 242 80386/33 MHz 322 80486/25 MHz 686 80486/33 MHz 982 80486/50 MHz 1382 Network Interface Cards There are two types of adapters: bus-mastered adapters and slave adapters. Bus mastered adapters have an on board processor that will place the packet in the buffer of the PC without intervention from the CPU, reducing processing cycles and increasing performance. Slave adapters require that the PC's CPU remove the packet from the card and place it in the machine's buffer. This could adversely impact the router's performance. * Classifying Router Applications * According to the type of traffic, the number of users, and the LAN or WAN media, a router application can be placed in the low, medium, or high speed category. Examples of these three categories are shown below 5. Once a router application is placed in a category, the appropriate components can be selected and a router configured that delivers reliable connectivity at a reasonable price. For more information on these categories refer to MPRRUL.TXT. Example of Low Speed Router Configuration ----------------------------------------- 20 MHz 386SX ISA Bus 8 MB RAM LAN and WAN Boards: Synchronous/+ Adapter NE1000 NE2000 Intel EtherExpress TPE Hub (ISA) other 8 or 16 bit boards Network Configurations: 1 LAN and 1 WAN, or 2 LANs Example of Medium Speed Router Configurations --------------------------------------------- 33MHz 386DX or 486SX ISA Bus 8 MB RAM LAN and WAN Boards: Synchronous/+ Adapter NE2000 Intel EtherExpress TPE Hub (ISA) other 16-bit boards Network Configurations: 1 LAN, 1 WAN or 1 LAN and 4 WANs, or 2 to 4 LANs Example: High Speed Router Configurations ----------------------------------------- 50 MHz 486DX EISA or MCA Bus 16 M RAM LAN and WAN Boards: Synchronous/+ Adapter (MCA) NE3200 NE/2 -32 Intel EtherExpress TPE Hub (EISA) other EISA or MCA boards Network Configurations: 1 LAN and up to 4 WANs * or 2 to 6 LANs * using 2 or more ports/board * Evaluating Performance Benchmarks * Types of Benchmarks Performance measurements for routers are usually based on packets per second (pps). Managers who select routers on this type of performance measurements usually spend more money for the routers than they need to. High packet per second rates are achieved using minimum (64 byte) packets. An actual network often uses a mixture of small and large packet sizes. A 64 byte packet is associated with data requests and terminal emulation, while 516 byte packets are used with data and file transfers. Selection Criteria - A User's Perspective Network managers should evaluate data throughput instead of packets per second. End users will perceive a network as fast when data throughput capabilities are high, regardless of the packets per second measurement. They will also perceive the network as slow if data throughput is low. Bandwidth Utilization Bandwidth utilization, the percentage of channel capacity consumed by a router, is an important measure of performance. The total bandwidth of a LAN or WAN determines how much data can pass through a network. In theory, the maximum bandwidth of an Ethernet LAN is 10 Mbps (14,880 pps). A Token Ring LAN has either a 4 Mbps or 16 Mbps (28,985 pps) bandwidth limitation. Wide area network bandwidth utilization is expressed in kilobits per second (Kbps). Bandwidth limitations for WANs: - X.25 services are generally available at 19.2 Kbps. In Western Europe and the United States they are available at 64 Kbps. - T1/E1 at 1.544/2.048 Mbps. In a typical LAN environment, the average bandwidth utilization is 30 to 40 percent. Short bursts of data can consume up to 80 percent of the available bandwidth. Routers that forward packets with utilization rates of 40 percent or more provide adequate performance for most LANs. For wide area connections, bandwidth utilization rates greater than 10% can saturate a T1/E1 link. * Test Results * The following charts show the bandwidth utilization, data throughput, and packets per second measurements of various router configurations. Please note that for large packet sizes, the MultiProtocol Router achieves more than 90 percent bandwidth utilization for both Ethernet and Token Ring, implying that the router can completely saturate a LAN. Driver limitations are responsible for lower performance, not the router software. Routers can be configured using a high performance system and LAN cards. One of the more interesting results is the 386SX with two NE2000 cards. It has 54 percent Ethernet bandwidth consumption. Most LANs run at rates far below this number, allowing customers to use a low-cost system and interface card on many LAN-to-LAN routers. * Testing Methodology * The performance test results were compiled by LANQuest Labs, an independent test lab. Packet generation and measurement equipment include: - Novell LANAlyzer - Bytex LAN test set - LANQuest Lab's FrameThrower Measurements were recorded at a sustained pps rate which is the number of packets received by the router equal to the number of packets routed out of the router without any packet loss. Novell encourages all network interface card and PC vendors to participate in LANQuest Lab's compilation of the MultiProtocol Router Benchmark. The MultiProtocol Router Benchmark provides users with information that will help them in choosing the right system for building manageable, global networks. For more information on LANQuest Lab's MultiProtocol Router Benchmark, call 1-800-487-7779. Benchmark Summary for IPX Routing with Ethernet System NIC Ports Packet Data Packets % of Size Thruput per Bandwidth Kbps Second Utilization ----------------------------------------------------------------- Epson NE2000 1 to 1 64 140 2180 11 320sx 516 531 1029 42 20MHz 1516 675 445 54 Compaq NE3200 1 to 1 64 207 3229 17 486 516 932 1807 75 33MHz 1516 1114 735 89 2 to 2 64 393 6135 16/port 516 1843 3572 74/port 1516 2210 1458 88/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. Benchmark Summary for IPX Routing with Token Ring 16M ----------------------------------------------------- System NIC Ports Packet Data Packets % of Size Thruput per Bandwidth Kbps Second Utilization ---------------------------------------------------------------------- Epson Madge AT 1 to 1 64 140 2182 7 320sx (16 M) 516 547 1060 27 20MHz 4100 1144 279 57 Compaq Madge 1 to 1 64 535 8354 27 486 EISA 516 1789 3468 89 33MHz (16 M) 4100 2005 489 100 AST Olicom 1 to 1 64 368 5748 18 486 EISA 516 1003 1943 50 33MHz 16/4 4100 1160 284 58 IBM Olicom 1 to 1 64 273 4266 14 PS/2 MCA 16/4 516 1009 1955 50 Model 95 4100 1255 306 63 IBM Olicom 2 to 2 64 246 3845 6/port PS/2 MCA 16/4 516 1062 2058 27/port Model 95 4100 1263 308 32/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. Benchmark Summary for AT Routing with Ethernet ---------------------------------------------- System Interface Ports Packet Data Packets/ % of Card Size Thruput Second Bandwidth (Kbps) Utilization ------------------------------------------------------------------- Compaq NE3200 1 to 1 64 196 3060 16 Ethernet 516 949 1839 76 2 to 2 64 281 4383 11/port 516 1660 3218 66/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. Benchmark Summary for AT Routing with Token Ring 16M ---------------------------------------------------- System Interface Ports Packet Data Packets/ % of Card Size Throughput Second Bandwith (Kbps) Utilization -------------------------------------------------------------------- Compaq Madge 1 to 1 64 272 4256 14 486 EISA 516 1325 2567 66 33MHz (16 M) AST Olicom 1 to 1 64 224 3498 11 486 EISA 516 869 1685 43 33MHz 16/4 IBM Olicom 1 to 1 64 221 3459 11 PS/2 MCA 16/4 516 953 1847 48 Model 95 2 to 2 64 208 3246 5/port 516 1015 1986 25/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. Benchmark Summary for IP Routing with Ethernet ---------------------------------------------- System Interface Ports Packet Data Packets/ % of Card Size Thruput Second Bandwidth (Kbps) Utilization -------------------------------------------------------- Compaq NE3200 1 to 1 64 205 3199 16 486 516 930 1802 74 33MHz 1516 1092 720 87 2 to 2 64 392 6121 16/port 516 1831 3584 74/port 1516 2233 1473 89/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. Benchmark Summary for IP Routing with Token Ring 16M ---------------------------------------------------- System Interface Ports Packet Data Packets/ % of Board Size Thruput Second Bandwidth (KBps) Utilization ------------------------------------------------------------------- Compaq Madge 1 to 1 64 488 7631 24 486 EISA 516 1677 3250 84 33MHz (16 M) 4100 1976 482 99 AST Olicom 1 to 1 64 367 5731 18 486 EISA 516 1001 1939 50 33MHz 16/4 4100 1197 292 60 IBM Olicom 1 to 1 64 259 4044 13 PS/2 MCA 16/4 516 1006 1950 50 Model 95 4100 1259 307 63 2 to 2 64 241 3768 6/port 516 1005 1947 25/port 4100 1251 305 31/port Note: The number in the Packets/Second column is the forwarding rate of the router. On a 2 to 2 configuration, this number is a combined throughput total. The specifications detailed in this document are subject to change without notice.