HP 9304M, 9308M 3.1 Throughput/Packet Loss Rate, 3. Performance Testing

HP ProCurve Routing Switch 9308M / 9304M Reviewer’s Guide

3. Performance Testing

Routing switches are normally performance tested under a variety of conditions, at Layer 2 and Layer 3, for three main parameters: throughput/packet loss rate, congestion control and latency. Definitions for throughput, packet loss rate and latency are specified in RFC 1242. See http://www.rfc­ editor.org/. Unfortunately, there is no clear definition for congestion control. There are numerous tests that test different aspects of congestion control, none of which is considered the definitive test for congestion.

Keep in mind that most of these tests are performed in an artificial environment intended to measure the outer limits of routing switches and rarely, if ever, reflect circumstances a switch would consistently find in an actual network environment. This is particularly true for congestion type testing. The applicability of these performance numbers to real world networks has to be interpreted carefully, as most routing switches in these tests perform to a level significantly exceeding the needs of real world networks.

Additional comments on each of the parameters follows.

3.1 Throughput/Packet Loss Rate

One of the major advantages that ASIC-based routing switches have over the traditional processor- based routers is that routing can be done near or at wire speed on all ports simultaneously for a cost considerably less than with a traditional router.

The throughput numbers given in Appendix A are shown for Gigabit-SX and 100Base-TX. The first table shows the situation where the packets don’t have to leave the module they came in on. Since each module in the 9300 routing switches routes on the module itself, this first table shows throughput without any packets traveling the backplane. As shown in the table, the 9308M produces complete wire speed routing on all ports simultaneously. The maximum number of packets per second that can travel through the routing switch is also shown in the 64 byte packet size column at 95,238,080. Although not explicitly measured, this figure would be 47,619,040 for the 9304M (one-half the rate of the 9308M.) This figure was measured with the routing switch routing IP. It would be the same for IPX routing or just switching at Layer 2.

The numbers for 100Base-T are full media-speed as well, but since the maximum number of packets that can be introduced to the switch is limited by the total number of ports possible at 18.8 Gbps (27,976,280 pps), the routing engines are not taxed at all in this situation.

In the situation where the throughput test is set up such that the destination for all packets requires a trip across the crosspoint matrix backplane, throughput is at wire speed, except for 64 byte packets where throughput is 91.3% of wire speed. This is due to the crosspoint matrix handling of 86,956,544 pps maximum. Note that this is a number of packets distinction, not data throughput. At larger packet sizes where the actual number of packets is less, the data rate through the backplane is at full wire speed. While this is a limitation in the total throughput of the routing switch, this only happens when all packets cross the backplane, all packets are 64 bytes in length, and all ports are receiving packets at wire speed full duplex simultaneously. This situation only happens in a testing environment. (If it happens momentarily in a real network, the excess packets would be buffered in the modules’ shared memory until the condition let up or shared memory became full.) To put this into perspective, if the average packet length through the routing switch exceeds 88 bytes, or if only 6 of the 8 gigabit ports on the modules have packets that cross the backplane, the backplane will run at full media speed. This clearly is not a limitation when dealing with real network environments.

This limitation is exactly the same for the 9304M. Since its crosspoint matrix is one-half the size of the 9308M, the matrix can handle 91.3% of media speed at 64 byte packets. The same comments given above as to why this really isn’t a problem in real environments also holds for the 9304M.

©1999 Hewlett-Packard Company Revision 4.0 – 4/1/1999 Page: 21 of 27