Fast and Furious

I Tested The 4507R, the only model that comes with the option for redundant Supervisor cards. I tested this option by pulling out the active Supervisor card while pinging through the box. It took about 30 seconds for the backup Supervisor card to kick in and start forwarding the pings. However, with a large route table such a switchover could take as long as a minute, Cisco reps say. I also pulled one of the redundant power supplies while running our most demanding throughput tests, and the 4507R didn't drop a packet.

For my tests, the 4507R was fully populated with five line cards containing six gigabit-fiber ports on each. There were also two active gigabit-fiber ports on the Supervisor card for a grand total of 32 gigabit ports. I hooked up a Spirent SmartBits 6000B tester to the box and blasted true wire-speed traffic, 100 percent utilization, through all 32 ports in full-duplex mode. I sent 1,518-byte packets and the worst-case 64-byte packets, which caused the 4507R to process more headers in the same amount of time. I also sent some in-between-sized packets and set the SmartBits to tolerate 0 percent drops--not a single packet was dropped.

Good • True wire-speed performance, even with access lists. • Heavy-duty redundancy. • Inline power. Bad • Expensive. • Limited wire-speed port density.

Next I configured a 30-line access list to be active on all the ports and specified that the whole list had to be traversed. The throughput was perfect, confirming Cisco's claim of wire-speed access-list processing. I left the access list on and did a latency test in full-mesh mode, which divided the traffic equally from each individual port to every other port and made the switch work a little harder. The best-case latency was about 3.4 microseconds with 64-byte packets. Even the worst case was low enough to be insignificant--about 73 microseconds with 1,518-byte packets.

Boosting the 4000

Cisco recently released some new software features for the Catalyst series. With the new software, the 4500 and even the 4000 series with the older Supervisor 3 card will support 802.1x wireless authentication and DHCP snooping, an impressive feature. In my tests the 4507R was able to prevent a DHCP server from answering DHCP requests on a port-by-port basis. This is a great capability for anyone who has had problems with rogue DHCP servers, which can wreak network havoc. Unfortunately, the server editions of Microsoft Windows 2000 practically beg the installer to turn on the DHCP server, making this a serious issue on some networks.

Vendor Info Cisco catalyst 4500 series, starts at $995. Cisco catalyst 4000 supervisor engine iv, $13,495. Cisco Systems, (800) 553-6387. www.cisco.com

You don't get this performance unless you use the six-port line cards. Although you can get higher-density 10/100 and gigabit line cards, the bandwidth between each line card and the chassis is limited to 6 gigabits in each direction. Cisco makes no secret of this fact and recommends higher densities for wiring-closet applications, where the lower, bursty traffic coming from the desktops is unlikely to need more than the 6 gigabits in each direction at any given time.

The higher-density cards maximize the per-port cost, which is already high--the 4507R retails for $65,000. But the switch has full routing and switching capabilities along with QoS (Quality of Service) features and wire-speed access-list functionality, so it can easily handle a medium-sized backbone. It's possible to get higher port densities and fewer features and upgrade later, which makes for a more attractive wiring-closet solution. Having the switching engine centralized in the Supervisor card makes incremental upgrades easier.

The good news for 4000 series users is that the line cards are interchangeable between the 4000 and 4500 series. The new Supervisor cards are engineered for further enhancements--next year you'll be able to get a field-installable daughter card that will provide detailed network usage statistics via a Netflow card.

Peter Morrissey is a contributing editor and columnist for Network Computing. Write to him at ppmorris@syr.edu.