Gigabit Ethernet Switches

Gigabit Performance
My goal in testing Gigabit Ethernet switches in the Lab was to get an idea of real-world performance gains rather than perform benchmark testing that would pit one product against another. To simulate a production network, I set up the Lab test network as follows.

The network consisted of 20 Fast Ethernet client PCs running Windows 2000 Professional and Windows NT Workstation and six Win2K servers. I first established a baseline by using these machines on the existing Fast Ethernet switches in the Lab. I recorded average throughput and elapsed time to perform certain tasks and monitored general network responsiveness during peak loads. After I established some numbers for comparison, I installed 64-bit Gigabit Ethernet cards in the six Win2K servers. Then, for each vendor, I replaced the existing Fast Ethernet switches with the vendor's Gigabit Ethernet equipment. I connected the client PCs to the desktop switch and used its 1000Base-T port to link it to the aggregation switch. Then, I connected the Win2K servers to the aggregation switch by using Cat 5 Enhanced (Cat 5e) patch cables for all the connections.

This test scenario might sound cut and dried, but I encountered many hurdles during my testing. You'll likely face similar problems when upgrading your networks, so I discuss them in the Web-exclusive sidebar "Upgrading to Gigabit Ethernet" (http://www.winnetmag.com, InstantDoc ID 24584). In general, I found that the right equipment with the right configuration can produce some worthwhile performance increases, although I never achieved anything close to true gigabit speed on the equipment I tested.

Two of the top-performing servers I tested were the Compaq ProLiant 7000 with a 64-bit 33MHz bus and the HP NetServer LT6000r with a 64-bit 66MHz bus. I used each server's high-performance SCSI RAID adapters in tandem with large arrays of Seagate Technology's Cheetah 10,000rpm SCSI drives. Intel PRO/1000 NICs were installed in each server, and I used Windows System Monitor to monitor throughput.

With some effort, I was able to coax the HP NetServer LT6000r to a maximum sustained network throughput of 460Mbps. I achieved this result during an application deployment test in which I pushed a 390MB Microsoft Office 2000 installation package to 18 Fast Ethernet clients while transferring similarly sized files to three other Gigabit Ethernet servers. To check general network responsiveness, I ran streaming multimedia content and used Windows Explorer to browse network resources on the two remaining Fast Ethernet clients. I didn't notice any image frame drop or jittering, and general network browsing was only slightly slower than usual. This test was simply impossible on Fast Ethernet equipment because it brought the network to a near standstill.

I noticed a significant improvement in multicast performance, measured by the time required to send a 690MB disk image to 18 multicast clients in one session. The HP NetServer LT6000r served as the multicast server, and the clients were using Fast Ethernet links to the desktop switch. On the Fast Ethernet network, the task took 19 minutes. On the Gigabit Ethernet network, the time was reduced to 9 minutes.

I measured the transfer of a large (1GB) file between the same hosts over Fast Ethernet and Gigabit Ethernet links with sustained network traffic (streaming media to multiple unicast clients). The file transfer took 230 seconds on Fast Ethernet and 88 seconds on Gigabit Ethernet.

Graph 1 and Graph 2 compare the peak performance numbers of Fast Ethernet and Gigabit Ethernet on three systems that I tested. The enterprise server realized a larger performance gain than the workstations because of higher I/O capacity. In essence, the Gigabit Ethernet equipment eliminated the network bottleneck and exposed the I/O bottleneck at the server or workstation.

Overall, my tests showed that Gigabit Ethernet provided a tangible performance improvement, but bottlenecks elsewhere kept the overall throughput lower than I had hoped. I was satisfied with Gigabit Ethernet performance relative to Fast Ethernet, and I was particularly impressed that general network responsiveness remained acceptable even during peak network loads. But I was disappointed not to be able to reach much beyond 450Mbps on the Lab's most capable server.

Although I saw tangible performance improvement on the Gigabit Ethernet servers, most networks will realize their greatest performance gain from Gigabit Ethernet over copper at aggregation points where multiple desktop switches bring hundreds of Fast Ethernet clients to converge on centralized server resources. Regardless of your deployment scenario, it's crucial that you do your homework beforehand so that you can forecast the performance gains you'll achieve in your particular environment. Then you can judge whether the gains will be worth the expense of upgrading.