Gigabit Over Copper: Bandwidth To Burn? -- Gigabit over copper delivers speed, but may come up short on reliability. Four vendors took our 'ISP Rack Collocation Challenge' to display their complete copper-gigabit solutions.
Network Computing, Sept 18, 2000 p48
nwc.com
By Conover, Joel
(NOTE: below is only a snip. use the link for the full article)
They said Gigabit Ethernet over standard copper wiring could not be done. But on June 23, 1999, IEEE members passed the 802.3ab Gigabit Ethernet Over Copper standard. Now, more than a year later, the industry is delivering its first gigabit-over copper-solutions to the mass market.
Gigabit over copper holds two strengths over its fiber brethren: lower cost and the ability to run over existing copper wiring. But the technology's reliability may not be enough for its target applications in data centers, ISP collocation sites and high-performance workstations and servers, where fault-tolerance is absolutely critical.
At Schneider National, one of our partner labs, in Green Bay, Wis., we tested the first wave of products. Although they deliver wire-speed throughput in the lab, we found gigabit over copper may deliver only four-nine reliability when pushed to its limits. The small chance that the physical layer could be to blame may be reason enough to cost-justify fiber.
The Gigabit Ethernet market is also surfing through an incredible commoditization phase. While the price rift between 1000BASE-SX optical and 1000BASE-TX gigabit over copper is slowly closing, 1000BASE-SX fiber ports still cost nearly 50 percent more per port than their copper equivalents. Network managers must decide if a 50 percent cost savings is worth a few thousandths of a percent risk.
The Manifest
Network Computing built a test plan and asked vendors to submit a complete end-to-end solution for a high-density ISP collocation site using gigabit-over-copper technology. We dubbed our test the ISP Rack Collocation Challenge. Each of the vendors had to deliver an end-to-end solution suitable for deployment in one 84-inch tall, 19-inch wide ISP collocation rack, using gigabit-over-copper technology and focusing on fault-tolerance, server load-balancing and space optimization.
Why the end-to-end approach? Because the vendors pitch such solutions to their customers and had begged us to test them this way. We invited 15 vendors to stop their huffing and to deliver on their bluster, yet only four vendors stepped up. We were not concerned with an apples-to-apples comparison; rather, we simply required each vendor to present its best gigabit-over-copper solution, keeping in mind collocation and server-room consolidation as primary targets for its technology.
In our test plan, we told vendors we'd be judging the solutions' physical fault-tolerance, fault management and availability. Furthermore, we asked each vendor to show us three features that significantly differentiate its solution from the competition.
We solicited participation from Alcatel, Alteon WebSystems, Asante Technologies, Cisco Systems, Compaq Computer Corp., Enterasys Networks, Extreme Networks, Foundry Networks, Hewlett-Packard Co., Intel Corp., Lucent Technologies, Nortel Networks, SysKonnect, Sun Microsystems and 3Com Corp. Alcatel, Compaq and HP claimed they had no products that fit the nature of our tests (gigabit over copper and server load-balancing). Lucent and Sun, though they had products, chose not to participate. Enterasys and 3Com sent no response.
Although Alteon, Asante and Nortel initially agreed to participate, we never got to test their products. Alteon informed us it felt a competitor's not-yet-shipping product was not a fair comparison with Alteon's already-shipping product. (It has always been Network Computing's policy to insist that products we test be shipping by the date of publication.) Nortel said its Passport 8600 gigabit-over-copper blades were not yet available for tests. Finally, we determined that the Asante FriendlyNet GX4 was not an appropriate contender among these large, enterprise-class products, and we removed it from our roundup.
Ultimately, we tested four complete solutions and two NICs. Representatives from Cisco, Extreme, Foundry and Intel trekked to our Wisconsin lab with a full rack of collocation gear. Products included server load-balancing hardware, firewalls, SSL accelerators, core switches and content-switching solutions-all gigabit-over-copper attached. SysKonnect and Intel sent NICs, which we tested separately (see "Gigabit-Over-Copper NICs," page 72).
Analyzing the Report Card
Our report card is broken down into five areas: price, management, fault-tolerance, density metric and advanced features. The vendor's price score is based on the cost of the solution in the rack configuration. The price metric includes all ancillary equipment. The management score is a measure of the overall ability to manage faults, performance monitoring and troubleshooting of the solution. Fault-tolerance is a function of the total solution. We asked each vendor to demonstrate that there was no single point of failure in the network and to show us the levels of fault management available in the solution. We weighted this metric equally with the level of connectivity and server density each vendor provided in the rack.
We also asked each vendor to show off how it manages its solution in terms of statistics, device management and overall server-farm health monitoring. These results, along with our analysis of the unique features, are reflected in the advanced-features score.
Final Analysis
In addition to evaluating each vendor's solution based on technical merit, feature set and fault-tolerance options, we contrived several metrics to quantify how fast and dense the vendor's overall solution was. Our primary metrics were bandwidth per inch (the higher the better, as it indicates more potential throughput), cost per inch (lower is better, providing quicker return on investment) and load per server (the lower the better, delivering faster response times) assuming an OC-12 connection to the upstream Internet provider. Based on the size and complexity of the products involved, typical server load never exceeded 20 Mbps, a load that almost any modern server can handle with ease. Of course, if larger (than 1U) servers are used, the load per server goes up accordingly. Several vendors also pointed out that a typical large ISP collocation effort would involve multiple sites connected at OC-3 speeds or below, to enhance fault-tolerance and response time on the global Internet. If that is true, the typical server will barely push more than 100 Mbps. Gigabit over copper leaves you with plenty of bandwidth to burn.
We chose Cisco's ArrowPoint solution as our Editor's Choice. Its content-aware networking gave it a specific and valuable advantage over the competition. Foundry and Intel tied for a close second. The small edge Cisco held over its competitors was largely due to its extremely useful reporting features and incredible content-aware switching. Extreme came in last, showing the green nature of its first-generation server load-balancing solution.
Executive Summary -- Gigabit-Over-Copper Solutions
Gigabit Ethernet over copper is finally appearing en masse in the market. This technology typically offers up to 50 percent cost savings over traditional fiber-based networks, and can make a low-cost interconnect medium for data centers, collocation facilities and high-performance clusters. But if 99.999 percent reliability is your priority, you'll want to think twice. Our tests showed gigabit over copper can be finicky-particularly sensitive to cabling-plant issues and external noise.
We invited vendors to display their gigabit-over-copper solutions as they would recommend them for a typical collocation environment. Cisco Systems, Extreme Networks, Foundry Networks and Intel showed up in our labs, touting complete collocation solutions. Cisco took top honors, winning our Editor's Choice award for its expensive but enticing solution based on its Catalyst 6500 series and ArrowPoint Communications content-switching products.
Because gigabit-over-copper solutions also require specialized NICs, we invited vendors to submit their copper-based cards for testing. Intel and SysKonnect chose to participate in these tests, with SysKonnect showing a production card that ran rings around Intel's beta effort. |
