Interactive buyer's guide Find the right switch via our search form. Scorecard and NetResults How we ranked the switches in several categories, feature and performance comparisons, key findings and vendor contact info. How we did it Our testing methodology. Switch shopping: How to find the one that's right for you Not yet a commodity But with prices dropping fast... Mier is president of Mier Communications, Inc., a network consultancy and product test center based in Princeton Junction, N.J. He also is an editor of "The Switch Book,'' a comprehensive special report on LAN switches, which is published annually by Mier Communications (see www.mier.com/ books.html). Smithers is manager of lab testing and Scavo is a systems integrator at Mier Communications. They can be reached at (609) 275-7311 or via e-mail to ed@ mier.com, rob@ mier.com or tom@ mier.com.

By Edwin E. Mier, Rob Smithers and Tom Scavo
Network World, 8/11/97

The third-leading cause of global warming - that's how hot the 100Base-T switch market is today.

The best of these devices offers blazing throughput, sizzling management and flow- and traffic-control features. But a check of the prices of some Fast Ethernet switches can throw ice water on your enthusiasm pretty quickly, and others' ability to live up to their performance promise is only lukewarm. Only half of the 100Base-T switches we just tested could handle all the traffic users could reasonably deliver.

Among the best switches, Hewlett-Packard Co.'s AdvanceStack Switch 800T, Cisco Systems, Inc.'s Catalyst 5002 and FORE Systems, Inc.'s modular ES-3810 showed excellent performance with eight, 12 and 16 ports, respectively. However, Cisco's switch was considerably more expensive than the others. The two lowest ranked switches, 3Com Corp.'s SuperStack II Switch 3000 TX and Network Peripherals, Inc.'s (NPI) FE-1200, each lack support for 10/100 autosensing, and each failed to handle a full traffic load.

When evaluating switches, comparing price and throughput helps determine the value you get for your equipment dollar. The story "Not yet a commodity'' highlights our pick for Best Buy among these products.

Selecting the test candidates

In the eight- to 16-port range, 100Base-T switches tend to be stackable; many are of fixed configuration. More than half of the switches we tested in this round support no modular configuration options.

We specified that all ports had to support 100Base-T operation at 100-M bit/sec. As it turned out, all but two of the switches support 10/100 autosensing, in which the same RJ-45 port accepts a 10Base-T or 100Base-TX link, and can automatically adjust to whichever is plugged in - a big benefit to users migrating from Ethernet to Fast Ethernet.

Of the switches tested, only NPI's FE-1200 and 3Com's SuperStack II Switch 3000 TX supported only 100Base-T ports at the full 100M bit/sec, though 3Com is reportedly shipping a newer model of its 3000 that provides 12 10/100 autosensing ports. However, the same switching fabric is used as in the eight-port model we tested, according to 3Com. Therefore, we'd expect the new model's performance to be the same as the lackluster performance we measured for the eight-port model.

Keep in mind that in designing switched networks, there are places where switches of differing port capacities are appropriate. For example, two eight-port switches don't necessarily equate to one 16-port switch; with two there's a bigger bottleneck when packets must travel between ports on different switches.

We expected the 100Base-T switches we tested to be able to accept maximum- or near maximum-rate packet streams on half of their ports and switch the packets in these streams to the other half of their ports. We expected an eight-port switch to be able to accept up to four input streams, each up to 148,000 packet/sec (an aggregate load of nearly 600,000 packet/sec) and be able to switch this load to its four other output ports.

For a 12-port switch, we delivered up to six input streams, a total of nearly 900,000 packet/sec, destined for the switch's other six output ports. And for a 16-port switch, we delivered up to eight streams, a load of nearly 1.2M packet/sec.

With our 64-byte packets, these loads equate to more than 300M bit/sec of throughput for four streams and slightly more than 600M bit/sec for eight streams.

These capacities would far overload even the largest routers today, which typically can handle no more than a few hundred thousand packets per second. This is one of the main reasons that switching has become such a popular alternative to routing.

Hewlett-Packard: A winner

handled 100% of the throughput load we threw at it and exhibited no head-of-line blocking or other performance problems.

This switch is rich in features; it has port mirroring and the capability to trunk as many as four 100Base-T links between switches.

With port mirroring, traffic on one port can be copied to another for external monitoring and analysis. There's an impressive set of traffic control features, including automatic broadcast control and some features for reducing broadcast volume at the network layer.

The vendor's useful and easy-to-use Windows-based management software is included with the switch. We found its drag-and-drop virtual LAN setup and on-screen help to be superb.

Cisco: High functionality and price

If you want advanced features, this switch has them. You can set up to four trunk links between switches. If set to full-duplex, this feature provides up to 800M bit/sec of bidirectional bandwidth between switches.

Various flow - and traffic-control features include the ability to set, on a per-port basis, the number of packets per second of broadcasts that are allowed to pass. You also can set a threshold for any port on the percentage of total traffic that broadcasts make up. After the threshold is reached, subsequent broadcasts are discarded.

The Catalyst 5002 supports port mirroring and includes a slick set of management wares in the form of CiscoWorks for Switched Internetworks. It offers additional features such as the ability to limit traffic on a particular port to only packets with a designated station's media access control (MAC) address, which some users might employ for increased security.

The bad news, though, is price. Configured as tested, the switch costs $18,285. At more than $1,300 per port, the Catalyst 5002 is un-realistically overpriced, in our view.

FORE: High performance

Throughput was stellar. Not a packet was lost of the nearly 1.2 million packet/sec we delivered via eight concurrent packet streams. We saw no performance problems.

FORE offers a good range of features, including a couple of configurable broadcast and multicast control parameters. The switch supports VLANs, but only those de-signed to run over an ATM backbone. And ATM is how the vendor would like its customers to handle high-speed switch-to-switch links, as well; there's no provision for running multiple 100Base-T links between switches.

At the time we tested FORE's 100Base-T switch, the product wasn't fully integrated in the shipping version of the vendor's FORE-View management software. Rather, complicated software patches and downloads were required, and this was a hassle. Complete integration with the NT version of FOREView won't be available until the next release, due in October.

NBase: Good all-around performance

For flow control, the switch uses a fairly unsophisticated technique we call jamming to prevent too many packets from inundating the switch. The carrier signal of Fast Ethernet's carrier-sense multiple access with collision detection (CSMA/CD) is temporarily raised on a per-port basis whenever the switch concludes that, without applying some jamming, the probability of an overload - and lost packets - is at hand.

The sender, upon sensing the carrier, is compelled to back off for a pseudo-random period of time before again attempting to transmit. This gives the switch a chance to clear out its buffers. We noted that with flow control off, the eight-port switch could handle about 97% of the full load we delivered to it via four input packet streams. Unfortunately, we also noted some head-of-line blocking with this switch.

We found NBase has a particularly extensive MAC-based filtering capability in this switch, one of the few we tested that offered this.

The switch falls squarely in the middle of the pack considering features, throughput, management and price.

Cabletron: Limited throughput

The FN100 cannot interact with Cabletron's high-end MMAC-Plus switching hubs using the features of the vendor's sophisticated SecureFast architecture. But the FN100 does include some sophisticated features on its own, including the ability to support as many as eight concurrently active 100Base-T links between switches. This feature, providing up to 800M bit/sec of bandwidth, can help to avoid switch-to-switch bottlenecks.

There's also some pretty fancy flow control, in which the user can set broadcast-limiting thresholds on a per-port basis.

We found some throughput limitations with the FN100, however. We couldn't attain more than about 80% load via four concurrent traffic streams. And we also found indications of head-of-line blocking.

Bay: Top performance, light on features

The good news is that the BayStack 350T tied with FORE's ES-3810 for the most overall throughput of all the switches we tested. The unit handled, and appropriately switched, every packet within the eight concurrent streams of high-volume traffic we delivered to it - nearly 1.2 million packet/sec.

While definitely high performance, the 350T lacks much of the more sophisticated features that many competitors' 100Base-T switches now include as standard. The 350T supports port-based VLANs, the Spanning Tree Protocol and similar basic switch capabilities. But there are no special traffic control features, and there is no multiple switch-to-switch link capability, so a single full-duplex 100Base-T link is the most you can carry between switches. Also, there's no port mirroring.

Compaq: Good basic switch

We walked away from Compaq's switch with a couple of performance concerns, however. We found that a heavy load can essentially shut down SNMP management access to the switch. Also, we found indications of head-of-line blocking.

The switch's feature set is pretty basic. Flow control is based on the same jamming technique we saw in NBase's switch. There's no special trunking to allow more than a single 100Base-T link between switches and no port mirroring.

Management is fairly easy with the Windows-based Intelligent Management Software application.

3Com: Lowest total throughput

We saw no evidence of head-of-line blocking, but we did observe insufficient horsepower to handle management queries under a heavy traffic load.

Still, the 3000 offers an impressive array of traffic-control features, in-cluding various per-port parameters for limiting broadcasts and multicasts. But this model of the switch supports just 100Base-T links - not the 10/100 autosensing that all but one of the other switches supports. This switch also does not have port mirroring.

NPI: Out of steam

The NPI switch has a fairly full set of features, though. For example, you can aggregate as many as four ports as "power links'' for switch-to-switch connections. Port mirroring is there, as well as the ability to set up to 32 MAC-based filters per port. There's also a built-in broadcast/multicast-limiting feature, which constrains broadcast or multicast packets to no more than 10% of a port's throughput.

If you're willing to sacrifice on throughput, would settle for 100Base-T-only ports and are not looking for expandable support for other technology uplinks such as ATM or Gigabit Ethernet, this switch is an affordable alternative.