REVIEWS

Good Things Come In Small Packages: Combo PC Cards

Motorola Marquis and 3Com's EtherLink III LAN+ Modem PC Card

They're not $300 an ounce or $3,000 a karat, but the surprises wrapped up in some of these combination Ethernet/Modem PC cards make them great small packages. We tested 10 different Ethernet/V.34 modem PC cards and found distinct differences in performance as well as packaging.

At the top of the heap rest the Motorola Marquis and 3Com's EtherLink III LAN+ Modem PC Card. These two cards are exactly the sa me in every respect, with the exception of their list prices. Motorola's suggested retail price is a mere $50 less than the 3Com card, giving Motorola a slight advantage. Both dominated the competition with clearly superior performance--a difference that makes either of these solutions a choice that's hard to resist.

While we were somewhat surprised at the 3Com/Motorola dominance, we were very surprised at the disappointing performance of products from veteran manufactures Xircom and US Robotics. US Robotics has corrected the errors that had caused us to stub our toe, but Xircom's card seemed to have no more than its name going for it.

Ethernet Performance To fully examine the 10BASE-T Ethernet performance component of these combo cards, we used Novell's DOS-based PERFORM3 utility. It sends different sized packets over a variable time interval and measures the maximum and ave rage bytes per second transferred. All of our tests transferred packets from 64 bytes to 4,096 bytes. We plotted the maximum and average values gathered over our entire Ethernet test. Results can be seen in the chart to the right.

File transfers were performed between a single 486/50 laptop through a 10BASE-T hub to a single NetWare 4.1 server running on a 486/25. Each card ran the tests over Novell's newest release of Virtual Loadable Modules (VLMs) with packet burst turned on. Only the laptop and file server were active on the segment during the test.

To compare the tested cards with a typical desktop card, we also ran the tests using an NE2000 running VLMs (also with packet burst mode enabled). The NE2000 transferred a maximum of 442.6 KB per second and averaged 343.3 KB per second across the entire range of packet sizes.

All of the cards did better than our control NIC. This surprised us, since we didn't think these little dudes were capable of such performance. While the majority of the cards did yield about the same performance, the 3Com and Motorola cards were, without a doubt, superior.

The 3Com EtherLink III LAN+Modem PC Card and the Motorola Marquis averaged 544 KB per second--more than 500 bytes per second faster than the next closest competitors, the Epson ELF-M28-NA and the ActionTec ComNet 288, when transferring over Ethernet. They had a peak transfer rate of 812 KB per second-- more than 900 bytes per second faster than the Olicom GoCard, the next closest competitor.

The rest of the pack congregated into two groups below those already mentioned. One group consists of US Robotics XJEM 3288, built by Megahertz, and the NovaLink NovaLAN 288. They have acceptable but middle-of-the-road performance. The remaining products form another group at the bottom of the pack: the Xircom CreditCard Ethernet+Modem 28.8 and the Silicom EtherModem.

We found, as expected, that the transfer rates sloped from higher to lower as packet size went from larger to small er. Our NE2000 control test returned a steady slope, from large to small.

For th e most part, even the poorest performers, Xircom and Silicom, showed this slope. Worth noting, twice during our tests these two cards jumped from a slowly downward slope to dramatically higher performance that would slowly dropped off again. This inconsistency cost them the race.

V.34 Modem Performance At the risk of sounding like a broken record, 3Com and Motorola were also the clear winners when it came to modem performance. They battered the next closest card by more than 15 percent at their highest rate. Again, the rest of the products migrate toward two categories--average and poor performance--but these results need a little explaining (see "Half the Story: How We Tested V.34 Modems").

3Com and Motorola transferred a peak of 8,340 cps--210 cps faster than the next fastest rate of 8,130 cps, turned in by a number of the second grouping of cards. Not only did 3Com and Motorola turn in the fastest peak time, but they clearly turned in fastest times more often than a ny of the other cards. This dominance continued into the next lower range of plotted results (between 7,000 cps and 8,000 cps).

At this point, the results start to become unclear, since most of the cards bunch together in the 7,000 cps and 6,000 cps ranges. The next best cards were from ActionTec and Olicom, followed closely by Ositech, Epson and Silicom. Seeing any real outstanding difference is difficult here, except when compared with the modem performance of the cards at the bottom of the performance heap.

Two cards clearly distinguished themselves as the worst modem performers: the Xircom CreditCard Ethernet+Modem 28.8 and the Megahertz XJEM3288. We were not surprised that the Xircom card faired poorly, since it had done the same in earlier testing. This card is completely non-aggressive, with all successful transfer rates in the 5,000 cps and below range.

US Robotics' card was a bit more of a su rprise, due to the fact that US Robotics usually builds some of the fastest and most reliable modems we've tested. In our test (which we performed over and over, giving US Robotics the benefit of the doubt), the XJEM3288 connected with a Penril Alliance V.34M, but with the exception of a small percentage of connections, could not make an LAPM error-correcting connection. The resulting raw connection had far too many errors when transferring data, resulting in a nearly zero bits transferred per second and a subsequent failure on almost all the tests. US Robotics claims that it currently has a solution to the problem, but we weren't able to test it by press time.

A result of the modem testing we also noted was the connection failure rate of each modem. Not surprisingly, the pair from 3Com and Motorola won in this category, turning in zero failures with a 100 percent successful connection ratio. The ActionTec and Epson modems both did well with less than a 1 percent failure rate. Modems from Olicom, Ositech and Silicom didn't do badly either, having less than a 4 percent fail ure rate. Still, this is where good connection rates end. The Xircom card turned in a miserable 24 percent failure rate that looked good in comparison to NovaLinks' 44 percent and US Robotics' 85 percent failure rates. Both NovaLink and US Robotics were shocked and worked to remedy the situation. We retested multiple times, but to no avail. US Robotics, as mentioned previously, has indicated that it now has a resolution for its failures.

Features When it comes to features, other cards in the field finally begin to distinguish themselves. Product features fell into two basic categories: card operations and add-on software/features. Card operations includes things like the number of PC Card controllers supported, the power requirements for the card, whether or not it's supported by Windows95 and support for functions like ISDN. In the add-on software/features category, we looked for things like mail software, digital line protection and remote node software. We rated based on quantity, rath er than making judgments about each features' quality. So if one feature is more important to you than another, consider it in your evaluation.

The Jack of Diamonds from Ositech had the best overall features score by offering the widest support for PC Card controllers, supporting eight different manufacturers, having a low power requirement and being able to dynamically manage the power the card draws when one or both of the functions of the card is not in use. Ositech also is very involved in adding function to its card, evident in the Deuce offering. For an additional cost, the Deuce is a 10BASE-T hub that supports the attachment of two additional machines. Ositech will be adding sound functions to its combo card in the future. (Anyone for multiplayer Doom at 30,000 feet?) Ositech was the only vendor to offer a NetWare server driver and Unix driver for its card--great for bringing your server to an on-the-road demo or class.

NovaLink Technologies provides two excellent add-on software packages with the purchase of its card. The Nova Mail software offers a fax and voicemail program that will digitize incoming voicemail and call an alphanumeric pager with a message. This allows for the transport of your mailbox and notification of a specific call without having to check voicemail. The second software package, the Remote Node client from Microcom, adds PPP remote node dial-in support along with Microcom's Carbon Copy Remote Control. This is not a bare-bones dialer, but a fully scripted, manageable, well-designed and easy-to-use remote client.

Not to be left out, the ActionTec ComNet 28.8 card comes with Remote Control software from Traveling software, which, like the remote node software provided by NovaLink, is not just a bundled crippled freebie but a well-designed piece of remote access technology. ActionTec also ships a Personal Information Manager from Sidekick to make this modem purchase a total solution for lapto p users.

Most of the modems support coax and UTP, b ut the Xircom CreditCard Ethernet+Modem 28.8 still has the best no-hassle telephone jack. This spring-loaded RJ11 slips back into the card when not being used and means one less connector to haul around or, in our case, forget. Silicom also offers a unique dual-function UTP/RJ11 jack that is about the size of Rhode Island (OK, it's really only 2.5 inches by 3.5 inches, and 1 inch thick), but it has the ability to service both the modem and NIC, and means carrying only one connector. It also offers the standard two-dongle approach.

Bruce Boardman can be reached at bboardman@nwc.com.

Half The Story: How We Tested V.34 Modems

For modem performance, we ran the now familiar suite of tests outlined in the Telecommunications Systems Bulletin (TSB)-37A and TSB-38. The suite consists of 165 tests that simulate all the known line conditions that exist today in the Public Switched Telephone Network (PSTN). We ran 36 of the tests, which account for 91 percent of line conditions. The 129 remaining tests make up the other 9 percent, with no single test being responsible for more than 0.2 percent of known line conditions.

We measured our results by plotting the characters per second (cps) transferred in each of the 36 tests. We weighted the results of each test by the likelihood that the conditions simulated would occur in the U.S. PSTN. This weighting is important, because connect speed and transfer rate are similar for most modems when connecting over clean phone lines. But the likelihood of getting a particular set of line characteristics is not equal. Therefore, we plotted each modem's throughput weighted by the likelihood that the conditions simulated would occur on the U.S. phone network.

So to score well, a modem had to transfer data more quickly on tests that simulated common conditions. In our chart, yellow and purple bars show faster transfer rates and are, therefore, more desirable. Blue bars sh ow the slowest transfer rates and are least de sirable. In our graph, if the bar does not go all the way to the 91 percent line, it is because the modem didn't connect or the connection was so poor that no data was successfully transferred. So, the optimum performance would be all yellow, all the way to 91 percent. The 3Com and Motorola products, which connected on all our tests, show their dominance with the longest yellow and purple bars.

We used Penril's Datability Networks Alliance V.34M as our control modem on the other end of the line. This modem won our last desktop modem roundup (November 15, 1995, page 101).

April 1, 1996