In our tests, cable modems provide faster Net access than analog modems or ISDN.
Andrew W. Davis and John W. Irza
Broadband data services for consumers are so hyped that many people probably wonder when they will even get such services, let alone whether they're any good. We're fortunate to live in suburban Boston, where several Internet service providers (ISPs) sell access via ISDN and cable modems, both of which promise substantial speed improvements over analog modems.
We compared these three types of devices in real-world tests that provide some of the first published results on commercially av
ailable cable modems. (BYTE's October 1997 Hardware Lab Report, "Bandwidth on a Budget: 34 Fast Modems," covered 56K de
vices, ISDN, and ADSL units; cable modems were not sufficiently standardized or widely available in time for those tests.)
Three Scenarios
To compare cable modems to analog modems and ISDN, we used three test vehicles. We tested a Bay Networks (
http://www.baynetworks.com
) cable modem using our cable provider, MediaOne (
http://www.mediaone.com
). This modem was connected to a Micron 166-MHz Pentium computer with 32 MB of RAM and Windows 95. Even thoug
h the Bay Networks device is capable of supporting 10 Mbps downstream and upstream, MediaOne throttles it down to 1.5 Mbps downstream and 300 Kbps upstream.
For our ISDN test, we used a U.S. Robotics (now 3Com;
http://www.3com.com
) Courier I-modem ISDN with V.Everything to connect to Netcom (
http://www.netcom.com
), a nationwide ISP that supports 64- and 128-Kbps ISDN connections. The test-bed we used was a Hitachi MX133 Notebook computer (a 133-MHz Pentium system with 32 MB of EDO DRAM) with an embedded 33.6-Kbps 3Com modem and running Windows 95.
The Courier I-modem uses 3Com's TurboPPP with AT commands, softwar
e that allows it to use both B-channels to send and receive data over ISDN. Another feature, called Dynamic Bandwidth Allocation, is intended to save money by using the second B-channel only when it's needed for data transfers. In practice, we found that the second B-channel was activated too rarely and erratically to provide meaningful test numbers.
Finally, for the analog modem tests, we used the same Hitachi MX133 portable. None of the four ISPs involved in this test supported 56K service, so all results shown are for 33.6-Kbps V.34 modems with compression enabled.
The
figure shows
average times for all three types of hardware. The cable modems are the clear winner in every case.
Test Sequence
Our test sequence consisted of three types of measurements. Each battery of tests began with 10 "pings" (short for packet Internet groper), a basic networking utility that tells if you can establish a connection to the server and gives the round-trip travel time from c
lient to server.
The second set of tests used "trace-route," which determines the number of hosts or routers between client and server. The number of such "hops" can affect performance since each router adds latency and, thus, potential delays.
The analog modem consistently made the connection in six hops, and the cable modem in eight. The ISDN connection, however, tended to be made over eight hops in the early morning and late evening, and over 12 or 16 hops during the day. We're not sure why this happened and hope to repeat the tests in the future using multiple ISPs with ISDN access.
The third battery of tests measured FTP "gets" and "puts" to determine upload and download speeds to a file server connected to the Internet via a T1 line. The file server we used is part of Marinet, a directory serving the marine-biology community and run by one of the authors.
The FTP calculation takes the file transfer in kilobytes and divides that by the elapsed time to calculate Kbps. We used a sm
all, highly compressible 250-KB file and a large, uncompressible 1.1-MB file.
Cable Modem Results
We created a program to run a test battery every hour and plotted the results to show upstream and downstream transfers as a function of time of day. The
figure compares
the weekday and weekend results to illustrate the effects of Internet congestion. Uploads were consistent at approximately 32.5 KBps when the data contained start and stop bits. Downloads are highly variable and primarily reflect backbone traffic and server loading. (The cable modem appears to use no compression, since the results were the same for all files.)
The figure below shows the averages of FTP results for cable modems. The fastest download took place at a rate of 135.9 KBps, while the slowest was 11.2 KBps; the average for 96 tests was 88 KBps.
The cable-modem ping results (96 runs with 10 pings per run) averaged 14.7 milliseconds, while the Trace Route test indicated six hops between ser
ver and client. This number was constant for all tests, an indication that the network configuration between endpoints didn't change.
But What About the Competition?
While ISDN file-transfer speeds occasionally topped 100 Kbps, the average speed was around 56 Kbps, not much above what the notoriously underachieving 56K analog modems might attain. We were disappointed in how seldom the second ISDN B-channel kicked in, but we were unable to determine where the limits to performance really lie.
Results with the analog modem were highly variable and exhibited the familiar dependency on time of day. Furthermore, for all V.34 tests that we ran with compression enabled (a likely real-world setup), the average FTP transfer rate was 31 Kbps (with start and stop bits). The slowest transfer rate was less than 500 bytes per second.
Taking the Lead
While Internet congestion continues to be a problem, it appears that all three access technologies are destined to deliver about 80 percent
to 90 percent of their rated performance when averaged over time.
Cable modems, though fast, face an imminent threat from Digital Subscriber Line (DSL) technologies, which promise to carry broadband data services over existing copper loops (see the sidebar "Cable Modems, For and Against"). But for now, cable companies are taking the lead in high-bandwidth residential services.
illustration_link (13 Kbytes)
illustration_link (35 Kbytes)
Andrew W. Davis (
andrewwd@ultranet.com
) is managing director of the Wainhouse Consulting Group, a market research and communications firm in Brookline, Massachusetts. John W. Irza (
jirza@sygnus.com
) is president of Sygnus Technology in Arlington, Massachusetts.
Lab Notes: Cable Modems Take the Early Lead
