Buyers often greet new technology with skepticism. They've seen too many vendors promise better throughput and easier management, then fail to keep those pledges. Digital subscriber line (DSL) technology in particular has been long on promises and short on delivery.
However, we're happy to report that Paradyne's Hotwire Multiple Virtual Lines (MVL) System seems poised to follow through on its potential of simple installation and strong throughput. We found that MVL can deliver 128K bit/sec throughput even under duress and as much as 768K bit/sec of bidirectional traffic under the best conditions. That throughput level is lower than some other DSL flavors we tested last year (NW, Nov. 3, 1997, page 55), but it's well above ISDN speed. MVL also lets you make voice calls over your line without dropping the connection to the Internet.
Speed and delay
Paradyne's MVL system comprises modems connected to remote PCs and a DSL Access Multiplexer (DSLAM) based on a telco's pre-mises. MVL allows data and voice transmissions simultaneously over the same twisted-pair wiring. The components negotiate data rates depending on the condition of the DSL loop. The negoti-ated rate is called the training rate.MVL modems are easy to set up. You just plug an RJ-11 cable into the phone socket of a line that's been provisioned for MVL service and hook up a second cable between the modem and a network adapter in a PC.
We connected the two sides of our MVL installation through a Consultronics DSL loop simulator. Rather than simply test throughput and latency under ideal conditions, we came up with a test suite that simulated real-world DSL loops under favorable, unfavorable and downright nasty conditions. We called the loops the good, the bad and the ugly. In the real world, customer conditions are better than or equal to these conditions 60%, 90% and 99% of the time, respectively. We then tested three aspects of the MVL equipment: single DSL link performance; on-hook/off-hook transitions; and the running of two modems over one DSL link. In each of our loop simulations, we noted worst-case training rates, which occurred when we injected noise at the receiving end of a traffic stream.
As expected, we achieved the highest training rate while the phone was on hook because all the bandwidth was available for the data channels. On our good link, this scenario resulted in an upstream and downstream rate of 768K bit/sec. The training rate is always in multiples of 64K bit/sec because bandwidth is provisioned in 64K bit/sec chunks.
While the phone was off-hook, we saw a slightly lower training rate of 512K bit/sec upstream and 640K bit/sec downstream because the voice channel slightly overlaps the data channel and takes over when activated. Transmission rates on the data channel are reduced even when no one is speaking. How-ever, user response time is still equivalent to that of a client on a moderately loaded LAN.
The bad loop yielded a 256K bit/sec training rate in upstream and downstream directions, and the ugly loop trained to 128K bit/sec in upstream and downstream directions. Even though 128K bit/sec is low compared with the maximum allowable training rate of 768K bit/sec, the fact that the equipment was able to train over such a poor loop is a feat in itself (see Figure 1).
In our throughput tests, we found a deficit of roughly 5% to 10% in Ethernet throughput compared with the DSL training rate. The shortfall is due to the inherent inefficiencies of converting Ethernet packets to run over the DSL link. We found cases in which Ethernet throughput ex-ceeded the training rate, but that was simply due to buffering. On bidirectional tests we found that upstream and downstream traffic streams perform as designed and share the negotiated bandwidth.
Our tests turned up one surprise: a particu-larly low throughput for 64-byte packets due to a processing limitation of the DSLAM and modem. The system could forward roughly 500 of these small packets per second upstream and about 1,000 per second downstream.
Paradyne was unaware of this problem, but offered that its modem is optimized for large packets to accommodate the bulk of Internet traffic. The company also speculates that throughput for smaller packet sizes is lower because of factors such as overhead and acknowledgements. At the time we went to press, Paradyne said it corrected the problem in a new software release scheduled to be ready this week. The new code should improve performance for all packet sizes by allowing an increased number of packets to be linked before being transmitted, Paradyne says.
Watch out for latency
While throughput rates are important, high device latency can also be a performance-killer. Overall, we found latencies virtually the same for upstream and downstream directions. The equipment had a latency of about 20 msec over the good and bad loops, but as much as five times that over the ugly loop. That's high compared with other network devices, such as packet switches and routers. In fact, 100 msec is comparable to the latency for a whole cross-country IP path across the Internet. That means on the worst lines users may see Web sites appear more slowly. On typical lines, high throughput would likely make up for any latency lags.Because MVL's splitterless modems are designed to be used simultaneously with telephones, we tested on-hook to off-hook transition times. When we took a phone off the hook, the bandwidth was greatly reduced and even shut off while the devices negotiated. It took approximately 10 to 12 seconds for the system to stabilize the data channel. This seems a little long, and it would be unacceptable if a phone were picked up and put down often. However, the transition times were predictable, and the system consistently stabilized itself after each disruption.
When we simultaneously fed traffic through two modems, the pair worked well together, sharing the traffic streams in a stable manner and dividing the bandwidth between themselves. We found that total bandwidth was less than the training rate because of interpacket gaps, Ethernet preamble bits and collisions.
Setup
While latency was sometimes a problem, Paradyne's easy installation claims held true. The MVL modem and DSLAM are nothing more than Layer 2 bridges, passing traffic from one physical medium to another. All routing takes place in a CO or, in some cases, in a campus telecommunica-tions room.To allow differentiation between traffic from the different DSL links, a 10Base-T interface on the DSLAM lets you connect it to a device that supports the 802.3Q standard for virtual LANs over Ethernet.
This feature allows the DSLAM to tag the Ethernet frames for the DSL link the packets are coming from or destined for. Users can choose their own devices to handle routing and other functions. Adding routing is easy when there's only one MVL modem on the loop, but there are few devices available today that can handle multiple 802.3Q streams per port.
Users manage the MVL system through serial and Ethernet interfaces; Paradyne doesn't provide a Web interface for management. You can monitor the training rates, data rates, packet rates and other operational parameters of the modem and the DSLAM.
Paradyne has published all its MVL documentation on the World Wide Web, which makes access easy. How-ever, all the manuals are posted in file form. We'd like to see the documentation also available in HTML to eliminate the need to download and print large files.
All in all, the MVL system is a good performer that can provide DSL service even to subscribers with poor-quality loops. The ability to service multiple modems on each DSL link allows subscribers to forgo Ethernet connections in a small office by using MVL over existing phone cabling. With those benefits, we give Paradyne's Hotwire MVL System a big thumbs up.
RELATED LINKS
Scorecard and NetResults
How we ranked Hotwire in several areas, vendor contact info
How we did it
Our test methodology.
DSL in the real world
We take Hotwire MVL out for a spin outside the lab. Network World, 11/16/98.
Can't get enough DSL
Early implementers tout the easy installation and attractive pricing of DSL service - when they can get it. Network World, 11/16/98.
Bass is a senior technical staff member at North Carolina State University's Centennial Networking Labs, which tests network equipment and network-attached devices for interoperability and performance. He can be reached at john_bass @ncsu.edu.
Bass is a member of the Network World Test Alliance, a cooperative of the premier reviewers in the network industry, each bringing to bear years of practical experience on every review.
