Another annoying behavior of the Comtrend modem I got from Sonic.net: it doesn’t always transfer real data at anywhere near the claimed “sync” speed.

As noted in my initial post about Sonic.net’s Fusion Broadband, I have 2 DSL lines from Sonic.net, which each in theory should support up to 20mbps downstream, but right now one of them syncs at 18mbps and one of them at only 4mbps, due to overly high attenuation in the wiring supplied by AT&T. Still, I should be able to get a combined 22mbps, right?

Well, with the modem deployed the way I intend to use it (in a back room near my other networking equipment and my UPS, connected to a DSL filter with a normal telephone patch cord), according to Sonic’s own speed test site, I get about 4mbps download and 1.83mbps upload. Whoa, that’s a lot lower.

To make things weirder, if I disconnect the slow line (the one that syncs at only 4mbps) and leave only the fast line (synced at 18mbps) connected, the same speed test result is 15.71mbps download, 0.95mbps upload. How is one line alone faster than the two lines together? At first I suspected a bug in the modem when bonding lines of different speeds, but then I repeated the test at the MPOE, and got the expected results: the two lines bonded together were slightly faster than the fast line alone.

Curious, and suspicious of everything at this point, I decided to run a bunch of tests: both at the MPOE (not depending on my own internal wiring) and in my network closet (adding ~50 feet of internal wiring), with and without the DSL filter supplied with the modem, and with different patch cables (the patch cable that came with the modem, the separate Y cable Sonic gave me, and an old patch cable I had lying around); for various combinations of each of these variables I tested both lines together, and where possible each line alone (that’s easy to do at the MPOE and hard to do at the server room, due to the way the modem is wired for single-line operation).

The results are pretty weird. I’ll list the results, one per line, with a description of the test configuration, the modem’s reported sync speed (down/up), the actual usable speed as reported by speedtest.sonic.net, and a % number which is the ratio of usable download speed to claimed downlink sync speed. First, testing one line at a time:

• At MPOE, fast line, no filter, Y cable:  sync 18.46/1.14, speed test 15.76/0.95 (85%)
• At network closet, fast line, with filter, Y cable: sync 18.49/1.14, speed test 15.71/0.95 (85%)
• At MPOE, slow line, no filter, Y cable: sync 4.93/1.08, speed test 4.02/0.90 (82%)
• At network closet, slow line, with filter, Y cable: sync 3.48/1.02, speed test won’t run (because modem has sync on slave line only)

The result of these single line tests shows that my internal wiring (between MPOE and network closet) and filter don’t seem to affect the speed at all — the modem syncs at the same speed, and tests at the same speed, in both locations.

But then, dual line tests (sorted by descending actual speed):

• At MPOE, both lines, no filter, Y cable: sync 23.60/2.23, speed test 20.06/1.83 (85%)
• At MPOE, both lines, no filter, Y cable + their patch cable: sync 22.99/2.23, speed test 18.94/1.83 (82%)
•  At network closet, both lines, no filter, my patch cable: sync 22.84/2.23, speed test 12.05/1.83 (53%)

• At network closet, both lines, no filter, their patch cable: sync 22.86/2.23, speed test 10.38/1.83 (45%) At MPOE,  both lines, with filter, Y cable + their patch cable: sync 22.09/2.23, speed test 6.12/1.83 (28%)

• At network closet, both lines, with filter, their patch cable: sync 21.83/2.23, speed test 4.23/1.83 (19%)

Note how the sync speed does not change appreciably across all these tests, but the actual speed changes hugely. Every single difference from the minimal MPOE configuration results in decreased speed!

Some additional notes: first, bonded upload speed is completely stable at 1.83mbps in every single test I’ve run. Second, I have tried various other speed test sites, and FTP of a large test file hosted by Sonic.net in case Flash/Java speed tests aren’t reliable indicators, but I got roughly equivalent results from all of them, so I stick with the speedtest.sonic.net numbers for comparison here.

First complaint: where are the bits going? Why is the sync speed not a reliable indicator of real speed? It’s annoying losing speed, and it’s annoying not being able to trust the sync speed, and this makes troubleshooting speed problems a lot harder (it takes longer to run a speed test than just check the sync speed, and Sonic.net’s support system shows them my sync speed but not, I presume, my actual transfer speeds). Presumably if the modem can’t actually transfer data at the sync rate, that’s due to errors and dropped frames at the ATM level, but the modem’s status page shows 0 for every ATM and DSL error category.

Second complaint: why do the filter, patch cable, and my internal wiring all result in a speed loss, but only in bonded operation, and not one line at a time, and without affecting the sync speed?

It’s looking increasingly unlikely that I’m going to get a fast and trouble- free connection via DSL. I mean, I could put the modem by the MPOE (with no UPS and no filter), but that wasn’t the intent, and I don’t understand why I should have to.

And one final note: in response to my earlier article about this modem, Sonic is already sending me a new one; I don’t really want to take time to repeat all of these experiments but I’ll at least see if the new modem or new filter does better in the cases that were worst here. (If nothing else, it should be easier to do apples-to-apples comparisons, using the same cables and getting all data points at both locations.)