The question is then, what makes a good modem? They certainly are not all the same in terms of aggression, from personal experience. Perhaps they offer similar performance on good lines. I wouldn't know.
For my dream modem I have a wish list of features
- correct input bandwidth, unless someone is going to tell me this is a non-issue
- please santa, monitoring of unused tones
- SRA, just in case (can but dream)
- retx (supported with ADSL2?), just in case (dream on)
- on LAN-side, use Ethernet flow control to keep queuing under control and let the router manage queues rather than the modem. Perhaps all modems already do this, I’m ashamed to say that I wouldn't know. Job for wireshark
- (having considered previous point) dear developers, pls implement QoS in L3, with sensible architecture of multiple queues, while I'm on a software development spending spree. - could do nice things such as prioritising DNS queries and TCP ACKs
For some reason I was rather disappointed with the Draytek Vigor 130 because of its less aggressive sync speed. Maybe I should retest it, because I certainly wouldn't mind getting back the free ~3% extra speed from the use of PPPoA, if my spreadsheet is not in error. Assuming TCP with 1500 byte IP MTU so L3/L4 overheads of IPv4+TCP headers, I compared the efficiency of a PPPoEoA with that PPPoA in terms the ratio of bytes over ATM to TCP SDU payload. (IPv6 comes out the same in this case.) Even if I could get the Draytek to simply match the sync rate obtained with the current Dlink DSL-320B-Z1 devices, it would be about 200kbps faster downstream - a free lunch.
(Although PPPoEoA is invariably bad news, in this case I could get a lot of that speed back if I went down to an MTU of 1492 because that would save me one whole ATM cell, 32 cells instead of 33, the only remaining difference being the reduction in efficiency because of the increased number of IP+TCP headers sent per byte of TCP payload due to the reduced MTU, but other than that, packets over PPPoA and over PPPoEoA would produce sequences of ATM cells of the same length. Of course PPPoEoA would still be worse, even worse, for short packets.)
It may be that the Draytek Vigor is not a “bad” modem on my line, it's just that the reliability vs speed tradeoff is set differently. If I could somehow tweak a Draytek to run at the same SNRM as the current DLinks, then it would be more of a fair comparison, comparing like with like.