Quantum DSL Bonding

Quantum DSL bonding is an innovative new bonding technique that overcomes the shortcomings of Multi-Link PPP (ML-PPP) on DSL lines. This new technology has been designed as a drop-in replacement for the old ML-PPP algorithm with full backwards compatibility. As such it will work with existing CPE (Customer Premises Equipment), such as Cisco or FireRack routers and firewalls.

What was wrong with Multi-Link PPP?

ML-PPP was originally designed to work with ISDN phone lines, which are very different to modern DSL connections. Each ISDN channel would provide exactly 64Kbit/sec of bandwidth, and ML-PPP did a very good job at bonding these channels together to give us 128K or 256K bonded ISDN services.

ADSL lines in particular have very different properties when compared to ISDN lines. These differences are:

  1. Unlike ISDN, DSL technologies like ADSL-Max, ADSL2+ and VDSL are adaptive rate services; The data-rate is not fixed and predictable. Two ADSL lines at the same premises can (and very often do) have very different speeds. The speeds of the individual lines can vary on a day-to-day basis. This imbalance can be problematic for standard ML-PPP.
  2. Different ADSL lines at the same site can have differing latencies (round-trip-times). ML-PPP works best when both lines have the same latency, as would be the case with ISDN lines. Large latency differences between channels can lead to packet loss, due to an overrun of the fragment reassembly buffer.
  3. The imbalance caused by the differing speeds can cause massively increased latency on slower channels, thus exacerbating the the second problem.
  4. ISDN lines did not waste bandwidth due to frame padding. DSL networks on the other hand (in the UK at least) use ATM as there delivery method. ATM sends a stream of fixed sized 53 byte cells carrying 48 bytes of data. This inevitably leads to some lost capacity due to ATM cell padding. This is not such a big issue on single-channel DSL services, as they tend to waste (on average) 24 bytes per packet on padding, whereas standard ML-PPP will tend to lose on average 24 bytes per channel to padding.

How is Quantum bonding different?

A number of techniques have been employed to address the above issues. One key technique employed by Quantum bonding is the modeling of the "Quantisation" of the underlying DSL/ATM network. This means that we do not treat the DSL line as you would a modem, or an ISDN line. We carefully construct data fragments so as to minimise wastage due to ATM cell padding. One key result of this modeling is that on average we lose only 24 bytes per packet to ATM cell padding per bonded bundle, as opposed to the usual average of 24 bytes per packet per channel.

Another important feature of Quantum bonding is that the size of the fragment reassembly buffer dynamically scales depending on the number of channels in the bundle. This change, along with other fragment handling enhancements means that large latency variations can be tolerated without any resulting packet loss.