[AusNOG] ADSL2+ Aggregation at LNS without MLPPP HOWTO

[Jarrad Mitchell]

I would expect that any form of 'single ip' ADSL Bonding is impractical for you.  If you must, you are best of to use Per-Destination Load Balancing with two separate ISPs, and perhaps route specific services through the better link. 

What follows is the technical reasoning and logical analysis that lead to the aforementioned conclusion

Why 'Bonding / Teaming Aggregating' two ADSL etc links is usually a Very Bad Idea 
ALL NETWORK INTERFACE (Including ADSL) 'Aggregation/Bonding/Teaming' that results in increased throughput across a single path (a connection between two computers/IPs eg your home pc to youtube.com) almost always results in a link that is at best just below TWICE the speed of the SLOWEST link.  
This is not always a problem.  For example, if you have two 100mbit fibre optic links over 1km, they are very unlikely to vary in their 'transmission' properties; that is to say, they are unlikely to vary considerably in how long it takes to ping the other end etc.  With a connection like this, where BOTH of the Aggregate / Team 'members' can be described as electrically/physically 'identical / significantly similar', a good quality, reliable increase in performance can be achieved.  

Key Insight 1: Multiple Link Aggregates that increase Single Path Throughput ONLY EVER make sense when using IDENTICAL Aggregate Members.
Knowing this, we next need to consider why ADSL Technology was developed.  Simply put, it was designed to leverage EXISTING, VERY OLD Balanced Transmission Line to deliver 'high speed' internet access.  And it does a wonderful job indeed.  But just what exactly is this existing infrastructure?  The PTSN is/has in most cases (if not all): 
- Based on old two pair non twisted transmission line that was never designed for data.- Had previous network modifications (been connected and disconnected, redesigned etc)- Been previously upgraded (Pulse to DTMF Dialing for example)- Been previously re-purposed (ISDN)- Been expanded well beyond its original design, inconsistantly (Loading Coils, Pair Gain Systems & RIMS).

Key Insight 2: Any Two Pairs between an Exchange and the Customer are VERY UNLIKELY to be 'identical / significantly similar'.
Some people might wish to point out that EFM / SHDSL & Similar use exactly the above to deliver a good service.  And they're right, from a delivered service (Marketing?) perspective.  Remember how I pointed out that a Aggregate will run at the speed of its slowest member?  EFM & SHDSL simply takes a bunch of pairs, and deliberately uses less bandwidth on all of them them than even the poorest can handle, then combines them.  
I have personally seen 8 PAIRS (16 wires) used to deliver 10mbit / 10mbit EFM!  To put this into perspective, a single near ideal pair can deliver 20mbit Simplex (one way) over 1Km using only 2MHz of bandwidth.  VDSL2+ over 500m, with its increased bandwidth would greatly exceed that!
Key Insight 3: YOU CAN make a bunch of DISSIMILAR (electrically) Links look Similar (logically) if you are prepared to make individual BANDWIDTH SACRIFICES.

ConclusionAnd there in lies the reality.  At say $30 per pair, it doesn't make sense economically to Aggregate a 18000/900 Kbps pair with a 9000/850 Kbps pair.  Because to do so reliably, you're likely to end up with 17000 / 1600 Kbps!!!!


 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20140912/0f123904/attachment.html>